perf(trace/opendc): Read records using low-level API

This change updates the OpenDC VM format reader implementation to use
the low-level record reading APIs provided by the `parquet-mr` library
for improved performance. Previously, we used the `parquet-avro` library
to read/write Avro records in Parquet format, but that library carries
considerable overhead.

14 files changed, 963 insertions, 257 deletions

diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableReader.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableReader.kt
index b82da888..7a01b881 100644
--- a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableReader.kt
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableReader.kt
@@ -22,38 +22,30 @@
 
 package org.opendc.trace.opendc
 
-import org.apache.avro.Schema
-import org.apache.avro.generic.GenericRecord
 import org.opendc.trace.*
 import org.opendc.trace.conv.*
+import org.opendc.trace.opendc.parquet.ResourceState
 import org.opendc.trace.util.parquet.LocalParquetReader
-import java.time.Duration
-import java.time.Instant
 
 /**
  * A [TableReader] implementation for the OpenDC virtual machine trace format.
  */
-internal class OdcVmResourceStateTableReader(private val reader: LocalParquetReader<GenericRecord>) : TableReader {
+internal class OdcVmResourceStateTableReader(private val reader: LocalParquetReader<ResourceState>) : TableReader {
     /**
      * The current record.
      */
-    private var record: GenericRecord? = null
-
-    /**
-     * A flag to indicate that the columns have been initialized.
-     */
-    private var hasInitializedColumns = false
+    private var record: ResourceState? = null
 
     override fun nextRow(): Boolean {
-        val record = reader.read()
-        this.record = record
+        try {
+            val record = reader.read()
+            this.record = record
 
-        if (!hasInitializedColumns && record != null) {
-            initColumns(record.schema)
-            hasInitializedColumns = true
+            return record != null
+        } catch (e: Throwable) {
+            this.record = null
+            throw e
         }
-
-        return record != null
     }
 
     override fun resolve(column: TableColumn<*>): Int = columns[column] ?: -1
@@ -67,36 +59,36 @@ internal class OdcVmResourceStateTableReader(private val reader: LocalParquetRea
         val record = checkNotNull(record) { "Reader in invalid state" }
 
         return when (index) {
-            COL_ID -> record[AVRO_COL_ID].toString()
-            COL_TIMESTAMP -> Instant.ofEpochMilli(record[AVRO_COL_TIMESTAMP] as Long)
-            COL_DURATION -> Duration.ofMillis(record[AVRO_COL_DURATION] as Long)
-            COL_CPU_COUNT -> getInt(index)
-            COL_CPU_USAGE -> getDouble(index)
-            else -> throw IllegalArgumentException("Invalid column")
+            COL_ID -> record.id
+            COL_TIMESTAMP -> record.timestamp
+            COL_DURATION -> record.duration
+            COL_CPU_COUNT -> record.cpuCount
+            COL_CPU_USAGE -> record.cpuUsage
+            else -> throw IllegalArgumentException("Invalid column index $index")
         }
     }
 
     override fun getBoolean(index: Int): Boolean {
-        throw IllegalArgumentException("Invalid column")
+        throw IllegalArgumentException("Invalid column or type [index $index]")
     }
 
     override fun getInt(index: Int): Int {
         val record = checkNotNull(record) { "Reader in invalid state" }
         return when (index) {
-            COL_CPU_COUNT -> record[AVRO_COL_CPU_COUNT] as Int
-            else -> throw IllegalArgumentException("Invalid column")
+            COL_CPU_COUNT -> record.cpuCount
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
         }
     }
 
     override fun getLong(index: Int): Long {
-        throw IllegalArgumentException("Invalid column")
+        throw IllegalArgumentException("Invalid column or type [index $index]")
     }
 
     override fun getDouble(index: Int): Double {
         val record = checkNotNull(record) { "Reader in invalid state" }
         return when (index) {
-            COL_CPU_USAGE -> (record[AVRO_COL_CPU_USAGE] as Number).toDouble()
-            else -> throw IllegalArgumentException("Invalid column")
+            COL_CPU_USAGE -> record.cpuUsage
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
         }
     }
 
@@ -106,28 +98,6 @@ internal class OdcVmResourceStateTableReader(private val reader: LocalParquetRea
 
     override fun toString(): String = "OdcVmResourceStateTableReader"
 
-    /**
-     * Initialize the columns for the reader based on [schema].
-     */
-    private fun initColumns(schema: Schema) {
-        try {
-            AVRO_COL_ID = schema.getField("id").pos()
-            AVRO_COL_TIMESTAMP = (schema.getField("timestamp") ?: schema.getField("time")).pos()
-            AVRO_COL_DURATION = schema.getField("duration").pos()
-            AVRO_COL_CPU_COUNT = (schema.getField("cpu_count") ?: schema.getField("cores")).pos()
-            AVRO_COL_CPU_USAGE = (schema.getField("cpu_usage") ?: schema.getField("cpuUsage")).pos()
-        } catch (e: NullPointerException) {
-            // This happens when the field we are trying to access does not exist
-            throw IllegalArgumentException("Invalid schema", e)
-        }
-    }
-
-    private var AVRO_COL_ID = -1
-    private var AVRO_COL_TIMESTAMP = -1
-    private var AVRO_COL_DURATION = -1
-    private var AVRO_COL_CPU_COUNT = -1
-    private var AVRO_COL_CPU_USAGE = -1
-
     private val COL_ID = 0
     private val COL_TIMESTAMP = 1
     private val COL_DURATION = 2
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableWriter.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableWriter.kt
index 01b9750c..97af5b59 100644
--- a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableWriter.kt
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceStateTableWriter.kt
@@ -22,84 +22,85 @@
 
 package org.opendc.trace.opendc
 
-import org.apache.avro.Schema
-import org.apache.avro.generic.GenericRecord
-import org.apache.avro.generic.GenericRecordBuilder
 import org.apache.parquet.hadoop.ParquetWriter
 import org.opendc.trace.*
 import org.opendc.trace.conv.*
+import org.opendc.trace.opendc.parquet.ResourceState
 import java.time.Duration
 import java.time.Instant
 
 /**
  * A [TableWriter] implementation for the OpenDC virtual machine trace format.
  */
-internal class OdcVmResourceStateTableWriter(
-    private val writer: ParquetWriter<GenericRecord>,
-    private val schema: Schema
-) : TableWriter {
+internal class OdcVmResourceStateTableWriter(private val writer: ParquetWriter<ResourceState>) : TableWriter {
     /**
-     * The current builder for the record that is being written.
+     * The current state for the record that is being written.
      */
-    private var builder: GenericRecordBuilder? = null
-
-    /**
-     * The fields belonging to the resource state schema.
-     */
-    private val fields = schema.fields
+    private var _isActive = false
+    private var _id: String = ""
+    private var _timestamp: Instant = Instant.MIN
+    private var _duration: Duration = Duration.ZERO
+    private var _cpuCount: Int = 0
+    private var _cpuUsage: Double = Double.NaN
 
     override fun startRow() {
-        builder = GenericRecordBuilder(schema)
+        _isActive = true
+        _id = ""
+        _timestamp = Instant.MIN
+        _duration = Duration.ZERO
+        _cpuCount = 0
+        _cpuUsage = Double.NaN
     }
 
     override fun endRow() {
-        val builder = checkNotNull(builder) { "No active row" }
-        this.builder = null
-
-        val record = builder.build()
-        val id = record[COL_ID] as String
-        val timestamp = record[COL_TIMESTAMP] as Long
+        check(_isActive) { "No active row" }
+        _isActive = false
 
-        check(lastId != id || timestamp >= lastTimestamp) { "Records need to be ordered by (id, timestamp)" }
+        check(lastId != _id || _timestamp >= lastTimestamp) { "Records need to be ordered by (id, timestamp)" }
 
-        writer.write(builder.build())
+        writer.write(ResourceState(_id, _timestamp, _duration, _cpuCount, _cpuUsage))
 
-        lastId = id
-        lastTimestamp = timestamp
+        lastId = _id
+        lastTimestamp = _timestamp
     }
 
-    override fun resolve(column: TableColumn<*>): Int {
-        val schema = schema
-        return when (column) {
-            RESOURCE_ID -> schema.getField("id").pos()
-            RESOURCE_STATE_TIMESTAMP -> (schema.getField("timestamp") ?: schema.getField("time")).pos()
-            RESOURCE_STATE_DURATION -> schema.getField("duration").pos()
-            RESOURCE_CPU_COUNT -> (schema.getField("cpu_count") ?: schema.getField("cores")).pos()
-            RESOURCE_STATE_CPU_USAGE -> (schema.getField("cpu_usage") ?: schema.getField("cpuUsage")).pos()
-            else -> -1
-        }
-    }
+    override fun resolve(column: TableColumn<*>): Int = columns[column] ?: -1
 
     override fun set(index: Int, value: Any) {
-        val builder = checkNotNull(builder) { "No active row" }
-
-        builder.set(
-            fields[index],
-            when (index) {
-                COL_TIMESTAMP -> (value as Instant).toEpochMilli()
-                COL_DURATION -> (value as Duration).toMillis()
-                else -> value
-            }
-        )
+        check(_isActive) { "No active row" }
+
+        when (index) {
+            COL_ID -> _id = value as String
+            COL_TIMESTAMP -> _timestamp = value as Instant
+            COL_DURATION -> _duration = value as Duration
+            COL_CPU_COUNT -> _cpuCount = value as Int
+            COL_CPU_USAGE -> _cpuUsage = value as Double
+        }
     }
 
-    override fun setBoolean(index: Int, value: Boolean) = set(index, value)
+    override fun setBoolean(index: Int, value: Boolean) {
+        throw IllegalArgumentException("Invalid column or type [index $index]")
+    }
 
-    override fun setInt(index: Int, value: Int) = set(index, value)
+    override fun setInt(index: Int, value: Int) {
+        check(_isActive) { "No active row" }
+        when (index) {
+            COL_CPU_COUNT -> _cpuCount = value
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
+        }
+    }
 
-    override fun setLong(index: Int, value: Long) = set(index, value)
+    override fun setLong(index: Int, value: Long) {
+        throw IllegalArgumentException("Invalid column or type [index $index]")
+    }
 
-    override fun setDouble(index: Int, value: Double) = set(index, value)
+    override fun setDouble(index: Int, value: Double) {
+        check(_isActive) { "No active row" }
+        when (index) {
+            COL_CPU_USAGE -> _cpuUsage = value
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
+        }
+    }
 
     override fun flush() {
         // Not available
@@ -113,12 +114,19 @@ internal class OdcVmResourceStateTableWriter(
      * Last column values that are used to check for correct partitioning.
      */
     private var lastId: String? = null
-    private var lastTimestamp: Long = Long.MIN_VALUE
-
-    /**
-     * Columns with special behavior.
-     */
-    private val COL_ID = resolve(RESOURCE_ID)
-    private val COL_TIMESTAMP = resolve(RESOURCE_STATE_TIMESTAMP)
-    private val COL_DURATION = resolve(RESOURCE_STATE_DURATION)
+    private var lastTimestamp: Instant = Instant.MAX
+
+    private val COL_ID = 0
+    private val COL_TIMESTAMP = 1
+    private val COL_DURATION = 2
+    private val COL_CPU_COUNT = 3
+    private val COL_CPU_USAGE = 4
+
+    private val columns = mapOf(
+        RESOURCE_ID to COL_ID,
+        RESOURCE_STATE_TIMESTAMP to COL_TIMESTAMP,
+        RESOURCE_STATE_DURATION to COL_DURATION,
+        RESOURCE_CPU_COUNT to COL_CPU_COUNT,
+        RESOURCE_STATE_CPU_USAGE to COL_CPU_USAGE,
+    )
 }
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableReader.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableReader.kt
index 4909e70e..6102332f 100644
--- a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableReader.kt
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableReader.kt
@@ -22,37 +22,30 @@
 
 package org.opendc.trace.opendc
 
-import org.apache.avro.Schema
-import org.apache.avro.generic.GenericRecord
 import org.opendc.trace.*
 import org.opendc.trace.conv.*
+import org.opendc.trace.opendc.parquet.Resource
 import org.opendc.trace.util.parquet.LocalParquetReader
-import java.time.Instant
 
 /**
- * A [TableReader] implementation for the resources table in the OpenDC virtual machine trace format.
+ * A [TableReader] implementation for the "resources table" in the OpenDC virtual machine trace format.
  */
-internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<GenericRecord>) : TableReader {
+internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<Resource>) : TableReader {
     /**
      * The current record.
      */
-    private var record: GenericRecord? = null
-
-    /**
-     * A flag to indicate that the columns have been initialized.
-     */
-    private var hasInitializedColumns = false
+    private var record: Resource? = null
 
     override fun nextRow(): Boolean {
-        val record = reader.read()
-        this.record = record
+        try {
+            val record = reader.read()
+            this.record = record
 
-        if (!hasInitializedColumns && record != null) {
-            initColumns(record.schema)
-            hasInitializedColumns = true
+            return record != null
+        } catch (e: Throwable) {
+            this.record = null
+            throw e
         }
-
-        return record != null
     }
 
     override fun resolve(column: TableColumn<*>): Int = columns[column] ?: -1
@@ -66,9 +59,9 @@ internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<G
         val record = checkNotNull(record) { "Reader in invalid state" }
 
         return when (index) {
-            COL_ID -> record[AVRO_COL_ID].toString()
-            COL_START_TIME -> Instant.ofEpochMilli(record[AVRO_COL_START_TIME] as Long)
-            COL_STOP_TIME -> Instant.ofEpochMilli(record[AVRO_COL_STOP_TIME] as Long)
+            COL_ID -> record.id
+            COL_START_TIME -> record.startTime
+            COL_STOP_TIME -> record.stopTime
             COL_CPU_COUNT -> getInt(index)
             COL_CPU_CAPACITY -> getDouble(index)
             COL_MEM_CAPACITY -> getDouble(index)
@@ -84,7 +77,7 @@ internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<G
         val record = checkNotNull(record) { "Reader in invalid state" }
 
         return when (index) {
-            COL_CPU_COUNT -> record[AVRO_COL_CPU_COUNT] as Int
+            COL_CPU_COUNT -> record.cpuCount
             else -> throw IllegalArgumentException("Invalid column")
         }
     }
@@ -97,8 +90,8 @@ internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<G
         val record = checkNotNull(record) { "Reader in invalid state" }
 
         return when (index) {
-            COL_CPU_CAPACITY -> if (AVRO_COL_CPU_CAPACITY >= 0) (record[AVRO_COL_CPU_CAPACITY] as Number).toDouble() else 0.0
-            COL_MEM_CAPACITY -> (record[AVRO_COL_MEM_CAPACITY] as Number).toDouble()
+            COL_CPU_CAPACITY -> record.cpuCapacity
+            COL_MEM_CAPACITY -> record.memCapacity
             else -> throw IllegalArgumentException("Invalid column")
         }
     }
@@ -109,30 +102,6 @@ internal class OdcVmResourceTableReader(private val reader: LocalParquetReader<G
 
     override fun toString(): String = "OdcVmResourceTableReader"
 
-    /**
-     * Initialize the columns for the reader based on [schema].
-     */
-    private fun initColumns(schema: Schema) {
-        try {
-            AVRO_COL_ID = schema.getField("id").pos()
-            AVRO_COL_START_TIME = (schema.getField("start_time") ?: schema.getField("submissionTime")).pos()
-            AVRO_COL_STOP_TIME = (schema.getField("stop_time") ?: schema.getField("endTime")).pos()
-            AVRO_COL_CPU_COUNT = (schema.getField("cpu_count") ?: schema.getField("maxCores")).pos()
-            AVRO_COL_CPU_CAPACITY = schema.getField("cpu_capacity")?.pos() ?: -1
-            AVRO_COL_MEM_CAPACITY = (schema.getField("mem_capacity") ?: schema.getField("requiredMemory")).pos()
-        } catch (e: NullPointerException) {
-            // This happens when the field we are trying to access does not exist
-            throw IllegalArgumentException("Invalid schema")
-        }
-    }
-
-    private var AVRO_COL_ID = -1
-    private var AVRO_COL_START_TIME = -1
-    private var AVRO_COL_STOP_TIME = -1
-    private var AVRO_COL_CPU_COUNT = -1
-    private var AVRO_COL_CPU_CAPACITY = -1
-    private var AVRO_COL_MEM_CAPACITY = -1
-
     private val COL_ID = 0
     private val COL_START_TIME = 1
     private val COL_STOP_TIME = 2
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableWriter.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableWriter.kt
index edc89ee6..cae65faa 100644
--- a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableWriter.kt
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmResourceTableWriter.kt
@@ -22,74 +22,82 @@
 
 package org.opendc.trace.opendc
 
-import org.apache.avro.Schema
-import org.apache.avro.generic.GenericRecord
-import org.apache.avro.generic.GenericRecordBuilder
 import org.apache.parquet.hadoop.ParquetWriter
 import org.opendc.trace.*
 import org.opendc.trace.conv.*
+import org.opendc.trace.opendc.parquet.Resource
 import java.time.Instant
-import kotlin.math.roundToLong
 
 /**
  * A [TableWriter] implementation for the OpenDC virtual machine trace format.
  */
-internal class OdcVmResourceTableWriter(
-    private val writer: ParquetWriter<GenericRecord>,
-    private val schema: Schema
-) : TableWriter {
+internal class OdcVmResourceTableWriter(private val writer: ParquetWriter<Resource>) : TableWriter {
     /**
-     * The current builder for the record that is being written.
+     * The current state for the record that is being written.
      */
-    private var builder: GenericRecordBuilder? = null
-
-    /**
-     * The fields belonging to the resource schema.
-     */
-    private val fields = schema.fields
+    private var _isActive = false
+    private var _id: String = ""
+    private var _startTime: Instant = Instant.MIN
+    private var _stopTime: Instant = Instant.MIN
+    private var _cpuCount: Int = 0
+    private var _cpuCapacity: Double = Double.NaN
+    private var _memCapacity: Double = Double.NaN
 
     override fun startRow() {
-        builder = GenericRecordBuilder(schema)
+        _isActive = true
+        _id = ""
+        _startTime = Instant.MIN
+        _stopTime = Instant.MIN
+        _cpuCount = 0
+        _cpuCapacity = Double.NaN
+        _memCapacity = Double.NaN
     }
 
     override fun endRow() {
-        val builder = checkNotNull(builder) { "No active row" }
-        this.builder = null
-        writer.write(builder.build())
+        check(_isActive) { "No active row" }
+        _isActive = false
+        writer.write(Resource(_id, _startTime, _stopTime, _cpuCount, _cpuCapacity, _memCapacity))
     }
 
-    override fun resolve(column: TableColumn<*>): Int {
-        val schema = schema
-        return when (column) {
-            RESOURCE_ID -> schema.getField("id").pos()
-            RESOURCE_START_TIME -> (schema.getField("start_time") ?: schema.getField("submissionTime")).pos()
-            RESOURCE_STOP_TIME -> (schema.getField("stop_time") ?: schema.getField("endTime")).pos()
-            RESOURCE_CPU_COUNT -> (schema.getField("cpu_count") ?: schema.getField("maxCores")).pos()
-            RESOURCE_CPU_CAPACITY -> schema.getField("cpu_capacity").pos()
-            RESOURCE_MEM_CAPACITY -> (schema.getField("mem_capacity") ?: schema.getField("requiredMemory")).pos()
-            else -> -1
-        }
-    }
+    override fun resolve(column: TableColumn<*>): Int = columns[column] ?: -1
 
     override fun set(index: Int, value: Any) {
-        val builder = checkNotNull(builder) { "No active row" }
-        builder.set(
-            fields[index],
-            when (index) {
-                COL_START_TIME, COL_STOP_TIME -> (value as Instant).toEpochMilli()
-                COL_MEM_CAPACITY -> (value as Double).roundToLong()
-                else -> value
-            }
-        )
+        check(_isActive) { "No active row" }
+        when (index) {
+            COL_ID -> _id = value as String
+            COL_START_TIME -> _startTime = value as Instant
+            COL_STOP_TIME -> _stopTime = value as Instant
+            COL_CPU_COUNT -> _cpuCount = value as Int
+            COL_CPU_CAPACITY -> _cpuCapacity = value as Double
+            COL_MEM_CAPACITY -> _memCapacity = value as Double
+            else -> throw IllegalArgumentException("Invalid column index $index")
+        }
     }
 
-    override fun setBoolean(index: Int, value: Boolean) = set(index, value)
+    override fun setBoolean(index: Int, value: Boolean) {
+        throw IllegalArgumentException("Invalid column or type [index $index]")
+    }
 
-    override fun setInt(index: Int, value: Int) = set(index, value)
+    override fun setInt(index: Int, value: Int) {
+        check(_isActive) { "No active row" }
+        when (index) {
+            COL_CPU_COUNT -> _cpuCount = value
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
+        }
+    }
 
-    override fun setLong(index: Int, value: Long) = set(index, value)
+    override fun setLong(index: Int, value: Long) {
+        throw IllegalArgumentException("Invalid column or type [index $index]")
+    }
 
-    override fun setDouble(index: Int, value: Double) = set(index, value)
+    override fun setDouble(index: Int, value: Double) {
+        check(_isActive) { "No active row" }
+        when (index) {
+            COL_CPU_CAPACITY -> _cpuCapacity = value
+            COL_MEM_CAPACITY -> _memCapacity = value
+            else -> throw IllegalArgumentException("Invalid column or type [index $index]")
+        }
+    }
 
     override fun flush() {
         // Not available
@@ -99,10 +107,19 @@ internal class OdcVmResourceTableWriter(
         writer.close()
     }
 
-    /**
-     * Columns with special behavior.
-     */
-    private val COL_START_TIME = resolve(RESOURCE_START_TIME)
-    private val COL_STOP_TIME = resolve(RESOURCE_STOP_TIME)
-    private val COL_MEM_CAPACITY = resolve(RESOURCE_MEM_CAPACITY)
+    private val COL_ID = 0
+    private val COL_START_TIME = 1
+    private val COL_STOP_TIME = 2
+    private val COL_CPU_COUNT = 3
+    private val COL_CPU_CAPACITY = 4
+    private val COL_MEM_CAPACITY = 5
+
+    private val columns = mapOf(
+        RESOURCE_ID to COL_ID,
+        RESOURCE_START_TIME to COL_START_TIME,
+        RESOURCE_STOP_TIME to COL_STOP_TIME,
+        RESOURCE_CPU_COUNT to COL_CPU_COUNT,
+        RESOURCE_CPU_CAPACITY to COL_CPU_CAPACITY,
+        RESOURCE_MEM_CAPACITY to COL_MEM_CAPACITY,
+    )
 }
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmTraceFormat.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmTraceFormat.kt
index 1a15c7b3..155f8cf3 100644
--- a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmTraceFormat.kt
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/OdcVmTraceFormat.kt
@@ -22,19 +22,19 @@
 
 package org.opendc.trace.opendc
 
-import org.apache.avro.Schema
-import org.apache.avro.SchemaBuilder
-import org.apache.avro.generic.GenericRecord
-import org.apache.parquet.avro.AvroParquetWriter
+import org.apache.parquet.column.ParquetProperties
 import org.apache.parquet.hadoop.ParquetFileWriter
 import org.apache.parquet.hadoop.metadata.CompressionCodecName
 import org.opendc.trace.*
 import org.opendc.trace.conv.*
+import org.opendc.trace.opendc.parquet.ResourceReadSupport
+import org.opendc.trace.opendc.parquet.ResourceStateReadSupport
+import org.opendc.trace.opendc.parquet.ResourceStateWriteSupport
+import org.opendc.trace.opendc.parquet.ResourceWriteSupport
 import org.opendc.trace.spi.TableDetails
 import org.opendc.trace.spi.TraceFormat
-import org.opendc.trace.util.parquet.LocalOutputFile
 import org.opendc.trace.util.parquet.LocalParquetReader
-import org.opendc.trace.util.avro.TIMESTAMP_SCHEMA
+import org.opendc.trace.util.parquet.LocalParquetWriter
 import shaded.parquet.com.fasterxml.jackson.core.JsonEncoding
 import shaded.parquet.com.fasterxml.jackson.core.JsonFactory
 import java.nio.file.Files
@@ -105,11 +105,11 @@ public class OdcVmTraceFormat : TraceFormat {
     override fun newReader(path: Path, table: String): TableReader {
         return when (table) {
             TABLE_RESOURCES -> {
-                val reader = LocalParquetReader<GenericRecord>(path.resolve("meta.parquet"))
+                val reader = LocalParquetReader(path.resolve("meta.parquet"), LocalParquetReader.custom(ResourceReadSupport()))
                 OdcVmResourceTableReader(reader)
             }
             TABLE_RESOURCE_STATES -> {
-                val reader = LocalParquetReader<GenericRecord>(path.resolve("trace.parquet"))
+                val reader = LocalParquetReader(path.resolve("trace.parquet"), LocalParquetReader.custom(ResourceStateReadSupport()))
                 OdcVmResourceStateTableReader(reader)
             }
             TABLE_INTERFERENCE_GROUPS -> {
@@ -128,24 +128,24 @@ public class OdcVmTraceFormat : TraceFormat {
     override fun newWriter(path: Path, table: String): TableWriter {
         return when (table) {
             TABLE_RESOURCES -> {
-                val schema = RESOURCES_SCHEMA
-                val writer = AvroParquetWriter.builder<GenericRecord>(LocalOutputFile(path.resolve("meta.parquet")))
-                    .withSchema(schema)
+                val writer = LocalParquetWriter.builder(path.resolve("meta.parquet"), ResourceWriteSupport())
                     .withCompressionCodec(CompressionCodecName.ZSTD)
+                    .withPageWriteChecksumEnabled(true)
+                    .withWriterVersion(ParquetProperties.WriterVersion.PARQUET_2_0)
                     .withWriteMode(ParquetFileWriter.Mode.OVERWRITE)
                     .build()
-                OdcVmResourceTableWriter(writer, schema)
+                OdcVmResourceTableWriter(writer)
             }
             TABLE_RESOURCE_STATES -> {
-                val schema = RESOURCE_STATES_SCHEMA
-                val writer = AvroParquetWriter.builder<GenericRecord>(LocalOutputFile(path.resolve("trace.parquet")))
-                    .withSchema(schema)
+                val writer = LocalParquetWriter.builder(path.resolve("trace.parquet"), ResourceStateWriteSupport())
                     .withCompressionCodec(CompressionCodecName.ZSTD)
                     .withDictionaryEncoding("id", true)
                     .withBloomFilterEnabled("id", true)
+                    .withPageWriteChecksumEnabled(true)
+                    .withWriterVersion(ParquetProperties.WriterVersion.PARQUET_2_0)
                     .withWriteMode(ParquetFileWriter.Mode.OVERWRITE)
                     .build()
-                OdcVmResourceStateTableWriter(writer, schema)
+                OdcVmResourceStateTableWriter(writer)
             }
             TABLE_INTERFERENCE_GROUPS -> {
                 val generator = jsonFactory.createGenerator(path.resolve("interference-model.json").toFile(), JsonEncoding.UTF8)
@@ -154,37 +154,4 @@ public class OdcVmTraceFormat : TraceFormat {
             else -> throw IllegalArgumentException("Table $table not supported")
         }
     }
-
-    public companion object {
-        /**
-         * Schema for the resources table in the trace.
-         */
-        @JvmStatic
-        public val RESOURCES_SCHEMA: Schema = SchemaBuilder
-            .record("resource")
-            .namespace("org.opendc.trace.opendc")
-            .fields()
-            .requiredString("id")
-            .name("start_time").type(TIMESTAMP_SCHEMA).noDefault()
-            .name("stop_time").type(TIMESTAMP_SCHEMA).noDefault()
-            .requiredInt("cpu_count")
-            .requiredDouble("cpu_capacity")
-            .requiredLong("mem_capacity")
-            .endRecord()
-
-        /**
-         * Schema for the resource states table in the trace.
-         */
-        @JvmStatic
-        public val RESOURCE_STATES_SCHEMA: Schema = SchemaBuilder
-            .record("resource_state")
-            .namespace("org.opendc.trace.opendc")
-            .fields()
-            .requiredString("id")
-            .name("timestamp").type(TIMESTAMP_SCHEMA).noDefault()
-            .requiredLong("duration")
-            .requiredInt("cpu_count")
-            .requiredDouble("cpu_usage")
-            .endRecord()
-    }
 }
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/Resource.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/Resource.kt
new file mode 100644
index 00000000..c6db45b5
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/Resource.kt
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import java.time.Instant
+
+/**
+ * A description of a resource in a trace.
+ */
+internal data class Resource(
+    val id: String,
+    val startTime: Instant,
+    val stopTime: Instant,
+    val cpuCount: Int,
+    val cpuCapacity: Double,
+    val memCapacity: Double
+)
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceReadSupport.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceReadSupport.kt
new file mode 100644
index 00000000..47cce914
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceReadSupport.kt
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.hadoop.conf.Configuration
+import org.apache.parquet.hadoop.api.InitContext
+import org.apache.parquet.hadoop.api.ReadSupport
+import org.apache.parquet.io.api.RecordMaterializer
+import org.apache.parquet.schema.*
+
+/**
+ * A [ReadSupport] instance for [Resource] objects.
+ */
+internal class ResourceReadSupport : ReadSupport<Resource>() {
+    override fun init(context: InitContext): ReadContext {
+        return ReadContext(READ_SCHEMA)
+    }
+
+    override fun prepareForRead(
+        configuration: Configuration,
+        keyValueMetaData: Map<String, String>,
+        fileSchema: MessageType,
+        readContext: ReadContext
+    ): RecordMaterializer<Resource> = ResourceRecordMaterializer(readContext.requestedSchema)
+
+    companion object {
+        /**
+         * Parquet read schema (version 2.0) for the "resources" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA_V2_0: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("submissionTime"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("endTime"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("maxCores"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("requiredMemory"),
+            )
+            .named("resource")
+
+        /**
+         * Parquet read schema (version 2.1) for the "resources" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA_V2_1: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("start_time"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("stop_time"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("cpu_count"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.DOUBLE)
+                    .named("cpu_capacity"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("mem_capacity"),
+            )
+            .named("resource")
+
+        /**
+         * Parquet read schema for the "resources" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA: MessageType = READ_SCHEMA_V2_0
+            .union(READ_SCHEMA_V2_1)
+    }
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceRecordMaterializer.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceRecordMaterializer.kt
new file mode 100644
index 00000000..3adb0709
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceRecordMaterializer.kt
@@ -0,0 +1,107 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.parquet.io.api.*
+import org.apache.parquet.schema.MessageType
+import java.time.Instant
+
+/**
+ * A [RecordMaterializer] for [Resource] records.
+ */
+internal class ResourceRecordMaterializer(schema: MessageType) : RecordMaterializer<Resource>() {
+    /**
+     * State of current record being read.
+     */
+    private var _id = ""
+    private var _startTime = Instant.MIN
+    private var _stopTime = Instant.MIN
+    private var _cpuCount = 0
+    private var _cpuCapacity = 0.0
+    private var _memCapacity = 0.0
+
+    /**
+     * Root converter for the record.
+     */
+    private val root = object : GroupConverter() {
+        /**
+         * The converters for the columns of the schema.
+         */
+        private val converters = schema.fields.map { type ->
+            when (type.name) {
+                "id" -> object : PrimitiveConverter() {
+                    override fun addBinary(value: Binary) {
+                        _id = value.toStringUsingUTF8()
+                    }
+                }
+                "start_time", "submissionTime" -> object : PrimitiveConverter() {
+                    override fun addLong(value: Long) {
+                        _startTime = Instant.ofEpochMilli(value)
+                    }
+                }
+                "stop_time", "endTime" -> object : PrimitiveConverter() {
+                    override fun addLong(value: Long) {
+                        _stopTime = Instant.ofEpochMilli(value)
+                    }
+                }
+                "cpu_count", "maxCores" -> object : PrimitiveConverter() {
+                    override fun addInt(value: Int) {
+                        _cpuCount = value
+                    }
+                }
+                "cpu_capacity" -> object : PrimitiveConverter() {
+                    override fun addDouble(value: Double) {
+                        _cpuCapacity = value
+                    }
+                }
+                "mem_capacity", "requiredMemory" -> object : PrimitiveConverter() {
+                    override fun addDouble(value: Double) {
+                        _memCapacity = value
+                    }
+
+                    override fun addLong(value: Long) {
+                        _memCapacity = value.toDouble()
+                    }
+                }
+                else -> error("Unknown column $type")
+            }
+        }
+
+        override fun start() {
+            _id = ""
+            _startTime = Instant.MIN
+            _stopTime = Instant.MIN
+            _cpuCount = 0
+            _cpuCapacity = 0.0
+            _memCapacity = 0.0
+        }
+
+        override fun end() {}
+
+        override fun getConverter(fieldIndex: Int): Converter = converters[fieldIndex]
+    }
+
+    override fun getCurrentRecord(): Resource = Resource(_id, _startTime, _stopTime, _cpuCount, _cpuCapacity, _memCapacity)
+
+    override fun getRootConverter(): GroupConverter = root
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceState.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceState.kt
new file mode 100644
index 00000000..9ad58764
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceState.kt
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import java.time.Duration
+import java.time.Instant
+
+internal class ResourceState(
+    val id: String,
+    val timestamp: Instant,
+    val duration: Duration,
+    val cpuCount: Int,
+    val cpuUsage: Double
+)
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateReadSupport.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateReadSupport.kt
new file mode 100644
index 00000000..17840ceb
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateReadSupport.kt
@@ -0,0 +1,105 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.hadoop.conf.Configuration
+import org.apache.parquet.hadoop.api.InitContext
+import org.apache.parquet.hadoop.api.ReadSupport
+import org.apache.parquet.io.api.RecordMaterializer
+import org.apache.parquet.schema.*
+
+/**
+ * A [ReadSupport] instance for [ResourceState] objects.
+ */
+internal class ResourceStateReadSupport : ReadSupport<ResourceState>() {
+    override fun init(context: InitContext): ReadContext {
+        return ReadContext(READ_SCHEMA)
+    }
+
+    override fun prepareForRead(
+        configuration: Configuration,
+        keyValueMetaData: Map<String, String>,
+        fileSchema: MessageType,
+        readContext: ReadContext
+    ): RecordMaterializer<ResourceState> = ResourceStateRecordMaterializer(readContext.requestedSchema)
+
+    companion object {
+        /**
+         * Parquet read schema (version 2.0) for the "resource states" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA_V2_0: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("time"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("duration"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("cores"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.DOUBLE)
+                    .named("cpuUsage")
+            )
+            .named("resource_state")
+
+        /**
+         * Parquet read schema (version 2.1) for the "resource states" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA_V2_1: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("timestamp"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("duration"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("cpu_count"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.DOUBLE)
+                    .named("cpu_usage")
+            )
+            .named("resource_state")
+
+        /**
+         * Parquet read schema for the "resource states" table in the trace.
+         */
+        @JvmStatic
+        val READ_SCHEMA: MessageType = READ_SCHEMA_V2_0.union(READ_SCHEMA_V2_1)
+    }
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateRecordMaterializer.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateRecordMaterializer.kt
new file mode 100644
index 00000000..f8b0c3c2
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateRecordMaterializer.kt
@@ -0,0 +1,102 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.parquet.io.api.*
+import org.apache.parquet.schema.MessageType
+import java.time.Duration
+import java.time.Instant
+
+/**
+ * A [RecordMaterializer] for [ResourceState] records.
+ */
+internal class ResourceStateRecordMaterializer(schema: MessageType) : RecordMaterializer<ResourceState>() {
+    /**
+     * State of current record being read.
+     */
+    private var _id = ""
+    private var _timestamp = Instant.MIN
+    private var _duration = Duration.ZERO
+    private var _cpuCount = 0
+    private var _cpuUsage = 0.0
+
+    /**
+     * Root converter for the record.
+     */
+    private val root = object : GroupConverter() {
+        /**
+         * The converters for the columns of the schema.
+         */
+        private val converters = schema.fields.map { type ->
+            when (type.name) {
+                "id" -> object : PrimitiveConverter() {
+                    override fun addBinary(value: Binary) {
+                        _id = value.toStringUsingUTF8()
+                    }
+                }
+                "timestamp", "time" -> object : PrimitiveConverter() {
+                    override fun addLong(value: Long) {
+                        _timestamp = Instant.ofEpochMilli(value)
+                    }
+                }
+                "duration" -> object : PrimitiveConverter() {
+                    override fun addLong(value: Long) {
+                        _duration = Duration.ofMillis(value)
+                    }
+                }
+                "cpu_count", "cores" -> object : PrimitiveConverter() {
+                    override fun addInt(value: Int) {
+                        _cpuCount = value
+                    }
+                }
+                "cpu_usage", "cpuUsage" -> object : PrimitiveConverter() {
+                    override fun addDouble(value: Double) {
+                        _cpuUsage = value
+                    }
+                }
+                "flops" -> object : PrimitiveConverter() {
+                    override fun addLong(value: Long) {
+                        // Ignore to support v1 format
+                    }
+                }
+                else -> error("Unknown column $type")
+            }
+        }
+
+        override fun start() {
+            _id = ""
+            _timestamp = Instant.MIN
+            _duration = Duration.ZERO
+            _cpuCount = 0
+            _cpuUsage = 0.0
+        }
+
+        override fun end() {}
+
+        override fun getConverter(fieldIndex: Int): Converter = converters[fieldIndex]
+    }
+
+    override fun getCurrentRecord(): ResourceState = ResourceState(_id, _timestamp, _duration, _cpuCount, _cpuUsage)
+
+    override fun getRootConverter(): GroupConverter = root
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateWriteSupport.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateWriteSupport.kt
new file mode 100644
index 00000000..e2f3df31
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceStateWriteSupport.kt
@@ -0,0 +1,105 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.hadoop.conf.Configuration
+import org.apache.parquet.hadoop.api.WriteSupport
+import org.apache.parquet.io.api.Binary
+import org.apache.parquet.io.api.RecordConsumer
+import org.apache.parquet.schema.*
+
+/**
+ * Support for writing [Resource] instances to Parquet format.
+ */
+internal class ResourceStateWriteSupport : WriteSupport<ResourceState>() {
+    /**
+     * The current active record consumer.
+     */
+    private lateinit var recordConsumer: RecordConsumer
+
+    override fun init(configuration: Configuration): WriteContext {
+        return WriteContext(WRITE_SCHEMA, emptyMap())
+    }
+
+    override fun prepareForWrite(recordConsumer: RecordConsumer) {
+        this.recordConsumer = recordConsumer
+    }
+
+    override fun write(record: ResourceState) {
+        write(recordConsumer, record)
+    }
+
+    private fun write(consumer: RecordConsumer, record: ResourceState) {
+        consumer.startMessage()
+
+        consumer.startField("id", 0)
+        consumer.addBinary(Binary.fromCharSequence(record.id))
+        consumer.endField("id", 0)
+
+        consumer.startField("timestamp", 1)
+        consumer.addLong(record.timestamp.toEpochMilli())
+        consumer.endField("timestamp", 1)
+
+        consumer.startField("duration", 2)
+        consumer.addLong(record.duration.toMillis())
+        consumer.endField("duration", 2)
+
+        consumer.startField("cpu_count", 3)
+        consumer.addInteger(record.cpuCount)
+        consumer.endField("cpu_count", 3)
+
+        consumer.startField("cpu_usage", 4)
+        consumer.addDouble(record.cpuUsage)
+        consumer.endField("cpu_usage", 4)
+
+        consumer.endMessage()
+    }
+
+    companion object {
+        /**
+         * Parquet schema for the "resource states" table in the trace.
+         */
+        @JvmStatic
+        val WRITE_SCHEMA: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("timestamp"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("duration"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("cpu_count"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.DOUBLE)
+                    .named("cpu_usage")
+            )
+            .named("resource_state")
+    }
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceWriteSupport.kt b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceWriteSupport.kt
new file mode 100644
index 00000000..14cadabb
--- /dev/null
+++ b/opendc-trace/opendc-trace-opendc/src/main/kotlin/org/opendc/trace/opendc/parquet/ResourceWriteSupport.kt
@@ -0,0 +1,114 @@
+/*
+ * Copyright (c) 2022 AtLarge Research
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+package org.opendc.trace.opendc.parquet
+
+import org.apache.hadoop.conf.Configuration
+import org.apache.parquet.hadoop.api.WriteSupport
+import org.apache.parquet.io.api.Binary
+import org.apache.parquet.io.api.RecordConsumer
+import org.apache.parquet.schema.*
+import kotlin.math.roundToLong
+
+/**
+ * Support for writing [Resource] instances to Parquet format.
+ */
+internal class ResourceWriteSupport : WriteSupport<Resource>() {
+    /**
+     * The current active record consumer.
+     */
+    private lateinit var recordConsumer: RecordConsumer
+
+    override fun init(configuration: Configuration): WriteContext {
+        return WriteContext(WRITE_SCHEMA, emptyMap())
+    }
+
+    override fun prepareForWrite(recordConsumer: RecordConsumer) {
+        this.recordConsumer = recordConsumer
+    }
+
+    override fun write(record: Resource) {
+        write(recordConsumer, record)
+    }
+
+    private fun write(consumer: RecordConsumer, record: Resource) {
+        consumer.startMessage()
+
+        consumer.startField("id", 0)
+        consumer.addBinary(Binary.fromCharSequence(record.id))
+        consumer.endField("id", 0)
+
+        consumer.startField("start_time", 1)
+        consumer.addLong(record.startTime.toEpochMilli())
+        consumer.endField("start_time", 1)
+
+        consumer.startField("stop_time", 2)
+        consumer.addLong(record.stopTime.toEpochMilli())
+        consumer.endField("stop_time", 2)
+
+        consumer.startField("cpu_count", 3)
+        consumer.addInteger(record.cpuCount)
+        consumer.endField("cpu_count", 3)
+
+        consumer.startField("cpu_capacity", 4)
+        consumer.addDouble(record.cpuCapacity)
+        consumer.endField("cpu_capacity", 4)
+
+        consumer.startField("mem_capacity", 5)
+        consumer.addLong(record.memCapacity.roundToLong())
+        consumer.endField("mem_capacity", 5)
+
+        consumer.endMessage()
+    }
+
+    companion object {
+        /**
+         * Parquet schema for the "resources" table in the trace.
+         */
+        @JvmStatic
+        val WRITE_SCHEMA: MessageType = Types.buildMessage()
+            .addFields(
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.BINARY)
+                    .`as`(LogicalTypeAnnotation.stringType())
+                    .named("id"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("start_time"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .`as`(LogicalTypeAnnotation.timestampType(true, LogicalTypeAnnotation.TimeUnit.MILLIS))
+                    .named("stop_time"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT32)
+                    .named("cpu_count"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.DOUBLE)
+                    .named("cpu_capacity"),
+                Types
+                    .required(PrimitiveType.PrimitiveTypeName.INT64)
+                    .named("mem_capacity"),
+            )
+            .named("resource")
+    }
+}
diff --git a/opendc-trace/opendc-trace-opendc/src/test/kotlin/org/opendc/trace/opendc/OdcVmTraceFormatTest.kt b/opendc-trace/opendc-trace-opendc/src/test/kotlin/org/opendc/trace/opendc/OdcVmTraceFormatTest.kt
index c8742624..dec0fef9 100644
--- a/opendc-trace/opendc-trace-opendc/src/test/kotlin/org/opendc/trace/opendc/OdcVmTraceFormatTest.kt
+++ b/opendc-trace/opendc-trace-opendc/src/test/kotlin/org/opendc/trace/opendc/OdcVmTraceFormatTest.kt
@@ -29,7 +29,9 @@ import org.junit.jupiter.api.assertThrows
 import org.junit.jupiter.params.ParameterizedTest
 import org.junit.jupiter.params.provider.ValueSource
 import org.opendc.trace.conv.*
+import java.nio.file.Files
 import java.nio.file.Paths
+import java.time.Instant
 
 /**
  * Test suite for the [OdcVmTraceFormat] implementation.
@@ -78,6 +80,37 @@ internal class OdcVmTraceFormatTest {
         reader.close()
     }
 
+    @Test
+    fun testResourcesWrite() {
+        val path = Files.createTempDirectory("opendc")
+        val writer = format.newWriter(path, TABLE_RESOURCES)
+
+        writer.startRow()
+        writer.set(RESOURCE_ID, "1019")
+        writer.set(RESOURCE_START_TIME, Instant.EPOCH)
+        writer.set(RESOURCE_STOP_TIME, Instant.EPOCH)
+        writer.setInt(RESOURCE_CPU_COUNT, 1)
+        writer.setDouble(RESOURCE_CPU_CAPACITY, 1024.0)
+        writer.setDouble(RESOURCE_MEM_CAPACITY, 1024.0)
+        writer.endRow()
+        writer.close()
+
+        val reader = format.newReader(path, TABLE_RESOURCES)
+
+        assertAll(
+            { assertTrue(reader.nextRow()) },
+            { assertEquals("1019", reader.get(RESOURCE_ID)) },
+            { assertEquals(Instant.EPOCH, reader.get(RESOURCE_START_TIME)) },
+            { assertEquals(Instant.EPOCH, reader.get(RESOURCE_STOP_TIME)) },
+            { assertEquals(1, reader.getInt(RESOURCE_CPU_COUNT)) },
+            { assertEquals(1024.0, reader.getDouble(RESOURCE_CPU_CAPACITY)) },
+            { assertEquals(1024.0, reader.getDouble(RESOURCE_MEM_CAPACITY)) },
+            { assertFalse(reader.nextRow()) },
+        )
+
+        reader.close()
+    }
+
     @ParameterizedTest
     @ValueSource(strings = ["trace-v2.0", "trace-v2.1"])
     fun testSmoke(name: String) {
@@ -95,6 +128,33 @@ internal class OdcVmTraceFormatTest {
     }
 
     @Test
+    fun testResourceStatesWrite() {
+        val path = Files.createTempDirectory("opendc")
+        val writer = format.newWriter(path, TABLE_RESOURCE_STATES)
+
+        writer.startRow()
+        writer.set(RESOURCE_ID, "1019")
+        writer.set(RESOURCE_STATE_TIMESTAMP, Instant.EPOCH)
+        writer.setDouble(RESOURCE_STATE_CPU_USAGE, 23.0)
+        writer.setInt(RESOURCE_CPU_COUNT, 1)
+        writer.endRow()
+        writer.close()
+
+        val reader = format.newReader(path, TABLE_RESOURCE_STATES)
+
+        assertAll(
+            { assertTrue(reader.nextRow()) },
+            { assertEquals("1019", reader.get(RESOURCE_ID)) },
+            { assertEquals(Instant.EPOCH, reader.get(RESOURCE_STATE_TIMESTAMP)) },
+            { assertEquals(1, reader.getInt(RESOURCE_CPU_COUNT)) },
+            { assertEquals(23.0, reader.getDouble(RESOURCE_STATE_CPU_USAGE)) },
+            { assertFalse(reader.nextRow()) },
+        )
+
+        reader.close()
+    }
+
+    @Test
     fun testInterferenceGroups() {
         val path = Paths.get("src/test/resources/trace-v2.1")
         val reader = format.newReader(path, TABLE_INTERFERENCE_GROUPS)