From 8d4d552e706ad5c5adebc774920337b4f201ac1f Mon Sep 17 00:00:00 2001 From: Fabian Mastenbroek Date: Sun, 12 Apr 2020 13:59:00 +0200 Subject: bug: Fix invalid FLOPs value for trace fragments This change fixes the invalid FLOPs values for the trace fragments which were multiplied twice by the number of cores of the VM. --- .../com/atlarge/opendc/compute/core/image/VmImage.kt | 2 +- .../atlarge/opendc/compute/virt/HypervisorEvent.kt | 4 ++++ .../opendc/compute/virt/driver/SimpleVirtDriver.kt | 20 ++++++++++---------- .../atlarge/opendc/experiments/sc20/Sc20Monitor.kt | 8 ++++++-- .../opendc/experiments/sc20/TestExperiment.kt | 2 ++ .../opendc/format/trace/sc20/Sc20TraceReader.kt | 16 +++++++++------- 6 files changed, 32 insertions(+), 20 deletions(-) (limited to 'opendc') diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/core/image/VmImage.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/core/image/VmImage.kt index 9ad88c17..b2a01804 100644 --- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/core/image/VmImage.kt +++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/core/image/VmImage.kt @@ -27,7 +27,7 @@ class VmImage( } else { val cores = min(fragment.cores, ctx.server.flavor.cpuCount) val burst = LongArray(cores) { fragment.flops / cores } - val usage = DoubleArray(cores) { fragment.usage } + val usage = DoubleArray(cores) { fragment.usage / cores } ctx.run(burst, usage, simulationContext.clock.millis() + fragment.duration) } diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/HypervisorEvent.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/HypervisorEvent.kt index 9ceb8bfc..7c088bc8 100644 --- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/HypervisorEvent.kt +++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/HypervisorEvent.kt @@ -60,6 +60,8 @@ public sealed class HypervisorEvent { * it did not have the capacity. * @property interferedBurst The sum of CPU time that virtual machines could not utilize due to performance * interference. + * @property cpuUsage CPU use in megahertz. + * @property cpuDemand CPU demand in megahertz. * @property numberOfDeployedImages The number of images deployed on this hypervisor. */ public data class SliceFinished( @@ -68,6 +70,8 @@ public sealed class HypervisorEvent { public val grantedBurst: Long, public val overcommissionedBurst: Long, public val interferedBurst: Long, + public val cpuUsage: Double, + public val cpuDemand: Double, public val numberOfDeployedImages: Int, public val hostServer: Server ) : HypervisorEvent() diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/driver/SimpleVirtDriver.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/driver/SimpleVirtDriver.kt index 5f15084d..cec9ce53 100644 --- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/driver/SimpleVirtDriver.kt +++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/driver/SimpleVirtDriver.kt @@ -246,14 +246,14 @@ class SimpleVirtDriver( // Divide the requests over the available capacity of the pCPUs fairly for (i in pCPUs) { - val remaining = hostContext.cpus.size - i - val availableShare = availableUsage / remaining - val grantedUsage = min(hostContext.cpus[i].frequency, availableShare) - val pBurst = ceil(duration * grantedUsage).toLong() + val maxCpuUsage = hostContext.cpus[i].frequency + val fraction = maxCpuUsage / maxUsage + val grantedUsage = min(maxCpuUsage, totalAllocatedUsage * fraction) + val grantedBurst = ceil(duration * grantedUsage).toLong() usage[i] = grantedUsage - burst[i] = pBurst - totalAllocatedBurst += pBurst + burst[i] = grantedBurst + totalAllocatedBurst += grantedBurst availableUsage -= grantedUsage } @@ -308,9 +308,7 @@ class SimpleVirtDriver( if (req.burst <= 0L || req.isCancelled) { hasFinished = true - } - - if (vm.deadline <= end && hostContext.server.state != ServerState.ERROR) { + } else if (vm.deadline <= end && hostContext.server.state != ServerState.ERROR) { // Request must have its entire burst consumed or otherwise we have overcommission // Note that we count the overcommissioned burst if the hypervisor has failed. totalOvercommissionedBurst += req.burst @@ -335,7 +333,9 @@ class SimpleVirtDriver( min(totalRequestedBurst, totalAllocatedBurst), min(totalRequestedBurst, totalGrantedBurst), // We can run more than requested due to timing totalOvercommissionedBurst, - totalInterferedBurst, // Might be smaller than zero due to FP rounding errors + totalInterferedBurst, // Might be smaller than zero due to FP rounding errors, + totalAllocatedUsage, + totalRequestedUsage, vmCount, // Some VMs might already have finished, so keep initial VM count server ) diff --git a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt index c0d6de03..245aa250 100644 --- a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt +++ b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/Sc20Monitor.kt @@ -17,7 +17,7 @@ class Sc20Monitor( private val lastServerStates = mutableMapOf>() init { - outputFile.write("time,duration,requestedBurst,grantedBurst,overcommissionedBurst,interferedBurst,numberOfDeployedImages,server,hostState,hostUsage,powerDraw,failedVms\n") + outputFile.write("time,duration,requestedBurst,grantedBurst,overcommissionedBurst,interferedBurst,cpuUsage,cpuDemand,numberOfDeployedImages,server,hostState,hostUsage,powerDraw,failedVms\n") } suspend fun onVmStateChanged(server: Server) {} @@ -32,6 +32,8 @@ class Sc20Monitor( 0, 0, 0, + 0.0, + 0.0, 0, server, duration @@ -49,6 +51,8 @@ class Sc20Monitor( grantedBurst: Long, overcommissionedBurst: Long, interferedBurst: Long, + cpuUsage: Double, + cpuDemand: Double, numberOfDeployedImages: Int, hostServer: Server, duration: Long = 5 * 60 * 1000L @@ -60,7 +64,7 @@ class Sc20Monitor( val usage = driver.usage.first() val powerDraw = driver.powerDraw.first() - outputFile.write("$time,$duration,$requestedBurst,$grantedBurst,$overcommissionedBurst,$interferedBurst,$numberOfDeployedImages,${hostServer.uid},${hostServer.state},$usage,$powerDraw") + outputFile.write("$time,$duration,$requestedBurst,$grantedBurst,$overcommissionedBurst,$interferedBurst,$cpuUsage,$cpuDemand,$numberOfDeployedImages,${hostServer.uid},${hostServer.state},$usage,$powerDraw") outputFile.newLine() } diff --git a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt index ede18b40..c75bde30 100644 --- a/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt +++ b/opendc/opendc-experiments-sc20/src/main/kotlin/com/atlarge/opendc/experiments/sc20/TestExperiment.kt @@ -201,6 +201,8 @@ fun main(args: Array) { event.grantedBurst, event.overcommissionedBurst, event.interferedBurst, + event.cpuUsage, + event.cpuDemand, event.numberOfDeployedImages, event.hostServer ) diff --git a/opendc/opendc-format/src/main/kotlin/com/atlarge/opendc/format/trace/sc20/Sc20TraceReader.kt b/opendc/opendc-format/src/main/kotlin/com/atlarge/opendc/format/trace/sc20/Sc20TraceReader.kt index 776cbc4e..3688864f 100644 --- a/opendc/opendc-format/src/main/kotlin/com/atlarge/opendc/format/trace/sc20/Sc20TraceReader.kt +++ b/opendc/opendc-format/src/main/kotlin/com/atlarge/opendc/format/trace/sc20/Sc20TraceReader.kt @@ -36,6 +36,7 @@ import java.io.BufferedReader import java.io.File import java.io.FileReader import java.util.UUID +import kotlin.math.max /** * A [TraceReader] for the internal VM workload trace format. @@ -82,7 +83,7 @@ class Sc20TraceReader( println(vmFile) val flopsHistory = mutableListOf() var vmId = "" - var cores = -1 + var maxCores = -1 var requiredMemory = -1L BufferedReader(FileReader(vmFile)).use { reader -> @@ -96,11 +97,13 @@ class Sc20TraceReader( vmId = vmFile.name val timestamp = (values[timestampCol].trim().toLong() - 5 * 60) * 1000L - cores = values[coreCol].trim().toInt() + val cores = values[coreCol].trim().toInt() val cpuUsage = values[cpuUsageCol].trim().toDouble() // MHz - requiredMemory = values[provisionedMemoryCol].trim().toLong() + requiredMemory = max(requiredMemory, values[provisionedMemoryCol].trim().toLong()) + maxCores = max(maxCores, cores) + + val flops: Long = (cpuUsage * 5 * 60).toLong() - val flops: Long = (cpuUsage * 5 * 60 * cores).toLong() if (flopsHistory.isEmpty()) { flopsHistory.add(FlopsHistoryFragment(timestamp, flops, traceInterval, cpuUsage, cores)) @@ -115,8 +118,7 @@ class Sc20TraceReader( oldFragment.flops + flops, oldFragment.duration + traceInterval, cpuUsage, - cores - ) + cores) ) } } @@ -136,7 +138,7 @@ class Sc20TraceReader( vmId, mapOf(IMAGE_PERF_INTERFERENCE_MODEL to relevantPerformanceInterferenceModelItems), flopsHistory, - cores, + maxCores, requiredMemory ) ) -- cgit v1.2.3