feat: Use Weilbull distribution for failures

4 files changed, 18 insertions, 9 deletions

diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
index 9fad2396..0f38eca1 100644
--- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
+++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/virt/service/SimpleVirtProvisioningService.kt
@@ -96,7 +96,6 @@ class SimpleVirtProvisioningService(
             } catch (e: InsufficientMemoryOnServerException) {
                 println("Unable to deploy image due to insufficient memory")
             }
-
         }
     }
 
diff --git a/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/UncorrelatedFaultInjector.kt b/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/UncorrelatedFaultInjector.kt
index 5155a25a..56706824 100644
--- a/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/UncorrelatedFaultInjector.kt
+++ b/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/UncorrelatedFaultInjector.kt
@@ -24,26 +24,35 @@
 
 package com.atlarge.opendc.core.failure
 
+import com.atlarge.odcsim.simulationContext
 import kotlinx.coroutines.delay
 import kotlinx.coroutines.launch
-import kotlin.math.ln
+import kotlin.math.ln1p
+import kotlin.math.pow
 import kotlin.random.Random
 
 /**
  * A [FaultInjector] that injects uncorrelated faults into the system, meaning that failures of the subsystems are
  * independent.
  */
-public class UncorrelatedFaultInjector(private val mu: Double = 1024.0, private val random: Random = Random.Default) : FaultInjector {
+public class UncorrelatedFaultInjector(private val alpha: Double, private val beta: Double, private val random: Random = Random) : FaultInjector {
     /**
      * Enqueue the specified [FailureDomain] to fail some time in the future.
      */
     override fun enqueue(domain: FailureDomain) {
         domain.scope.launch {
-            val d = random.expovariate(mu)
-            delay(d)
+            val d = random.weibull(alpha, beta) * 1e3 // Make sure to convert delay to milliseconds
+            
+            // Handle long overflow
+            if (simulationContext.clock.millis() + d <= 0) {
+                return@launch
+            }
+
+            delay(d.toLong())
             domain.fail()
         }
     }
 
-    private fun Random.expovariate(mu: Double) = (-mu * ln(1 - nextDouble())).toLong()
+    // XXX We should extract this in some common package later on.
+    private fun Random.weibull(alpha: Double, beta: Double) = (beta * (-ln1p(-nextDouble())).pow(1.0 / alpha))
 }
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 d3b2d5c6..40cb9719 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
@@ -18,10 +18,11 @@ class Sc20Monitor(
     private var failed: Int = 0
 
     init {
-        outputFile.write("time,requestedBurst,grantedBurst,numberOfDeployedImages,server,hostUsage,powerDraw\n")
+        outputFile.write("time,requestedBurst,grantedBurst,numberOfDeployedImages,server,hostUsage,powerDraw,failedVms\n")
     }
 
     override suspend fun onUpdate(server: Server, previousState: ServerState) {
+        println("${simulationContext.clock.instant()} ${server.uid} ${server.state}")
         if (server.state == ServerState.ERROR) {
             failed++
         }
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 a1619fe2..69174f0f 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
@@ -89,7 +89,6 @@ fun main(args: Array<String>) {
         val provider = ServiceLoader.load(SimulationEngineProvider::class.java).first()
         val system = provider("test")
         val root = system.newDomain("root")
-
         val chan = Channel<Unit>(Channel.CONFLATED)
 
         root.launch {
@@ -118,7 +117,8 @@ fun main(args: Array<String>) {
 
             root.launch {
                 chan.receive()
-                val faultInjector = UncorrelatedFaultInjector(mu = 2e7)
+                // Parameters from A. Iosup, A Framework for the Study of Grid Inter-Operation Mechanisms, 2009
+                val faultInjector = UncorrelatedFaultInjector(alpha = 9.66772, beta = 12.23796)
                 for (node in bareMetalProvisioner.nodes()) {
                     faultInjector.enqueue(node.metadata["driver"] as FailureDomain)
                 }