feat: Fix failure duration parameters

3 files changed, 20 insertions, 9 deletions

diff --git a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
index cec1d1a7..2d95b489 100644
--- a/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
+++ b/opendc/opendc-compute/src/main/kotlin/com/atlarge/opendc/compute/metal/driver/SimpleBareMetalDriver.kt
@@ -307,11 +307,11 @@ public class SimpleBareMetalDriver(
         get() = domain
 
     override suspend fun fail() {
-        // serverContext?.unavailable = true
+        serverContext?.unavailable = true
     }
 
     override suspend fun recover() {
-        // serverContext?.unavailable = false
+        serverContext?.unavailable = false
     }
 
     override fun toString(): String = "SimpleBareMetalDriver(node = ${nodeState.value.uid})"
diff --git a/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/CorrelatedFaultInjector.kt b/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/CorrelatedFaultInjector.kt
index f363bf45..f46ce512 100644
--- a/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/CorrelatedFaultInjector.kt
+++ b/opendc/opendc-core/src/main/kotlin/com/atlarge/opendc/core/failure/CorrelatedFaultInjector.kt
@@ -90,7 +90,7 @@ public class CorrelatedFaultInjector(
                 ensureActive()
 
                 // Make sure to convert delay from hours to milliseconds
-                val d = lognvariate(iatScale, iatShape) * 3600 * 1e6
+                val d = lognvariate(iatScale, iatShape) * 3.6e6
 
                 // Handle long overflow
                 if (simulationContext.clock.millis() + d <= 0) {
@@ -99,14 +99,19 @@ public class CorrelatedFaultInjector(
 
                 delay(d.toLong())
 
+
                 val n = lognvariate(sizeScale, sizeShape).toInt()
                 val targets = active.shuffled(random).take(n)
+
+                println("[${simulationContext.clock.instant()}] FAIL $targets")
+
+
                 for (failureDomain in targets) {
                     active -= failureDomain
                     failureDomain.fail()
                 }
 
-                val df = lognvariate(dScale, dShape) * 3600 * 1e6
+                val df = lognvariate(dScale, dShape) * 6e4
 
                 // Handle long overflow
                 if (simulationContext.clock.millis() + df <= 0) {
@@ -115,8 +120,12 @@ public class CorrelatedFaultInjector(
 
                 delay(df.toLong())
 
+                println("[${simulationContext.clock.instant()}] RECOVER $targets")
+
                 for (failureDomain in targets) {
                     failureDomain.recover()
+
+                    // Re-enqueue machine to be failed
                     enqueue(failureDomain)
                 }
             }
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 400cef33..ca7e31ea 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
@@ -104,10 +104,11 @@ class ExperimentParameters(parser: ArgParser) {
  */
 fun createFaultInjector(domain: Domain, random: Random): FaultInjector {
     // Parameters from A. Iosup, A Framework for the Study of Grid Inter-Operation Mechanisms, 2009
+    // GRID'5000
     return CorrelatedFaultInjector(domain,
-        iatScale = -1.39, iatShape = 1.03,
+        iatScale = -1.39, iatShape = 1.03, // Hours
         sizeScale = 1.88, sizeShape = 1.25,
-        dScale = 1.88, dShape = 1.25,
+        dScale = 9.51, dShape = 3.21, // Minutes
         random = random
     )
 }
@@ -245,11 +246,12 @@ fun main(args: Array<String>) {
                                 monitor.onVmStateChanged(it.server)
 
                             // Detect whether the VM has finished running
-                            if (it.server.state == ServerState.ERROR || it.server.state == ServerState.SHUTOFF) {
+                            if (it.server.state == ServerState.SHUTOFF) {
                                 running -= server
+                            }
 
-                                if (running.isEmpty() && (!reader.hasNext() || availableHypervisors == 0))
-                                    finish.send(Unit)
+                            if (running.isEmpty() && !reader.hasNext()) {
+                                finish.send(Unit)
                             }
                         }
                         .collect()