Support independent CPU and GPU workloads on the same node

buildSrc/src/main/kotlin/testing-conventions.gradle.kts

@@ -27,6 +27,10 @@ plugins {
 tasks.test {
     useJUnitPlatform()
 
+    minHeapSize = "512m"
+    maxHeapSize = "3072m"
+    jvmArgs = listOf("-XX:MaxMetaspaceSize=512m")
+
     reports {
         html.required.set(true)
         junitXml.required.set(true)
@@ -42,26 +46,26 @@ dependencies {
     testRuntimeOnly(versionCatalog["junit.jupiter.engine"])
 }
 
-tasks.register<Test>("testsOn18") {
-    javaLauncher.set(javaToolchains.launcherFor {
-        languageVersion.set(JavaLanguageVersion.of(18))
-    })
-
-    useJUnitPlatform()
-
-    minHeapSize = "512m"
-    maxHeapSize = "1024m"
-    jvmArgs = listOf("-XX:MaxMetaspaceSize=512m")
-}
-
-tasks.register<Test>("testsOn19") {
-    javaLauncher.set(javaToolchains.launcherFor {
-        languageVersion.set(JavaLanguageVersion.of(19))
-    })
-
-    useJUnitPlatform()
-
-    minHeapSize = "512m"
-    maxHeapSize = "1024m"
-    jvmArgs = listOf("-XX:MaxMetaspaceSize=512m")
-}
+//tasks.register<Test>("testsOn18") {
+//    javaLauncher.set(javaToolchains.launcherFor {
+//        languageVersion.set(JavaLanguageVersion.of(18))
+//    })
+//
+//    useJUnitPlatform()
+//
+//    minHeapSize = "512m"
+//    maxHeapSize = "1024m"
+//    jvmArgs = listOf("-XX:MaxMetaspaceSize=512m")
+//}
+//
+//tasks.register<Test>("testsOn19") {
+//    javaLauncher.set(javaToolchains.launcherFor {
+//        languageVersion.set(JavaLanguageVersion.of(19))
+//    })
+//
+//    useJUnitPlatform()
+//
+//    minHeapSize = "512m"
+//    maxHeapSize = "1024m"
+//    jvmArgs = listOf("-XX:MaxMetaspaceSize=512m")
+//}

opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/host/SimHost.kt

@@ -58,6 +58,7 @@ public class SimHost(
     private val engine: FlowEngine,
     private val machineModel: MachineModel,
     private val cpuPowerModel: CpuPowerModel,
+    private val accelPowerModel: CpuPowerModel,
     private val embodiedCarbon: Double,
     private val expectedLifetime: Double,
     private val powerDistributor: FlowDistributor,
@@ -136,6 +137,7 @@ public class SimHost(
                 this.machineModel,
                 this.powerDistributor,
                 this.cpuPowerModel,
+                this.accelPowerModel,
             ) { cause ->
                 hostState = if (cause != null) HostState.ERROR else HostState.DOWN
             }
@@ -340,6 +342,11 @@ public class SimHost(
         return guest.getCpuStats()
     }
 
+    public fun getAccelStats(task: ServiceTask): GuestCpuStats {
+        val guest = requireNotNull(taskToGuestMap[task]) { "Unknown task ${task.name} at host $name" }
+        return guest.getAccelStats()
+    }
+
     override fun hashCode(): Int = name.hashCode()
 
     override fun equals(other: Any?): Boolean {
@@ -352,7 +359,11 @@ public class SimHost(
      * Convert flavor to machine model.
      */
     private fun Flavor.toMachineModel(): MachineModel {
-        return MachineModel(simMachine!!.machineModel.cpuModel, MemoryUnit("Generic", "Generic", 3200.0, memorySize))
+        return MachineModel(
+            simMachine!!.machineModel.cpuModel,
+            simMachine!!.machineModel.accelModel,
+            MemoryUnit("Generic", "Generic", 3200.0, memorySize),
+        )
     }
 
     /**

opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/internal/Guest.kt

@@ -256,6 +256,25 @@ public class Guest(
         )
     }
 
+    /**
+     * Obtain the GPU statistics of this guest.
+     */
+    public fun getAccelStats(): GuestCpuStats {
+        virtualMachine!!.updateCounters(this.clock.millis())
+        val counters = virtualMachine!!.performanceCounters
+
+        return GuestCpuStats(
+            counters.cpuActiveTime / 1000L,
+            counters.cpuIdleTime / 1000L,
+            counters.cpuStealTime / 1000L,
+            counters.cpuLostTime / 1000L,
+            counters.cpuCapacity,
+            counters.cpuSupply,
+            counters.cpuDemand,
+            counters.cpuSupply / cpuLimit,
+        )
+    }
+
     /**
      * Helper function to track the uptime and downtime of the guest.
      */

opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/provisioner/HostsProvisioningStep.kt

@@ -127,6 +127,7 @@ public class HostsProvisioningStep internal constructor(
                         engine,
                         hostSpec.model,
                         hostSpec.cpuPowerModel,
+                        hostSpec.accelPowerModel,
                         hostSpec.embodiedCarbon,
                         hostSpec.expectedLifetime,
                         hostDistributor,

opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/telemetry/table/task/TaskTableReader.kt

@@ -107,6 +107,16 @@ public interface TaskTableReader : Exportable {
      */
     public val cpuDemand: Double
 
+    /**
+     * The GPU given to this task (in MHz).
+     */
+    public val accelUsage: Double
+
+    /**
+     * The GPU demanded by this task (in MHz).
+     */
+    public val accelDemand: Double
+
     /**
      * The duration (in seconds) that a CPU was active in the task.
      */

opendc-compute/opendc-compute-simulator/src/main/kotlin/org/opendc/compute/simulator/telemetry/table/task/TaskTableReaderImpl.kt

@@ -143,6 +143,14 @@ public class TaskTableReaderImpl(
         get() = _cpuDemand
     private var _cpuDemand = 0.0
 
+    override val accelUsage: Double
+        get() = _accelUsage
+    private var _accelUsage = 0.0
+
+    override val accelDemand: Double
+        get() = _accelDemand
+    private var _accelDemand = 0.0
+
     override val cpuActiveTime: Long
         get() = _cpuActiveTime - previousCpuActiveTime
     private var _cpuActiveTime = 0L
@@ -186,6 +194,7 @@ public class TaskTableReaderImpl(
         }
 
         val cpuStats = _host?.getCpuStats(task)
+        val accelStats = _host?.getAccelStats(task)
         val sysStats = _host?.getSystemStats(task)
 
         _timestamp = now
@@ -194,6 +203,8 @@ public class TaskTableReaderImpl(
         _cpuLimit = cpuStats?.capacity ?: 0.0
         _cpuDemand = cpuStats?.demand ?: 0.0
         _cpuUsage = cpuStats?.usage ?: 0.0
+        _accelDemand = accelStats?.demand ?: 0.0
+        _accelUsage = accelStats?.usage ?: 0.0
         _cpuActiveTime = cpuStats?.activeTime ?: _cpuActiveTime
         _cpuIdleTime = cpuStats?.idleTime ?: _cpuIdleTime
         _cpuStealTime = cpuStats?.stealTime ?: _cpuStealTime

opendc-compute/opendc-compute-topology/src/main/kotlin/org/opendc/compute/topology/TopologyFactories.kt

@@ -165,22 +165,51 @@ private fun HostJSONSpec.toHostSpec(clusterName: String): HostSpec {
             )
         }
 
+    val accelUnits =
+        if (accel == null) {
+            List(1) {
+                CpuModel(
+                    globalCoreId++,
+                    0,
+                    0.0,
+                )
+            }
+        } else {
+            List(accel.count) {
+                CpuModel(
+                    globalCoreId++,
+                    accel.coreCount,
+                    accel.coreSpeed.toMHz(),
+                )
+            }
+        }
+
     val unknownMemoryUnit = MemoryUnit(memory.vendor, memory.modelName, memory.memorySpeed.toMHz(), memory.memorySize.toMiB().toLong())
     val machineModel =
         MachineModel(
             units,
+            accelUnits,
             unknownMemoryUnit,
         )
 
     val powerModel =
         getPowerModel(powerModel.modelType, powerModel.power.toWatts(), powerModel.maxPower.toWatts(), powerModel.idlePower.toWatts())
 
+    val accelPowerModel =
+        getPowerModel(
+            accelPowerModel.modelType,
+            accelPowerModel.power.toWatts(),
+            accelPowerModel.maxPower.toWatts(),
+            accelPowerModel.idlePower.toWatts(),
+        )
+
     val hostSpec =
         HostSpec(
             createUniqueName(this.name, hostNames),
             clusterName,
             machineModel,
             powerModel,
+            accelPowerModel,
         )
     return hostSpec
 }

opendc-compute/opendc-compute-topology/src/main/kotlin/org/opendc/compute/topology/specs/HostSpec.kt

@@ -37,6 +37,7 @@ public data class HostSpec(
     val clusterName: String,
     val model: MachineModel,
     val cpuPowerModel: CpuPowerModel,
+    val accelPowerModel: CpuPowerModel,
     val embodiedCarbon: Double = 1000.0,
     val expectedLifetime: Double = 5.0,
 )

opendc-compute/opendc-compute-topology/src/main/kotlin/org/opendc/compute/topology/specs/TopologySpecs.kt

@@ -77,8 +77,10 @@ public data class ClusterJSONSpec(
 public data class HostJSONSpec(
     val name: String = "Host",
     val cpu: CPUJSONSpec,
+    val accel: CPUJSONSpec? = null,
     val memory: MemoryJSONSpec,
     val powerModel: PowerModelSpec = PowerModelSpec.DFLT,
+    val accelPowerModel: PowerModelSpec = PowerModelSpec.NONE,
     val count: Int = 1,
 )
 
@@ -139,6 +141,14 @@ public data class PowerModelSpec(
                 maxPower = Power.ofWatts(400.0),
                 idlePower = Power.ofWatts(200.0),
             )
+
+        public val NONE: PowerModelSpec =
+            PowerModelSpec(
+                modelType = "constant",
+                power = Power.ofWatts(0),
+                maxPower = Power.ofWatts(0),
+                idlePower = Power.ofWatts(0),
+            )
     }
 }
 

opendc-compute/opendc-compute-workload/src/main/kotlin/org/opendc/compute/workload/ComputeWorkloadLoader.kt

@@ -38,6 +38,7 @@ import org.opendc.trace.conv.resourceMemCapacity
 import org.opendc.trace.conv.resourceNature
 import org.opendc.trace.conv.resourceStateCpuUsage
 import org.opendc.trace.conv.resourceStateDuration
+import org.opendc.trace.conv.resourceStateIsGpu
 import org.opendc.trace.conv.resourceSubmissionTime
 import java.io.File
 import java.lang.ref.SoftReference
@@ -79,6 +80,7 @@ public class ComputeWorkloadLoader(
         val durationCol = reader.resolve(resourceStateDuration)
         val coresCol = reader.resolve(resourceCpuCount)
         val usageCol = reader.resolve(resourceStateCpuUsage)
+        val isGpuCol = reader.resolve(resourceStateIsGpu)
 
         val fragments = mutableMapOf<String, Builder>()
 
@@ -88,12 +90,13 @@ public class ComputeWorkloadLoader(
                 val durationMs = reader.getDuration(durationCol)!!
                 val cores = reader.getInt(coresCol)
                 val cpuUsage = reader.getDouble(usageCol)
+                val isGpu = reader.getBoolean(isGpuCol)
 
                 val builder =
                     fragments.computeIfAbsent(
                         id,
                     ) { Builder(checkpointInterval, checkpointDuration, checkpointIntervalScaling, scalingPolicy, id) }
-                builder.add(durationMs, cpuUsage, cores)
+                builder.add(durationMs, cpuUsage, cores, isGpu)
             }
 
             fragments
@@ -231,10 +234,11 @@ public class ComputeWorkloadLoader(
             duration: Duration,
             usage: Double,
             cores: Int,
+            isGpu: Boolean,
         ) {
             totalLoad += (usage * duration.toMillis()) / 1000 // avg MHz * duration = MFLOPs
 
-            builder.add(duration.toMillis(), usage, cores)
+            builder.add(duration.toMillis(), usage, cores, isGpu)
         }
 
         /**

opendc-experiments/opendc-experiments-base/src/main/kotlin/org/opendc/experiments/base/runner/ScenarioReplayer.kt

@@ -120,7 +120,8 @@ public suspend fun ComputeService.replay(
                 }
 
                 val workload = entry.trace
-                val meta = mutableMapOf<String, Any>("workload" to workload)
+//                val meta = mutableMapOf<String, Any>("workload" to workload)
+                val meta = mutableMapOf<String, Any>()
 
                 val nature =
                     if (entry.nature == "deferrable") {