add slicing

Author: Graeme22

README.md

@@ -0,0 +1 @@
+# rlskedge

cluster.jl

@@ -10,7 +10,7 @@ using .job
 
 export ClusterEnv, QUEUE_SIZE, ZONES
 
-const QUEUE_SIZE, ZONES = 128, 32
+const QUEUE_SIZE, ZONES, SLICE_SIZE = 128, 32, 50_000
 
 struct Metrics
     avg_bounded_slowdown::Float32
@@ -23,6 +23,7 @@ mutable struct ClusterEnv <: AbstractEnv
     workload::Workload
     time::Int
     next_job_index::Int
+    last_job_index::Int
     reward::Float32
     done::Bool
     cluster::Vector{Job}
@@ -33,15 +34,19 @@ mutable struct ClusterEnv <: AbstractEnv
     metrics::Union{Nothing, Metrics}
 end
 
+choose_index(wl::Workload) = rand(1:length(wl.jobs) - SLICE_SIZE)
+
 function RLBase.reset!(env::ClusterEnv)
-    #workload = Workload(rand(WORKLOADS))
-    #env.workload = workload
+    workload = Workload(rand(WORKLOADS))
+    env.workload = workload
+    index = choose_index(workload)
     env.time = env.workload.jobs[1].submit_time
-    env.next_job_index = 2
+    env.next_job_index = index + 1
+    env.last_job_index = index + SLICE_SIZE
     env.reward = 0
     env.done = false
     env.cluster = []
-    env.queue = [env.workload.jobs[1]]
+    env.queue = [env.workload.jobs[index]]
     env.are_pending_jobs = true
     env.available_cores = env.workload.cores
     env.utilization = []
@@ -50,34 +55,20 @@ end
 
 function ClusterEnv()
     workload = Workload(rand(WORKLOADS))
+    index = choose_index(workload)
     time = workload.jobs[1].submit_time
-    next_job_index = 2
-    reward = 0
-    done = false
-    cluster = []
-    queue = [workload.jobs[1]]
-    are_pending_jobs = true
-    available_cores = workload.cores
-    utilization = []
-    metrics = nothing
-
-    ClusterEnv(workload, time, next_job_index, reward, done, cluster, queue, are_pending_jobs, available_cores, utilization, metrics)
-end
-
-function ClusterEnv(index)
-    workload = Workload(WORKLOADS[index])
-    time = workload.jobs[1].submit_time
-    next_job_index = 2
+    next_job_index = index + 1
+    last_job_index = index + SLICE_SIZE
     reward = 0
     done = false
     cluster = []
-    queue = [workload.jobs[1]]
+    queue = [workload.jobs[index]]
     are_pending_jobs = true
     available_cores = workload.cores
     utilization = []
     metrics = nothing
 
-    ClusterEnv(workload, time, next_job_index, reward, done, cluster, queue, are_pending_jobs, available_cores, utilization, metrics)
+    ClusterEnv(workload, time, next_job_index, last_job_index, reward, done, cluster, queue, are_pending_jobs, available_cores, utilization, metrics)
 end
 
 RLBase.action_space(env::ClusterEnv) = Base.OneTo(QUEUE_SIZE)
@@ -214,7 +205,7 @@ function (env::ClusterEnv)(action)
         end
 
         # check for termination
-        if env.next_job_index > length(env.workload.jobs)  # no more pending jobs!
+        if env.next_job_index > env.last_job_index  # no more pending jobs!
             env.are_pending_jobs = false
         end
         # we've finished!
@@ -233,8 +224,8 @@ function (env::ClusterEnv)(action)
             bslds = [max((j.simulated_wait_time + j.simulated_run_time) / max(j.simulated_run_time, 10), 1) for j in env.workload.jobs]
             sum_bslds = +(bslds...)
             avg_bsld = sum_bslds / length(env.workload.jobs)
-            # currently: negative average bounded slowdown
-            env.reward = -avg_bsld
+            # currently: negative average bounded slowdown relative to SJF
+            env.reward = env.workload.sjf_bsld - avg_bsld
             env.done = true
             env.metrics = Metrics(avg_bsld, avg_wait_time, max_wait_time, avg_utilization)
             break

data/ANL-Intrepid-2009-1.swf

@@ -17,14 +17,14 @@
 ; fcfs_average_wait: 15236.789f0
 ; fcfs_max_wait: 260871
 ; fcfs_utilization: 0.403677f0
-; rand_bounded_slowdown: ?
-; rand_average_wait: ?
-; rand_max_wait: ?
-; rand_utilization: ?
-; sjf_bounded_slowdown: ?
-; sjf_average_wait: ?
-; sjf_max_wait: ?
-; sjf_utilization: ?
+; rand_bounded_slowdown: 33.204773f0
+; rand_average_wait: 29764.49f0
+; rand_max_wait: 880932
+; rand_utilization: 0.4077924f0
+; sjf_bounded_slowdown: 6.728023f0
+; sjf_average_wait: 6257.9917f0
+; sjf_max_wait: 812701
+; sjf_utilization: 0.40214235f0
 
 ; Note: Scheduler is Cobalt (http://trac.mcs.anl.gov/projects/cobalt/)
     1  0  6680  7560  2048  -1  -1  2048  10800  -1  -1  1  -1  -1  1  -1  -1  -1

data/CEA-Curie-2011-2.1-cln.swf

@@ -77,14 +77,14 @@
 ; fcfs_average_wait: 23738.648f0
 ; fcfs_max_wait: 181225
 ; fcfs_utilization: 0.37693802f0
-; rand_bounded_slowdown: ?
-; rand_average_wait: ?
-; rand_max_wait: ?
-; rand_utilization: ?
-; sjf_bounded_slowdown: ?
-; sjf_average_wait: ?
-; sjf_max_wait: ?
-; sjf_utilization: ?
+; rand_bounded_slowdown: 951.0125f0
+; rand_average_wait: 34300.434f0
+; rand_max_wait: 448601
+; rand_utilization: 0.37612095f0
+; sjf_bounded_slowdown: 340.9844f0
+; sjf_average_wait: 14785.063f0
+; sjf_max_wait: 1055787
+; sjf_utilization: 0.37800074f0
 
 272393 31656837  33492  86395  256     -1    -1  256  86400    -1  1  87  95  -1 -1  4 -1 -1
 272394 31664481 249861   1189 2048     -1    -1 2048  72000    -1  0  21  89  -1 -1  4 -1 -1

data/CTC-SP2-1996-3.1-cln.swf

@@ -23,18 +23,18 @@
 ; Queue:  3 piofs    
 ; Queue:  4 informix 
 
-; fcfs_bounded_slowdown: 679.31323f0
-; fcfs_average_wait: 23738.648f0
-; fcfs_max_wait: 181225
-; fcfs_utilization: 0.37693802f0
-; rand_bounded_slowdown: ?
-; rand_average_wait: ?
-; rand_max_wait: ?
-; rand_utilization: ?
-; sjf_bounded_slowdown: ?
-; sjf_average_wait: ?
-; sjf_max_wait: ?
-; sjf_utilization: ?
+; fcfs_bounded_slowdown: 7639.569f0
+; fcfs_average_wait: 1.3953139f6
+; fcfs_max_wait: 4140831
+; fcfs_utilization: 0.26490992f0
+; rand_bounded_slowdown: 2746.2961f0
+; rand_average_wait: 870947.9f0
+; rand_max_wait: 4445812
+; rand_utilization: 0.25893745f0
+; sjf_bounded_slowdown: 618.7664f0
+; sjf_average_wait: 237426.42f0
+; sjf_max_wait: 29790142
+; sjf_utilization: 0.24841218f0
 
 ;
 ; Note: THIS IS A CLEANED VERSION OF THE LOG!

data/HPC2N-2002-2.2-cln.swf

@@ -21,18 +21,18 @@
 ; Partition:  1  DEFAULT    
 ; Partition:  2  ALL    
 
-; fcfs_bounded_slowdown: 679.31323f0
-; fcfs_average_wait: 23738.648f0
-; fcfs_max_wait: 181225
-; fcfs_utilization: 0.37693802f0
-; rand_bounded_slowdown: ?
-; rand_average_wait: ?
-; rand_max_wait: ?
-; rand_utilization: ?
-; sjf_bounded_slowdown: ?
-; sjf_average_wait: ?
-; sjf_max_wait: ?
-; sjf_utilization: ?
+; fcfs_bounded_slowdown: 161.85031f0
+; fcfs_average_wait: 17225.17f0
+; fcfs_max_wait: 1164283
+; fcfs_utilization: 0.39946938f0
+; rand_bounded_slowdown: 129.56932f0
+; rand_average_wait: 15103.185f0
+; rand_max_wait: 1625164
+; rand_utilization: 0.39942858f0
+; sjf_bounded_slowdown: 127.50789f0
+; sjf_average_wait: 12791.929f0
+; sjf_max_wait: 1631664
+; sjf_utilization: 0.39929992f0
 
 ;
 ;