[Test] Improve scalability envelope (ray-project#14406)

* fixed. * fix. * Update the result. * Addressed code review.
daikeshi · Mar 2, 2021 · b1e0409 · b1e0409
1 parent eab53a8
commit b1e0409
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 10 deletions.
diff --git a/benchmarks/README.md b/benchmarks/README.md
@@ -8,7 +8,7 @@ All distributed tests are run on 64 nodes with 64 cores/node. Maximum number of
 
 | Dimension                                       | Quantity |
 | ---------                                       | -------- |
-| # nodes in cluster (with trivial task workload) | 250+     |
+| # nodes in cluster (with trivial task workload) | 1000+     |
 | # actors in cluster (with trivial workload)     | 10k+     |
 | # simultaneously running tasks                  | 10k+     |
 | # simultaneously running placement groups       | 1k+      |

diff --git a/benchmarks/distributed/config.yaml b/benchmarks/distributed/config.yaml
@@ -15,24 +15,24 @@ auth:
 available_node_types:
     head_node:
         node_config:
-            InstanceType: m5.16xlarge
-            ImageId: ami-098555c9b343eb09c 
+            InstanceType: r5dn.16xlarge # Network optimized.
+            ImageId: ami-0a2363a9cff180a64 
         resources:
+          CPU: 0
           node: 1
           small: 1
-        max_workers: 999999
     worker_node:
         node_config:
             InstanceType: m5.16xlarge
-            ImageId: ami-098555c9b343eb09c 
+            ImageId: ami-0a2363a9cff180a64
         resources:
           node: 1
-        min_workers: 63
-        max_workers: 63
+        min_workers: 64
+        max_workers: 64
     small_worker_node:
         node_config:
             InstanceType: m5.xlarge
-            ImageId: ami-098555c9b343eb09c 
+            ImageId: ami-0a2363a9cff180a64
         resources:
           node: 1
         max_workers: 999999

diff --git a/benchmarks/distributed/test_distributed.py b/benchmarks/distributed/test_distributed.py
@@ -6,11 +6,11 @@
 from time import sleep, perf_counter
 from tqdm import tqdm, trange
 
-TEST_NUM_NODES = 64
+TEST_NUM_NODES = 65
 MAX_ACTORS_IN_CLUSTER = 10000
 MAX_RUNNING_TASKS_IN_CLUSTER = 10000
 MAX_PLACEMENT_GROUPS = 1000
-MAX_NUM_NODES = 250
+MAX_NUM_NODES = 1000
 
 
 def num_alive_nodes():