Tune down the end-to-end runtime of tao_bench_standalone_mini

benchpress/config/jobs.yml

@@ -1157,21 +1157,22 @@
     - '--ipv4={ipv4}'
   vars:
     - 'num_servers=0'
-    - 'memsize=0.5'
+    - 'memsize=0.25'
     - 'fast_threads_ratio=0.75'
     - 'slow_to_fast_ratio=3'
     - 'warmup_time=0'
-    - 'test_time=30'
+    - 'test_time=15'
     - 'num_clients=2'
-    - 'clients_per_thread=0'
+    - 'clients_per_thread=190'
     - 'bind_cpu=1'
     - 'bind_mem=1'
-    - 'stats_interval=5000'
+    - 'stats_interval=1000'
     - 'port_number_start=11211'
     - 'disable_tls=0'
-    - 'timeout_buffer=30'
+    - 'timeout_buffer=10'
     - 'num_slow_threads=0'
     - 'num_fast_threads=0'
+    # 0 = scale the client with usable cores: max(1, cores-6, cores*0.8).
     - 'num_client_threads=0'
     - 'warmup_timeout_buffer=0'
     - 'test_timeout_buffer=0'

packages/tao_bench/README.md

@@ -202,6 +202,12 @@ minimal settings to enable quick testing and validation.
 A `breakdown.csv` file will be generated in the results to track runtime breakdowns,
 which can be used to filter metrics for the actual benchmark execution time.
 
+#### How to tune the memory usage
+
+`memsize` (the server cache) is the key knob that determines the memory usage of the benchmark.
+The default value is `0.25` which allocates about ~200MB memory.
+
+
 ### Result reporting
 
 Once the tao_bench benchmark finishes, benchpress will report the results in JSON format