Details
-
Bug
-
Resolution: Fixed
-
Critical
-
5.5.0
-
Untriaged
-
Centos 64-bit
-
Yes
Description
Throughput degraded because of:
MB-22010: Change the default eviction policy to statistical_counter
Setup:
- 2 nodes
- 1 bucket
- 100M items
Steps:
bin/ycsb load couchbase2 -P workloads/workloadd -p writeallfields=true -threads 25 -p couchbase.host=triton-srv-01-ip6.perf.couchbase.com -p couchbase.bucket=bucket-1 -p couchbase.upsert=true -p couchbase.epoll=true -p couchbase.boost=48 -p exportfile=ycsb_load_0.log -p couchbase.sslMode=none -p couchbase.certKeystoreFile=../ -p couchbase.certKeystorePassword= -p couchbase.password=password -p recordcount=100000000
|
bin/ycsb run couchbase2 -P workloads/workloadd -p writeallfields=true -threads 160 -p couchbase.host=triton-srv-01-ip6.perf.couchbase.com -p couchbase.bucket=bucket-1 -p couchbase.upsert=true -p couchbase.epoll=true -p couchbase.boost=48 -p exportfile=ycsb_run_0.log -p couchbase.sslMode=none -p couchbase.certKeystoreFile=../certificates/data.keystore -p couchbase.certKeystorePassword=storepass -p couchbase.password=password -p operationcount=1000000000 -p maxexecutiontime=900 -p recordcount=100000000
|
Timeline and logs: https://bit.ly/2GoORqM
(There'a also an earlier ~10% regression due to MB-25179 - tracked by MB-29046).
Attachments
Issue Links
- relates to
-
MB-29046 Throughput of YCSB Workload D dropped by 10% due to connection buffer recycling
- Closed