Details
-
Bug
-
Resolution: Fixed
-
Critical
-
3.0
-
Security Level: Public
-
Build 3.0.0-805
Platform = Physical
OS = CentOS 6.5
CPU = Intel Xeon E5-2680 v2 (40 vCPU)
Memory = 256 GB
Disk = RAID 10 SSD
-
Untriaged
-
Centos 64-bit
-
-
Yes
-
June 30 - July 18
Description
Originally reported as "~2400% CPU consumption by memcached during ongoing workload with five (5) buckets ".
The CPU usage for memcached process is more than two times the usage in the previous release. This is due to increased scheduling overhead from the shared thread pool.
Workaround: Reduce the number of threads on systems that have more than 30 cores.
2 nodes, 5 buckets
1M docs (clusterwise), equally distributed, non-DGM
10K mixed ops/sec (85% reads, 1% creates, 1% deletes, 13% updates; clusterwise), equally distributed
CPU utilization in 2.5.1: ~300%
CPU utilization in 3.0.0: ~2400%
Attachments
Issue Links
| For Gerrit Dashboard: MB-11405 | ||||||
|---|---|---|---|---|---|---|
| # | Subject | Branch | Project | Status | CR | V |
| 38337,4 | MB-11405 optimize by avoiding string constructions | master | ep-engine | Status: MERGED | +2 | +1 |
| 38620,3 | MB-11405 Reduce CPU use in active system | master | ep-engine | Status: MERGED | +2 | +1 |
| 38756,8 | MB-11405 Wake only as many threads as ready tasks | master | ep-engine | Status: MERGED | +2 | +1 |
| 39475,7 | MB-11405 MB-11434 limit max writers to 8 (configurable) | master | ep-engine | Status: MERGED | +2 | +1 |
| 40080,6 | MB-11405 Dedicated Threads that poll & sleep on own Q only | master | ep-engine | Status: MERGED | +2 | +1 |
| 40084,10 | MB-11405 Spawn at least 10 global threads | master | ep-engine | Status: MERGED | +2 | +1 |
| 192732,4 | MB-56032 [1/3]: Increase AuxIO default thread limit from 8 to 128 | neo | kv_engine | Status: MERGED | +2 | +1 |