benchmarking: gearman, couchdb and redis
Translation of the original in Catalan. Read it in: CatalàCastellano
These are not serious and strict performance tests, but at least in my case they have served me to get an idea of the performance of these applications and to be able to design different architectures with a bit more knowledge of the subject.
In case you don't know the tools:
- gearman: task server
- couchdb: non-relational database system
- redis: sistema de caché similar a memcached, but much better from my point of view
System on which the tests have been made:
- HP ML110 G5 - Xeon 2GHz - 4GB RAM - HD via NFS
- Disk performance: Timing buffered disk reads: 26 MB in 3.00 seconds = 8.66 MB/sec
- SO Hypervisor: VMWare ESXi 3.5
- Virtual server: 1 CPU 2GHz and 512Mb RAM
- SO Guest: Ubuntu 8.04 Hardy
Test results:
- gearman client, makes 5.000 requests to the server:
- gearman backend: default, non-persistent queue
- cmd: gearmand -vvv -u root
- time: ~32s - performance: ~156req/s
- gearman backend: sqlite3, persistent queue
- cmd: gearmand -vvv -u root --libsqlite3-db=/tmp/gearman_sqlite3.cache -q libsqlite3
- time: ~11m10s - performance: ~0.8req/s
- gearman backend: tokyo cabinet btree, persistent queue
- cmd: gearmand -q libtokyocabinet --libtokyocabinet-file=/tmp/gearmand.tcb -vvv -u root
- time: ~2m3s - performance: ~40req/s
- gearman backend: tokyo cabinet hash, persistent queue
- cmd: gearmand -q libtokyocabinet --libtokyocabinet-file=/tmp/gearmand.tch -vvv -u root
- time: ~2m5s - performance: ~40req/s
- gearman backend: tokyo cabinet RAM, non-persistent queue
- cmd: gearmand -q libtokyocabinet --libtokyocabinet-file="*" -vvv -u root
- time: ~17s - performance: ~294req/s
- gearman backend: default, non-persistent queue
- we insert 5.000 documents into couchdb:
- time: ~14s - performance: ~357req/s
- redis we do 10.000 operations of type:
- SET: time: ~0.35s - performance: ~28.375req/s
- GET: time: ~0.59s - performance: ~16.920req/s
- PING: time: ~0.33s - performance: ~30.471req/s