-
Notifications
You must be signed in to change notification settings - Fork 138
Distributed Channel Index Performance Testing
Adam Fraser edited this page Sep 16, 2015
·
3 revisions
Proposed performance testing matrix, ordered by priority:
Topology (gatling x SG x CBS) | Scenario | Load (writers/readers per gatling node) | Description |
---|---|---|---|
1 x 2 x 3 | gateload, in-memory cache | 5K/8K | Baseline |
1 x 2 x 3 | gateload, distributed index | 5K/8K | Distributed index baseline |
1 x 2 x 3 | gateload, distributed index | 10K/16K | Higher load on baseline |
1 x 5 x 5 | gateload, distributed index | 5K/8K | Large cluster. |
1 x 20 x 5 | gateload, distributed index | 5K/8K | Larger SG cluster - tests higher contention on writes. |
1 x 10 x 10 | gateload, distributed index | 20K/20K | Ramping up load, cluster size |
1 x 2 x 3 | access, in-memory cache | 5K/8K | Baseline |
1 x 2 x 3 | access, distributed index | 5K/8K | Distributed index baseline |
Notes:
- Haven't yet described tests where only a subset of the SG nodes are identified as index writers. In those scenarios, gateload traffic should be directed to the non-index-writer nodes.