Pavel Voronkin created IGNITE-10681:
---------------------------------------
Summary: PME benchmarks become unstable at high number of
partitions per cache.
Key: IGNITE-10681
URL: https://issues.apache.org/jira/browse/IGNITE-10681
Project: Ignite
Issue Type: Improvement
Reporter: Pavel Voronkin
With increased number of partitions per cache (20-32k), some of the stages
happening during PME become very unstable both per node and per run. This must
be investigated because it blocks us from fine-tuning PME time optimizations
further.
Example: same configuration, same servers, same version (8.5.1-p150)
a run 2018-11-13:
{noformat}
Exchange [13,0] during LEAVE 1 server(s): 8942 msec, 8952 msec
New topology version after JOIN 1 server(s): 14
Exchange [14,0] during JOIN 1 server(s): 7558 msec, 8225 msec
Exchange [14,1] during JOIN 1 server(s): 19510 msec, 19562 msec
{noformat}
a run 2018-11-14:
{noformat}
Exchange [13, 0] during LEAVE 1 server(s): 14434 msec, 14448 msec
New topology version after JOIN 1 server(s): 14
Exchange [14, 1] during JOIN 1 server(s): 9089 msec, 9512 msec
Exchange [14, 0] during JOIN 1 server(s): 14455 msec, 14671 msec
{noformat}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
