Hey folks,
I'm trying to set up a small ES cluster (2x m1.xlarge machines to handle
~4M small docs). Searches are generally low latency (<20ms), but I can just
barely get ~130qps without latency starting to go up and the load average
for each machine sitting at ~10. I should also mention there is about 20qps
of background /update requests, but without search load the cpu usage is
minimal.
Here's what my setup looks
like: https://gist.github.com/anonymous/5010665a1d1f6c5a8b7c . There's no
faceting going on and I think the entire index should be in kernel cache
(it's about 2G in size and the machine has 15G of RAM).
The reason username.parsed has such a weird analysis is for autocomplete.
In particular, given a username of "Dwight SchRute", I want any of the
following to match the doc:
- Dwight Schrute
- DwIght S
- rute
- schrute
but at the same time, I don't want "schrute dwight" to match. Regardless,
removing username.parsed from the query only yields an extra ~30qps.
Is this performance expected? Did I misconfigure something?
Thanks for the help,
Mike.
--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.