<div dir='auto'><div><br><div class="gmail_extra"><br><div class="gmail_quote">Am 26.01.2023 21:25 schrieb Marc <Marc@f1-outsourcing.eu>:<br type="attribution"><blockquote class="quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><p dir="ltr">>
<br>
> Generally speaking the setup performs OK, I wrote a JMeter Roundcube script to
<br>
> test the Roundcube. This includes actions like listing mails and fetching
<br>
> them. I am hitting a somehow dubious limit of 600 req/s though. (This includes
<br>
> all kinds of HTTP calls to Roundcube, not only the ones triggering IMAP). When
<br>
> the dovecot becomes unavailable, the performance without mails jumps to 7700
<br>
> req/s. Testing is done with 4 JMeter-servers and one client.
<br>
>
<br>
<br>
Should you not be testing also dovecot performance directly? So you can see what is it's maximum. I can't imagine http interfaces are the bottle neck.
<br><br>
</p>
</blockquote></div>Right, I also set up imaptest and tested with that. I seem to hit a limit around 600 auth/s (auth workers 64, hashing algorithm does not seem to matter, tried all 3) and get about 1100 append/s and 1100 fetch/s.</div></div><div class="gmail_extra" dir="auto"><br></div><div class="gmail_extra" dir="auto">But I must say that the imaptest ran in parallel (10) for the append and fetch test on a single machine. Now that I think about it, the auth test may have been limited by the single imaptest instance, as I did not run that in parallel.</div></div>