Dear Folks, Using head of imaptest with dovecot 2.1, I am attempting to put a heavy load on the server, with insufficient success. I made 15000 user accounts, put them in to a file, one per line. Then I ran imaptest with ./imaptest userfile=../../imap-test-userlist-15001.txt clients=15001 pass=SECRETPASSWORD But CPU load is only reaching 20. I put an mbox containing 67 messages into ~/mail/dovecot-crlf So what is the best way to put a really heavy load on an imap server? The documentation describes testing correctness of the server without really emphasising performance. I would like to do something like dnsperf, which adds a linearly increasing load until the server is unable to cope. This machine has 24G RAM and 8 cores. I'll be grateful for any constructive suggestions. -- Nick Urbanik http://nicku.org nicku at nicku.org GPG: 7FFA CDC7 5A77 0558 DC7A 790A 16DF EC5B BB9D 2C24 ID: BB9D2C24
On 16.3.2012, at 12.08, Nick Urbanik wrote:> Using head of imaptest with dovecot 2.1, I am attempting to put a > heavy load on the server, with insufficient success. > > I made 15000 user accounts, put them in to a file, one per line. Then > I ran imaptest with > > ./imaptest userfile=../../imap-test-userlist-15001.txt clients=15001 > pass=SECRETPASSWORDA single imaptest process can't handle that many simultaneous clients. You'd need to run multiple imaptests in parallel.> So what is the best way to put a really heavy load on an imap server? > The documentation describes testing correctness of the server without > really emphasising performance.Yes, imaptest is mainly meant to test server correctness (i.e. for me to test that Dovecot is bugfree). It spends a lot of time checking and tracking things that is irrelevant when you simply want to load the server. You could add no_tracking parameter to get rid of some of it.
On 2012-03-16 9:17 AM, Timo Sirainen <tss at iki.fi> wrote:> imaptest is mainly meant to test server correctness (i.e. for me to > test that Dovecot is bugfree). It spends a lot of time checking and > tracking things that is irrelevant when you simply want to load the > server. You could add no_tracking parameter to get rid of some of it.Maybe imaptest could be duped+modified somehow to produce a new imap_load_test utility...? -- Best regards, Charles