Paul J Stevens wrote:
> Michael Monnerie wrote:
>
>> I make an imapsync to my mailserver, which connects
to the PostgreSQL
>> db, and I get these numbers:
>>
>> mailsrv -> db 15.482.309
>> db -> mailsrv 648.119.851
>>
>> client -> mailsrv 1.771.458
>> mailsrv -> client 15.956.176
>>
>> So with ~18MB IMAP traffic I produce ~663MB
database traffic? Can
>> somebody explain me this?
>
> Interesting figures. Can you produce a base-line? Like
say: traffic
> numbers for syncing a single account, containing a
single folder with a
> single message of a known size, blah blah. Log the
query patterns
> triggered, and the traffic numbers involved. Out of
curiosity. Perhaps
> the explanation will jump out if the logs.
>
> One hypothesis a priori: we don't cache as much as we
could. In fact,
> dbmail caches *very* little and will at times happily
call the same
> query a thousand times. And it doesn't even use
prepared statements (yet).
>
I also had a hard time migrating 20GB of mbox email to
dbmail, it actually
took about 12 days, 3 imapsync runs with wrappers alerting
me when things get
ugly (granted I had the emails evenly distributed between
accounts, so I could
arrange a staged migration).
The figures you are talking about can be easily derived
using my torture
script which I wrote to track down the 2.2 memory leaks some
months ago. You
guys even included it in the test-scripts section. On a
quick glance it will
need the following modifications:
* in init_db() function, so that all resulting emails will
be identical (now
the subject is an incrementing number)
* in torture() imapsync is supplied a skipheader that may no
longer be relevant
* at the end of the script the torture runs twice with and
without
G_SLICE=always-malloc which is irrelevant
* instead of t_memcheck() the dbmail-imapd daemon should be
started by t_normal()
Cheers!
_______________________________________________
DBmail mailing list
DBmail dbmail.org
htt
ps://mailman.fastxs.nl/mailman/listinfo/dbmail
|
