On Apr 11, 2005, at 1:52 AM, Marc G. Fournier wrote:

> Does anyone have any #s on how fast Slony works?
>

In normal operation, I cannot make slony fall behind more than about 15 
seconds where the boxes are connected over Gig ethernet.  This is with 
a whole heck of a lot of inserts and updates happening all the time, 
and a lot of querying happening on the replica.

The key is very fast I/O systems (ie, no Dell, Compaq, etc.)  At one 
time I had a "slow" machine which would regularly fall behind by 10 or 
30 minutes with no queries running on it -- just the replication for 
backup purposes.

When I do my monthly "old data purge", the replica (currently only one 
-- about to add the second at the end of this week) can fall behind by 
about 20 or 25 minutes.  The data purge deletes perhaps several 
thousand rows in one table, but the rows referencing these rows in 
other tables are cascade-deleted, and those add up to several millions 
of rows.  I do it in small transactions since Pg tends to run out of 
memory when tracking so many locks for the FKs...

Vivek Khera, Ph.D.
+1-301-869-4449 x806

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2476 bytes
Desc: not available
Url : http://gborg.postgresql.org/pipermail/slony1-general/attachments/20050412/b2bc88e2/smime.bin