Forum OpenACS Q&A: Response to OpenACS scalability, and testing.

Collapse
Posted by Don Baccus on
That's quite a site - Slashdot themselves only get 25 hits/sec
during an average day (they mentioned this after September 11,
because on that day they were getting 50/60 hits/sec and were having
trouble keeping the site up).

This site right here stood up to a slashdotting just fine when Ben
published his "Why not MySQL?" paper and it got mentioned on
slashdot.  I'm not sure what it's running on today, back then it was
a dual P400 that was also Ben's server for development.  It barely
broke out a sweat but there wasn't much DB activity going on.

The big thing with PG is to VACUUM your tables nightly.  PG's
storage manager is a non-overwriting one so tables grow as you
update rows and don't automatically reclaim space.  VACUUM, then, is
a bit like garbage collection.  VACUUM in PG 7.1 and earlier
acquires an exclusive lock on each table it VACUUMs so can make a
site very slow while it is in progress.  Unless you're content's
huge, though, a VACUUM only takes a few minutes a day.

PG 7.2 will have two forms of VACUUM, one the current model and the
other a lighterweight model that won't acquire locks, just touching
non-busy pages.  It won't shrink files but will allow recycling of
dead space at much less cost than a full VACUUM.

Sourceforge has been running on PG for about a year.  Tim Purdue
(you can get his e-mail from the phpbuilder.com, I think) can
probably give you some good information regarding durability of a PG
installation.

They're apparently switching to Oracle but that seems to be because
of their desire to sell an enhanced "Enterprise Edition" sourceforge
to corporate clients and they think that Oracle is the only thing
that will sell there.