> > The site http://slashdot.org ?[...]
I had two people ask me privately about Slashdot's setup so I'll
answer here for both of them (and anyone else who cares). At peak
hours, we send a bit over 100 requests/sec through Pound (image
requests go elsewhere), so I doubt we're one of the larger sites
using it.
We have a total of 16 webheads running Apache. Of those, 5 are also
running pound (5 just for redundancy). Pound takes about 4% of the
CPU on those machines; its resource set size stays around 8 MB
but its virtual memory size gets up around 50-90 MB. All this is on
Linux 2.4.
We have a hardware load balancer that round-robins all incoming HTTP
requests to the 5 pounds. If one goes down, for example when we
restart for a new version or new config of Pound, the LB stops
sending it requests pretty quickly. You may ask why we need Pound
if we already do LB in hardware; the reason is that Pound can
distribute requests to different machines more intelligently.
We're not using Pound's session tracking. Mainly we use it to
segregate traffic so if we have a resource crunch with e.g. heavy
comment reading, or a DDoS designed to wedge our search database,
the damage is limited to a subset of our webheads. So our pound.cfg
actually has 12 sections to it, but we're just filtering with
UrlGroup, and in one place we do a HeadRequire Cookie ".*user.*"
just to test whether there's a cookie for a logged-in user at all.[...]
|
