On 13.03.2012 23:44, Peter Gaughran wrote:
Hi folks,
We're in a large-number-of-users, high bandwidth/usage situation
(average 80 gigs per hour during business hours) and so have opted
for
a couple of new proxies (one for fail-over) which we're about to make
'live'. Currently, our cache_mem and cache_dir look like the
following
cache_mem 23424 MB
cache_dir aufs /cache1 183000 32 256
cache_dir aufs /cache2 183000 32 256
cache_dir aufs /cache3 183000 32 256
cache_dir aufs /cache4 183000 32 256
With internal testing (from a team of 5!), of course, everything is
fine. I'm concerned about what will happen when we roll them out, or
if there's something (blindingly) obvious we might have missed or
that
could be tweaked.
Don't forget the 10-15 MB of RAM index per GB of total cache space.
(~11 GB RAM index).
Each of those live transactions making up those ~20MBps will also
consume a ~64KB or more of RAM.
All of that *plus* cache_mem needs to fit into the available box RAM.
By my rough calc that is over 35GB of RAM needed on each box.
Amos
