[erlang-questions] large scale deployments and netsplits

Scott Lystig Fritchie <>
Wed Sep 16 22:02:14 CEST 2009


Joel Reymont <> wrote:

jr> I have 468,819 bots started on 100 small EC2 instances, 1 VM per
jr> instance.

Joel, just to rule out something that I know has bitten me (and is worth
mentioning to others in the community)...

... are the sizes of the Erlang VM processes big enough to run the risk
of paging virtual memory pages to disk?  If yes, then all bets are off
for timely responses to network distribution communication, including
liveness checks.

Many Linux kernels have a value of /proc/sys/vm/swappiness equal to 40.
That's just damn annoying for anyone who want to have a single OS
process grow to a significant fraction of total RAM without having some
of your RAM pages flushed to disk for "no good reason".  IMHO, most
Linux servers should have swappiness's value to 0, period.

-Scott


More information about the erlang-questions mailing list