[erlang-questions] Low disk logging performance in SMP
Sun Apr 26 09:21:36 CEST 2009
I guess this is similar to the findings in the blogs below. I didn't
know selective receives are so slow but apparently you are right, sync
processing is much better for these kind of situations. I always
thought this is due to bad IO performance.
Having said that, I'm curious of a solution to the selective receive
slowness in erlang. Does any one think a solution like appending
messages to BDB-style persistent queues or memcacheq helps? This is
particularly useful in cases where high load could lead to large
queues and possibly a crash. I've seen such a solution in a telecom
application which is working quite fine.
On Sat, Apr 25, 2009 at 1:05 AM, Per Melin <> wrote:
> Hynek Vychodil:
>> I have no problem with mine version, it does about 5MB/s for me.
> I would expect your version to be fast, because you empty the message
> queue before you write to disk and thus avoids any problems caused by
> selective recieves in the file modules.
> erlang-questions mailing list
More information about the erlang-questions