Thanks, I will look into that. Hans
On Tue, Apr 28, 2015 at 1:24 AM, dormando <[email protected]> wrote: > You'll run into a few problems... -m 1 won't work the way you think since > it'll force-allocate one slab page per slab class. we don't have a > fallback mode to cut that down. > > We also don't do anything to detune the thread stack size (which is > probably 2mb still?), plus a handful of allocations around each worker > thread (the new event base, connection buffers, etc). > > That still seems sort of high, but thread stacks + etc would account for > quite a bit of it. You can also try reducing the size of the hash table.. > that could show up as VSZ if you haven't written anything into it yet. > > On Mon, 27 Apr 2015, Hans Kramer wrote: > > > Hi, > > I am using memcached on a system with very modest RAM: 64Mb to be > precise (an Artila Matrix 504). > > I have specified -M 1 -t 1 to reduce the memory allocated by memcached. > > Now RSS is always modest and around an explainable value, however, VSZ > is at least 20Mb and with every > > thread I add VSZ increases with at least 6 Mbytes. If I understand this > correctly, this relative "high" VSZ, increases the oom_score > > a lot. Hence I would like to reduce unnecessary alloc as much as > possible. > > I have briefly looked at the memcached source and I cannot any > significant mallocs. Is this perhaps coming from the libevent > > library? > > > > Any insights will be helpful. > > > > Kind regards > > > > Hans > > > > -- > > > > --- > > You received this message because you are subscribed to the Google > Groups "memcached" group. > > To unsubscribe from this group and stop receiving emails from it, send > an email to [email protected]. > > For more options, visit https://groups.google.com/d/optout. > > > > > -- --- You received this message because you are subscribed to the Google Groups "memcached" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
