[Systems] Memcached

David Farning dfarning at sugarlabs.org
Sat Mar 28 13:37:30 EDT 2009

On Sat, Mar 28, 2009 at 11:36 AM, Ivan Krstić
<krstic at solarsail.hcs.harvard.edu> wrote:
> On Mar 28, 2009, at 4:51 PM, David Farning wrote:
>> (among other things) is still growing pretty quickly.
> I'm not sure what you're trying to say.

I am trying to say that aslo is a memory hog and we should take that
into account when considering the the future infrastructure needs of
Sugar Labs.

 Memcached is a slab-allocated cache
> with a LRU ejection policy which is triggered only when the slab class
> selected for a new allocation is full. Optional expiration times for stored
> items are not taken into account by any kind of real-time garbage collector;
> they're just given precedence in the LRU calculation.
> In other words, by definition, memcached memory utilization will keep
> growing until it hits whatever limit you gave it. This isn't bad. It's
> exactly how it's supposed to be, which is why the graphs you have are
> practically useless. You care about the cache utilization rate, not bytes
> used.

This might be heresy to you, and other developers, but I don't really
care how the infrastructure works.  All I see is a load (combined with
a doubling/per week of activity downloading) going up at a high enough
rate that it makes me wonder if someone who knows about these things
should look into it.

As a last resort, I try to work on it myself until we can engage
someone willing and more capable to do the job.


More information about the Systems mailing list