slabs issue / idea

[Serhat Sakarya]

Hi guys,

I'm having some issues with using both large and small objects in the same
memcached instance: the large slab classes eat up all the memory before
enough small slab classes are allocated. Yes, there is pre-allocation with
the 1.2.0 and 1.1.13 branches, but having only 1MB to work with isn't so
great :-)

What are the common solutions for this problem? Does it make sense to run a
separate instance for the small slabs and a separate one for the large ones?

I don't know if this has been considered before, but is it a reasonable idea
to split the larger chunks into a linked list of 1-10KB chunks? This would
reduce the number of slab classes significantly, albeit at the expense of
performance.

Regards,

Serhat
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.danga.com/pipermail/memcached/attachments/20060916/f8cc8999/attachment.htm