malloc and caching misses

[Lasse Karstensen]

On Wed, Jul 29, 2015 at 10:37:11AM +0200, Tobias Honacker wrote:
> first of all we are using malloc 16G and varnish currently using > 25G of
> our memory so the vm begin to swap. n_lru_nuked value raise up quickly.

For reducing the resident memory size, try reducing fetch_chunksize to 8KB.

If that doesn't bring the memory consumption down, reduce the malloc
size.

Since you're doing a lot of return(pass), there will be a lot of
transient memory usage for buffering the objects as they are passed
through Varnish.

Note that there is a concurrency related panic in 4.0.2 when running
with a bit of load. You should upgrade to 4.0.3.

-- 
Lasse Karstensen
Varnish Software AS