Re: [squid-users] Urgent Help Needed: Any suggestion on CPU utiliztion increased from 73% to 97.5 after cache_mem increased from 4 GB to 8 GB

From: Henrik Nordstrom <henrik_at_henriknordstrom.net>
Date: Sun, 15 Jun 2008 05:05:32 +0200

On sön, 2008-06-15 at 10:57 +0800, Arkin Y wrote:

> Acctually , I tried to use Linux OS default malloc, gnu malloc , dlmalloc,
> dlmalloc could help me get the best performance among the three different
> malloc .

Interesting.

For me dlmalloc fails completely when the process size goes above 2GB.

> I checked all my mgr:info log , when I set the cache_mem to different size ,
> the stroe entries are different too .
> When the cache memory is 4GB , the Store Entries are ~519396
> When the cache memory is 6 GB , the StoreEntries are ~776836
> When the cache memory is 8 GB , the StoreEntries are ~1026097

Are you running without an on-disk store? (cache_dir)

> The average size of the stored objects almost ~8KB, so I am not sure
> whether such problem was caused by large objects .

Ok.

> I checked a hash_lookup method , If in the worst cases in 8G scenario,it
> looks like almost 2 times of hash keys will be traversed to get the
> storeEntry than in 4 G scenario.

The size of the hash table is dependent on the store size.

But I guess this calculation screws up if there is no on-disk cache of
reasonable size.

On startup, what is reported NN for "Using NN Store buckets" in
cache.log?

Regards
Henrik

Received on Sun Jun 15 2008 - 03:05:38 MDT

This archive was generated by hypermail 2.2.0 : Mon Jun 16 2008 - 12:00:03 MDT