Re: Size of cache-digests ...

From: Alex Rousskov <rousskov@dont-contact.us>
Date: Thu, 8 Oct 1998 16:35:28 -0600 (MDT)

On Fri, 9 Oct 1998, Henny Bekker wrote:

> I'm curious to the size of the cache-digest and if it is locked into RAM
> memory..

Both peer and local digests are stored in memory (and on disk). Storing peer
digests in memory is unavoidable because we need to do fast lookups. We could
get rid of in-memory copy of a local digest. However,
        - same memory will be re-allocated every hour to rebuild
          the digest,
        - memory requirement is probably relatively small,
        - old digest will have to be swapped in if we want to
          support digest "diff"s or "delta"s to further decrease
          network overhead.
Keeping that in mind, we decided to have in-memory copy of a local digest.

> On our toplevel cache with 512-MByte RAM and a cache-disk of
> 16-MGByte whe have had cache-digest enabled.. We also have a number of
> Squid v2.0 clients and a number of Squid v2.0 sibling caches which have
> also their cache-digest enabled.. I wonder how much RAM it will occupy
> and if this might be the reason the system was swapping...

Consider using cache manager interface. Check out "Cache Digest and ICP
blob" and "Store Digest" report the size of each digest...

What bit-utilization do your digests have?

> PS What does the "Enable Forw/Via database" option does ??

I think it was some experimental code to plot/analyze the Global Cache Mesh
based on X-Forwarded-For and Via headers. Remember the "Plankton" tool
demonstrated on the last Caching Workshop? I think they used that..

> Is it being used in coherence with cache-digest ??

No, there is no connection between Cache Digests and "Forw/Via database".
You probably should not enable the latter unless you are collecting those
hierarchical stats...

Alex.
Received on Thu Oct 08 1998 - 15:37:23 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:42:23 MST