Re: hit rates and maximum_object_size

Neale Banks wrote:

> That's the "Storage LRU Expiration Age"? ....

Yes.

> Cache information for squid:
> Request Hit Ratios: 5min: 27.1%, 60min: 30.7%
> Byte Hit Ratios: 5min: -3.5%, 60min: -1.4%
> Storage Swap size: 7416851 KB
> Storage Mem size: 102368 KB
> Storage LRU Expiration Age: 7.96 days
> Mean Object Size: 11.25 KB
> Requests given to unlinkd: 0

If this has been running for some days, then you do not have need for
much more disk.

> Mmmm... given that I don't have any cache-digest happening, those
> negative "Byte Hit Ratios" (as discussed in separate thread just now)
> are a bit of concern. Maybe something like the quick-abort needs
> tweaking? It's currently:
>
> quick_abort_min 16 KB
> quick_abort_max 16 KB
> quick_abort_pct 95

delayed abort can be fooled on some requests, and my recommendation is
to configure quick_abort as restrictive as possible until the function
is properly verified.
  quick_abort_min 0 KB
  quick_abort_max 0 KB

Also, there are problems with HEAD requests on FTP objects, and some
range requests. Patches for these two latter problems are available from
my Squid page <http://hem.passagen.se/hno/squid/>

> FWIW, I am seeing in the cache.log:
>
> aio_queue_request: WARNING - Running out of I/O theads
> aio_queue_request: Perhaps you should increase NUMTHREADS
> aio_queue_request: Or install more disks to share the load
>
> But that _is_ a separate (performance) issue, right?

Right.

I also have a couple of patches related to this message...

/Henrik
Received on Mon Dec 13 1999 - 14:05:59 MST