Re: [squid-users] RockStore squid 3.2

 On Fri, 23 Sep 2011 11:58:09 +0200 (CEST), Fred B wrote:
> Hi,
>
> Actually i use squid with workers and diskd
>
> workers 2
> if ${process_number} = 1
> cache_dir diskd /cache1 130000 128 512
> else
> cache_dir diskd /cache2 130000 128 512
> endif
>
> But i lost cache performance because the caches are not shared
> How to use (test) the new feature RockStore with
> squid-3.2.0.12-20110923-r11343 (compilation, squid.conf) ?
>
> Thanks

 rock is just another cache_dir type. With this one you need to set
 max-size to something reasonable in the KB range.
 Make sure you run squid -z to create the directory properly first.

 There are a couple of quirks which you may need to work around:
  * a full restart is needed to change dirs, reconfigure wont work
 (yet).
  * there seems to be a bug in -z where dirs hidden inside if statements
 are not always built. So put them outside the "if" for the -z setup step
 to be safe. After that it seems fine for the run operations.

 From the docs which will get published next release:
 "
         The rock store type:

             cache_dir rock Directory-Name Mbytes <max-size=bytes>

         The Rock Store type is a database-style storage. All cached
         entries are stored in a "database" file, using fixed-size
 slots,
         one entry per slot. The database size is specified in MB. The
         slot size is specified in bytes using the max-size option. See
         below for more info on the max-size option.

         swap-timeout=msec: Squid will not start writing a miss to or
         reading a hit from disk if it estimates that the swap operation
         will take more than the specified number of milliseconds. By
         default and when set to zero, disables the disk I/O time limit
         enforcement. Ignored when using blocking I/O module because
         blocking synchronous I/O does not allow Squid to estimate the
         expected swap wait time.
 "

 On the whole rock fills the niche where COSS was used previously. So
 objects over something like 128KB would be best kept in diskd or AUFS
 dirs. If you like you could experiment with a set of rock dirs using
 min-size/max-size to split the files into bands for efficient space
 usage. We are interested in things like that to figure out what working
 practices should be advised.

 Amos
Received on Sat Sep 24 2011 - 07:05:15 MDT