> Alex Rousskov wrote:
>> 1. optimizing Rock code so that it can rebuild faster
>> 2. optimizing your OS so that it can read disk faster
>> 3. optimizing your disks so that they can read faster
On 02/13/2014 10:13 AM, k simon wrote:
> I wonder to
> know does the rock store's 32KB slot means that I can use 32KB block size ?
IIRC, Squid code does not know about file system block sizes so Squid
should work with any block size. I do not know whether using 32KB fs
block sizes will improve your overall disk cache performance, but it is
a good idea to test that (as a part of item #2 above).
It may be tricky to configure this correctly though, for several
reasons, including:
* If your rock slot size is not exactly the same as the disk block size,
you may force the disk to do a lot more I/O than necessary.
* If the majority of cached entries are smaller than 16KB but your disk
block size is 32KB, you may be forcing the disk to do a lot more I/O
than necessary, especially if Squid does not write the whole slot at all
times. I do not recall whether it writes the whole slot -- the behavior
may depend on the Squid version. Ideally, Squid should round up the
write size to be on the next page and fs block size boundary, but the
code is not that smart yet.
In summary, some experimentation and testing may be needed to get this
right. Some "optimizations" like that may actually make things worse so
be careful...
Good luck,
Alex.
Received on Thu Feb 13 2014 - 17:43:45 MST
This archive was generated by hypermail 2.2.0 : Thu Feb 13 2014 - 12:00:05 MST