tis 2007-03-20 klockan 20:13 +0545 skrev Tek Bahadur Limbu:
> I admit there is no such rule but I am using it as a base for
> measurement and comparison. Obviously, req/sec is an easier and a
> better unit than say open_files/sec.
What you want to monitor is the number of seeks/s, and if not that the
amount of time something is waiting for i/o.
the first isn't very easily collected in most os:es, but the latter is
usually available via sar, iostat etc.
the big difference between ufs and aufs (and also diskd) is that with
aufs Squid does not wait while there is disk i/o, continuing network
operations as the disk i/o takes place.
With ufs each millisecond spent in iowait means network activity was
paused..
> I have read articles and posts regarding ufs and they do suggest that
> ufs usually peaks out at 30-40 requests per second. However I have not
> really performed rigorous tests and benchmarks regarding ufs so I
> could be a little biased here.
The limit is caused by the seek time limitation of the disks, and this
has not improved significantly over the years.
However, trowing a lot of memory at it helps the situation, and the
average amount of memory per Squid server has increased significantly so
it is not unlikely the numbers may have improved a bit.
> These complied values are working for my servers up till now. Well I
> once had a problem regarding diskd without the above compilation
> options. But that was long ago and you may be right that I won't see
> any difference with or without it.
On FreeBSD and many other OS:es diskd requires OS tuning. The
requirements is documented in the FAQ. These is the minimum
requirements.
> > if you want to tune diskd read first a lot of postgres sql tuning matter
> > which are the only lonly guys which seem ever having worked serious
> > (except me of course ;) ) with this IPC stuff on FreeBSD. What you find on
> > squid's website regarding FreeBSD makes diskd work on old versions but not
> > tuned.
Would be great if someone with insight in FreeBSD shm & ipc
configuration would update the wiki with more current information.
Regards
Henrik
This archive was generated by hypermail pre-2.1.9 : Sat Mar 31 2007 - 13:00:02 MDT