On Wed, Mar 28, 2001, David Rumble wrote:
> Hello
>
> As Squid is an I/O intensive app, I have been working with a major user of
> Squid running on Linux RH 2.2.16.(bespoke). Hardware no name 512MB, single
> p2 300. Two test machines have been setup, One with a platypus Qikdrive
> www.platypus.net (SSD 2GB capacity, upto 10MBps/15k Tps/Access times - 3/10
> MICROSECONDS. The other with 2GB Scsi HDD.
>
> After considerable testing there is no performance gain as far as HIT access
> times are concerned. According to the drive spec/requestsPs equation one
> would expect a significant improvement. (requests/sec1000000/access time).
> Cpu and Bandwidth are within tolerances.
>
> Does Squid use DISKD to cache to RAM? Are there any configs that need
> changing? Any advice,comments would be greatly appreciated.
The equation for a "HIT" is a function of network behaviour and
disk access, rather than just disk access.
You might want to try reiserfs on it.
Alternatively, watch this space for the squid FS being written, and
I _do_ believe it will run nicely on a SSD. :)
On a side note: I know of at least one large company running a squid-like
application through both RAM, SSDs and normal disks. Their throughput
is actually rather damned impressive, since they shift the non-popular
stuff off RAM -> SSD -> disk .
And no, diskd doesn't cache any disk IO. thats up to the OS, and a little
bit up to squid.
Adrian
Received on Wed Mar 28 2001 - 05:05:14 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:59:00 MST