Re: [squid-users] Hard Drive Latency

From: James Gray <james_gray@dont-contact.us>
Date: Mon, 30 Aug 2004 17:43:55 +1000

Joe Cooper wrote:
> James Gray wrote:
>
>> Is there anything to be gained by switching to AUFS in a SCSI RAID 0+1
>> (real hardware RAID with 64MB cache etc) environment? How long is a
>> piece of string....I know - just curious from a "theoretical" point of
>> view.
>>
>> We often see system (i.e., not "nice" or "user") utilisation go up
>> under heavy Squid load on FreeBSD, but I can't say whether or not this
>> results in a noticeable performance degradation - we've never measured
>> it and the users haven't complained (about proxy performance anyway).
>
>
> FreeBSD is not a supported environment for AUFS, as far as I know. You
> would use diskd instead, but it doesn't do anything on a single device
> system, as I assume you have. AUFS on a single disk system does make a
> measurable difference (regardless of controller and RAID).
>
> RAID of any sort is usually not recommended for Squid if performance
> matters to you at all.

Thanks for the info. Actually we have a 4 drive setup; 4x18Gb 10k RPM
U160 in RAID 0+1 on a Compaq SmartArray5i controller with 64MB cache,
Xeon 1.2GHz (IIRC) with 512MB ECC RAM :). Performance isn't such and
issue for us, but redundancy is. Sure we could've gone RAID5 with one
hot spare and probably got lower latency but we decided throughput was
more important. Our bench-marking showed we got fractionally better
throughput with RAID 0+1 over RAID 5 - the trade-off was slightly higher
latency. Besides the whole thing is over-engineered by several orders
of magnitude; it only supports 15 concurrent users on a 4Mbps link!!

Like I said - theoretical curiosity more than anything. The subject
about latency spiked my interest due to the performance figures we'd
seen previously in our testing.

Cheers,

James
Received on Mon Aug 30 2004 - 02:51:00 MDT

This archive was generated by hypermail pre-2.1.9 : Wed Sep 01 2004 - 12:00:03 MDT