At the risk of flogging a dead horse, I'd like to add my 2p to this
discussion. No matter what implementation of RAID you are using, it does 1
of 3 things:
1. Provides for recovery in the event of a hardware failure by storing data
redundantly; either (a) mirroring it completely (RAID 1), or (b) storing
composite information in a parity area (RAID 4 or 5).
2. Speeds up I/O operations on filesystems by spreading them across multiple
physical drives ("Striping" or RAID 0).
3. Both of the above (RAID 0+1).
Problems usually arise when you have more than one system trying to do the
same thing, because they end up conflicting or competing with each other.
This most often occurs in the area of striping because lots of software can
also be configured for load-spreading to improve I/O performance.
My advice to anyone (not just squid bods) would be to decide which product
you want to do the task: if squid, set up multiple cache_dir's but don't
stripe; if RAID, set up striping with a single cache_dir on a striped
volume.
If you are also worried about recovery then your decision may also be
directed by the need to set up redundant datasets. The Linux native software
RAID, for example, does not support RAID 0+1, so you would be obliged to use
RAID for the redundancy and squid for the load-balancing. If performance is
an issue, RAID 1 is MUCH faster than RAID 5, often outperforming a single
disk.
In general, configuring a specific product such as squid often has
performance benefits over something like RAID; but these may be minimal, and
striping is a solution which can provide benefit for any software on your
system.
Winston Gutkowski
Received on Tue Jun 04 2002 - 14:49:30 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:08:26 MST