On Donnerstag, Jänner 30, 2003, at 10:21 Uhr, Henrik Nordstrom wrote:
> Raid2-5 is not good for Squid cache drives due to their high write
> penalty for small random writes.
I see, thank you.
>
> For the cache use either individual drives, or two RAID1 sets.
>
I shall create two logical drives then, spanning 2 disks each with Raid
1
> What was the reasoning to select RAID3?
>
Basically these characteristic advantages of RAID level 3
Very high Read data transfer rate
Very high Write data transfer rate
Disk failure has an insignificant impact on throughput
Low ratio of ECC (Parity) disks to data disks means high efficiency
I am aware, that RAID 3 has its limitation namely:
Transaction rate equal to that of a single disk drive at best (if
spindles are synchronized)
but I disregarded them, since we are on IDE only anyways and this is
not a multiple thousand user cache.
Controller design is fairly complex but the one we are using is a
pretty good one and doing just fine.
>
> reiserfs is a good filessytem candidate.
>
I am testing 7 cache dirs with reiserfs now and a larger one with jfs,
where I am storing teh small objects in the reiserfs partitions and the
larger ones on the jfs.
> There is no benefit in performance to split the cache in multiple
> partitions. What matters for cache performance is how many of your
> drives you can keep busy doing different operations..
Well in this case the RAID controller does play a significant role.
Since the drives are logical and purely handled by the controller it
also decides about teh read/write and queue operations trying to keep
all 4 drives equally busy and the spindles in sync.
> but for
> maintenance and recovery purposes it is a good idea to not make the
> cache partition too large, perhaps not more than twice the amount of
> cache you plan on using.
Thank you for all the tips in advanced, I will test and report back.
>
Received on Thu Jan 30 2003 - 02:35:33 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:12:59 MST