[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: RAID Suggestion for webserver



> >
> > The CPU won't be handling this... the 3ware RAID card (hardware) will
> > perform the parity calculations, so RAID 5 won't cause that type of
> > slowdown due to additional CPU utilization.
>
> Mmm, this is one of the rare IDE RAID cards that are true hardware RAID.
> But you still have only 3 disks doing the work.
> How about the number of disk-IO operations needed to perform a read or
> write operation??

Yes, this is true hardware RAID, and not those "soft" hardware raid cards.

Disk IO is not very high... just low to medium.


> In degraded mode there are no computations to be made. It's just the
> same as RAID0. The card just reads whatever is on the two disks and only
> sorts the blocks in the right order. The same is with the RAID10 setup.
> Although this setup still has 3 disks, one of those 3 only has half of a
> stripe and can't be used in the array, so in this case the degraded
> RAID10 is also something like a RAID0.

Oh... I thought running RAID 5 in degraded mode impacts badly on
performance because the actual data on the failed disk has to be simulated
from the parity and data on the other disks...?

> The hardware's chips will be used for rebuilding the array on the spare
> disk. How much this will impact the normal operation depends on how fast
> you want the rebuild to take place. And even with dedicated chips this
> will degrade normal operation because it needs to read all information
> on the 2 disks in order to calculate what to put on the 3rd disk.

True... there will always be a performance decrease during the rebuild
process. I was just saying that using hardware RAID should help to
minimize the slowdown compared to using software RAID or those fake
hardware raid cards.

> > It seems RAID5 would be a safer solution as long as another failure
does
> > not occur during the reconstruction onto the spare. Hum... how long
would
> > the reconstruction take for a 40G hd? I'm guessing 30-40 minutes?
Would
> > that be about right?
>
> I'm not sure how fast today's disks really are, but remember that the
> rebuild time also depends on how much normal work the array has to do (I
> asume the card has a setting that allows you to tune how the card
> divides the rebuilding work versus the normal work).

We bought those nice new IBM 120GXP hard disks... they have the highest
performance when conducting multiple read/write operations (high IO)
simultaneously due to IBM's excellent adaptive cache algorithms. Maxtor is
good if the IO load is low (eg. non server environment) due to it's fast
access times, but due to bad cache algorithms (or perhaps poor HD chip
performance) it gets overloaded in a high IO environment (file server,
database server, etc.)



Reply to: