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

Re[2]: NFS problems (also contains RAID report!) - LONG!



udif@bigfoot.com (Udi Finkelstein) wrote:

>  >Forget about hdparm, it's a toy (I mean, hdparm -t). Use bonnie.
>  What is bonnie?

A reliable filesystem I/O benchmarker. http://www.textuality.com/bonnie/

>  >>  Now comes the RAID part. I made a linear raid
>  >
>  >Why not RAID0? Generally, you'd get much better perfromance and could
>  tune
>  >it to your specific needs. I've once done more or less extensive
>  benchmarks;
>  Why? because linear, unlike RAID0, can be expanded more easily if I want
>  to
>  (if I add a 4th disk). 

"More easily" means without backup & restore? But I believe you realized that
with the linear (or level 0) raid the probability of the crash of the total
array storage is quite high (proportional to the number of disks) so you want
to run backups frequently anyway, right?

>  Besides, for if the 3 disks used for RAID are also used
>  for other things (swap & system) I will not be surprised to see linear
>  gets
>  better results if swap is on disk1, system on disk2, and the linear array
>  starts on disk3. 

The best for swap would be to split it over all the disks and let kernel
decide which to use in a given moment (it uses basically a raid0 algorithm,
AFAIK); and for the system better is to be on a separate disk...

>  Also, with 16MB/s coming out of the
>  disks even in linear mode, it outperforms the 100MBit/s ethernet it's
>  shared through, so it really doesn't matter.

Well, if it's the sole use of this box... (but why you're worrying about swap
then?). And still, the sustained output is not the only important parameter.
For DB applications the random seeks may be crucual, and with linear you'll
never approach even close to raid0.

Regards,

Evgeny


--
   ____________________________________________________________
  / Evgeny Stambulchik  <fnevgeny@plasma-gate.weizmann.ac.il>  \
 /  Plasma Laboratory, Weizmann Institute of Science, Israel \  \
 |  Phone : (972)8-934-3610  == | == FAX   : (972)8-934-3491 |  |
 |  URL   :    http://plasma-gate.weizmann.ac.il/~fnevgeny/  |  |
 |  Finger for PGP key >=====================================+  | 
 |______________________________________________________________| 

=================================================================
To unsubscribe, send mail to linux-il-request@linux.org.il with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail linux-il-request@linux.org.il