[Asterisk-Users] RAID affecting X100P performance...

Michael Welter mike at introspect.com
Wed Jul 21 13:31:24 MST 2004


Mike Benoit wrote:
> I have a P3-800 with two IDE drives in a software RAID1 configuration.
> Each drive is on a separate IDE channel. Now anytime there is HD
> activity, I hear "beeps" and "cutting out" on a call using the X100P
> card. 
> 
> I ran the zttest program, and discovered HD activity would drop the
> accuracy down to between 2% and 50%. 
> 
> However I noticed if I disabled one drive in the RAID1 array, zttest
> would always report 99.98% or higher. So one drive running works fine,
> but as soon as I enable the second drive, all hell breaks loose. DMA and
> 32-bit mode are enabled on both drives as well.
> 
> I have a backup server with two Promise PCI IDE controllers in it, with
> 4 drives in a software RAID5 configuration, so just out of curiosity
> sake, I stuck a X100P card in it and tried running zttest while the RAID
> was re-syncing. The results were pretty bad. 
> 
> --- Results after 384 passes ---
> Best: 36.779785 -- Worst: 1.562500
> 
> Is this a poor mainboard issue, or is it actually not possible to do IDE
> software RAID on a machine running Asterisk with X100P cards?
> 
> Is anyone currently doing it?
> 
> Thanks.
> 
I tried software RAID1 with *.  It worked well until we had to recover a 
drive.  During drive recovery (recopy) the CPU is pegged, and it takes 
~16 hours to recover the drive.  I believe this is a Linux bug--there is 
no reason for such a high CPU usage during a disk copy.  Note that any 
abnormal Linux shutdown will cause a drive recovery.

I've since switched to 3Ware PCI RAID card.  The driver is in the Linux 
distribution which makes it very easy to configure.  The card is only 
about $100 US.

For a new client, I'm about to purchase a dual processor Athlon 
motherboard with an onboard RAID chip.  Stay tuned.

Cheers,

-- 
Michael Welter
Introspect Telephony Corp.
Denver, Colorado
+1 303 674 2575
mike at introspect.com
www.introspect.com




More information about the asterisk-users mailing list