[Asterisk-Users] Hardware for Asterisk

Steven Critchfield critch at basesys.com
Fri Jan 16 10:27:19 MST 2004


On Fri, 2004-01-16 at 06:47, Andrew Kohlsmith wrote:
> > If you value your data, don't use software raid. If you value
> > performance don't use software raid. If you value uptime/stability don't
> > use any raid on IDE.
> 
> That's pure bullshit -- I use software RAID *specifically* because I value 
> my data.  I don't want to buy two hardaware RAID controllers to have one 
> sit on the shelf just in case the first dies... and if the second dies 
> you're SOL because they've lasted long enough that they're no longer 
> available.  Linux software RAID is available on any Linux system and if the 
> system blows up I can put the drives in another system and *not* worry 
> about it not being detected.
> 
> As far as performance goes, I have some bonnie++ tests that I've run that 
> show that at least on the few systems I've tested, software RAID 1 beat out  
> hardware RAID 1 (these systems were IDE, SCSI-2 and Ultra320, with DPT RAID 
> controllers for SCSI on P4 and I think regular Promise IDE RAID controllers 
> on P3) -- not a huge difference in speed but one that at least tosses your 
> "if you value performance don't use software raid" argument.
> 
> Perhaps on a _heavily_ loaded server you might be right, but then again I 
> feel that you're stupid for letting a server get so loaded up that it can't 
> handle the simple mirroring algorithms in addition to normal file servering 
> functions without degrading performance to a noticable degree.
> 
> I used to believe that HW RAID was the only way to go.  With RAID5 I still 
> feel that is true to an extent.  However if you're just mirroring there is 
> _no_ significant advantage to choosing hardware RAID over software RAID.  
> Not on IDE, and not on SCSI.  In fact, there are advantages to choosing 
> software RAID over hardware RAID, as I've mentioned above.

Have you experienced a hardware failure yet that you had to come back
from? If you loose a drive, it is a high probability that you will loose
the controller. So unless you have a add on card, or some motherboard
with 4 IDE ports, you will corrupt the second drive of a mirror. If the
second drive is corrupted, then you are only a hair above not having
anything. If you don't trust that, check out the GOOD IDE raid
controllers. You are only allowed to place 1 drive per port, and they
only use 1 port on a IDE controller. 

Even the large NAS devices that use IDE have the IDE controller built
into the sled that holds the drive and use PCI hotswap technology. 

I don't buy it that any truly redundant raid system is as fast in
software as in hardware on a machine doing anything significant. In raid
1, you are double or more writing all data to the drives. in a read
environment, it might be able to share the load out to more than 1 drive
and help, but I don't expect it would be much better than a dedicated
controller handling the load. Any load of a software raid solution takes
processor time away from the processes it is trying to complete. So take
our VoIP application, if I am spending time getting the voice recording
to 2 or more drives and the software to get it there, you have
significantly reduced the amount of time available to the CPU to handle
the VoIP packets in a timely manner. This only gets worse as call volume
goes up. If it is hardware raid, you know it will be a single write and
the controller deals with the problems. 


> > What matters as far as the computers being used is that you are unlikely
> > to get your hands on a real server class motherboard without having
> > bought it in a Dell or Compaq. It also matters as to the supporting
> 
> Again I call bullshit -- Where do you think Dell and Compaq get their 
> motherboards from?  (ok compaq might actually manufacture them) -- I can 
> get server-class motherboards from Asus, Gigabyte, Intel, Tyan, and a host 
> of manufacturers without having to buy into the proprietary nature of 
> anything Name Brand.

On server hardware, Dell has their own boards. IBM had their own boards.
Compaq and HP also produce their own boards. Maybe they don't produce
their own boards in the desktop models, but they do in the server class
machines. While you can buy Intel, Tyan, and SuperMicro boards, I
wouldn't consider any of the remaining ones you list as truly server
class.  

> > hardware. If the PSU isn't quality enough, then it doesn't matter what
> > motherboard you use. Dell doesn't want to deal with your system after
> > sales. They will put a few extra dimes into the PSU so it stays in shape
> > for a few more years. The companies you are most likely to purchase a
> > case from will usually expect you to not come after them if the PSU
> > fails. So why would they bother to spend the extra money to make the PSU
> > last longer.
> 
> I can also put some extra dimes into the power supply... or fans... or 
> anything.  Dell/Compaq/whoever does not mean high quality by default.

Maybe not by default, but if you get into the hot swap PSUs you
absolutely are talking quality. 

> > Also Dell is more likely to have a part to fix your machine in the mail
> > within hours instead of you waiting till you can get to the store to
> > purchase your replacement part before RMAing the part and waiting the
> > couple of weeks for the replacement.
> 
> This is true.
> 
> > In general, you get what you pay for, and less so when you go bargain
> > hunting. It all comes down to the same old problem of figuring out what
> > your time and downtime are worth.
> 
> Agreed.  Personally I'd rather have a complete second system on the shelf 
> that I can swap out within 15 minutes than rely on anyone plus a courier, 
> but that's just me.

While I'll agree that a complete spare is a good idea, if you are
looking for the bargains now, I don't have faith that you would also be
the person who would buy 2 and leave the second untouched until failure
occurs. I'll admit I couldn't leave a fully functioning machine just
laying around not doing something.
-- 
Steven Critchfield  <critch at basesys.com>




More information about the asterisk-users mailing list