On Dec 2, 11:51*pm, wrote:
On Nov 29, 12:01*pm, Torbjorn Lindgren wrote:



wrote:
So you dont thinkthe7200 rpm (single port)? sas drives (ES2.1) from
seagate.. ie: 500gb, would be worththebother over 32MB enterprise
drive equivalents that are sata?

Based onthespec it'sthesame drive mechanism except less cache on
theSAS model,thefirmare is somewhat different though. It depends on
whether you need to mix in SAS 15k disks inthesame chassi/controller
ornot, ifnotI'd savethemoney and go forthelower cost SATA ES.2
drives.

There CAN be differences in other areas, in this case it appears to be
thattheSAS controller is faster (up to 50% faster CPU, slightly
faster memory). It can also handle more cache (4GB instead of 2GB),
though in most cases 2GB is probably enough.

TheCPU/memory difference should give it higher streaming read/write,
if given enough disks (as a guess it'll may start showing up at
perhaps 10-12 disk for read and RAID1/10/5 write, 8-10 disks for RAID6
write?).

However, for most people STR has little or NO effect, it's high enough
that abouttheonly possible exception I can think is HD video editing
using *uncompressed* video and I'mnotsure even those guys would
typically notice it (probably run into CPU utilization limits even on
a dual/quad before a STR limit that high is noticeable).

I dont think we take a ton offileI/O hits now.. but ourfilesystem
is stored onthehost server side.. withthenew server it will be
from within a VM.

I'm leaning towards building a nice array of SAS drives onthenew
server just for a VM or two that will do nothing butfilesharing
(network shares)..

Or maybe use that SAS array to power say 2-3 VMS,therest will be on
a sata array..

That's where it starts to get complicated and where there's few good
guides. It MAY work well to use both SATA and SAS disks on a
controller, and it maynot! Ifnotyou do wanttheSAS ES.2 models for
bulk storage instead (or two different controllers but that
complicates other things).

There's a few anecdotes butnotmuch beyond that.

Withfilesharing right now, onthemain host, we have about 60 open
files.. but for most other sql apps that fall under our one sql vm,
i'm guessing we dont have more than a handful or maybe 15 atthemost
simultaneous hits as of now, though this may grow.

Number of files isn't a particularly usefull metric for someone
outside which doesn't have knownledge of yoursystem.

If you want to size this try to figure how much read IOPS and write
IOPS you have, then try to figure out how many of those can be
gathered via a large write-back cache (this will have a big impact if
you use RAID5 or RAID6 and 10+% writes).

A fresh thought occurred.. i wonder if doing a mirror of two 15K SAS
drives would prove beneficial?

I looked at 450GB seagates that were 15K drives.. i'd need 5 to come
close to 2TB in a raid5 array.. thats $3000 alone.

You mentioned looking at Seaburg for the cpu.. how is it different
than harpertown.. do you have a particular model in mind.

The model of harpertown i'm going to slightly upgrade to is the 2.66
1333 fsb 5430 series. *This is up from 2.5 and 1066 fsb on our other
server.

I saw some passmark disk mark ratings on a sas 15k 4 drive array.. the
values were around 15,000 or so if i remember correctly, which blows
away our raid6 8 drive 32mb 500gb array that is sataII..

It still is a bit steep for a 2TB 5 drive SAS array ($3000), vs the
$799 6 or 7TB raid6 or raid5 array.. but i'm betting the Vm'S that ran
on those arrays would probably have impressive disk marks in
themselves... but for us, for now we settled for Sata II's, maybe
going with SAS in a year, when sas 2.0 is out (price drop), unless we
find our file systems suffer in the VM's they are moved to (network
shares).