I'm talking about a rackmount computer (Tyan GT24) with 4 drive cages
for SATA drives which plug into a backplane. If I have the drives laid
out atop the computer with the SATA cables running over the open case
into the RAID card, the drives are very fast (70-80 MB/second). Now if I
merely install the drives into the drive cages in the rackmount while
keeping them directly connected to the SATA cables (i.e., backplane
removed), the drives are way slower (20-30 MB/second). This is very
reproducible behavior and it occurs with not only our 750 GB Seagates
but also with an Hitachi and WD.

[ Post a follow-up to this message ]

Maurice Volaski wrote:
> I'm talking about a rackmount computer (Tyan GT24) with 4 drive cages
> for SATA drives which plug into a backplane. If I have the drives laid
> out atop the computer with the SATA cables running over the open case
> into the RAID card, the drives are very fast (70-80 MB/second). Now if I
> merely install the drives into the drive cages in the rackmount while
> keeping them directly connected to the SATA cables (i.e., backplane
> removed), the drives are way slower (20-30 MB/second). This is very
> reproducible behavior and it occurs with not only our 750 GB Seagates
> but also with an Hitachi and WD.

One obvious way is when vibration disturbs the disk head enough to cause
the drive to have to re-read data (in your case, since you're clearly
streaming data it might have to retry a couple of times per revolution
before it got everything right).  This is reportedly one area in which
'enterprise' drives work better (i.e., are more resistant to such
disturbance) than desktop drives, and people have specifically reported
difficulties in multi-drive enclosures where mechanical coupling allowed
one drive's vibration to affect other drives.

No guarantee that this is what you're seeing, but it is one possibility.
I suppose you could also be getting some kind of electrical
interference in your cabling in the box that does not occur when the
cables are spread around outside it, but it's more difficult to see how
that could generate that level of performance loss that you're experiencing.

I'm not sure whether 'S.M.A.R.T.' information in the drives includes
data on the kinds of retries that vibration-related problems would
produce, but it wouldn't hurt to take a look at before-and-after stats
in both configurations.

- bill

[ Post a follow-up to this message ]

In article < OKadnSgQMPZDoGvYnZ2dnUVZ_hisnZ2d@metroca
stcablevision.com>,
Bill Todd <billtodd@metrocast.net> wrote:

> Maurice Volaski wrote: 
>
> One obvious way is when vibration disturbs the disk head enough to cause
> the drive to have to re-read data (in your case, since you're clearly
> streaming data it might have to retry a couple of times per revolution
> before it got everything right).  This is reportedly one area in which
> 'enterprise' drives work better (i.e., are more resistant to such
> disturbance) than desktop drives, and people have specifically reported
> difficulties in multi-drive enclosures where mechanical coupling allowed
> one drive's vibration to affect other drives.
>
> No guarantee that this is what you're seeing, but it is one possibility.

You are right. It's the fans.

[ Post a follow-up to this message ]