Another slow disk thing

[Mercutio]

Hey while we're at it...

I recently became responsible for an Intel OEM Server. I don't know what model it is, but it's about six years old and uses an integrated LSI SATA controller instead of ICH* or IDE-emulated SATA.

The machine has a hardware RAID1, but they way I came to be involved with it is that neither member of the array was mountable one day and the guy who owns the server just called the only computer company he had any current relationship with.

I wound up recreating the array with a new set of disks and using a TrueImage boot CD to clone the data from the other drives on to my new ones since the guy didn't actually know how his backups worked or where they were located. Eventually I found out his backups were an external 160GB drive that must've been full for the last three years, based on the date of the last successful backup.

Old drives were 80GB Hitachi SATA drives. Replacement drives are Hitachi 250GB SATA drives. The new drives are SATA II and the original drives are original SATA.

OK.

Here's the problem: Data transfers on the new drives seem to top out at about 2.5MB/sec, slow enough that the machine takes 45 minutes to boot, a 100MB Quickbooks file takes about 1 minute to load up from a client, and a database report that used to run in "about a minute" now takes about a half hour.

I don't know how things used to be with this guy's stuff, 'cause I wasn't around for that, but he says it's unbearably slow now, and the only thing this system does is authenticate people on a domain and serve files. Everything DOES work, just... slowly.

As a further source of annoyance, his disk-related performance counters all look normal, and his controller isn't fancy enough to have a monitor of its own. The RAID BIOS shows that the current drives are synchronized and healthy.

I'm telling this guy that the real fix would be a new server, but he doesn't want to pay for licenses again, and at this point I'm actually kind of intrigued to figure out where the issue is.

If I disable RAID mode on the controller, I get a BSOD on boot. If I disable the RAID driver in Windows, I get a BSOD on boot. If I do both, I get a BSOD on boot. Safe mode doesn't work either.

I even cloned his data back to another set of the same old 80GB Hitachi drives and saw the same slowness.

I'm really quite puzzled by this.

 

[Will Rickards]

Can you force the new drives into SATA I mode? By jumper or hitachi's maintenance program?

 

[Mercutio]

Apparently not on the 7K1000, but I did try an second set of 80GB 7k250 that are SATA I anyway.

 

[Will Rickards]

My only idea is to bypass the onboard ports and use an add-in card to give you SATAII support. If they still want the raid you could do it. But it seems like you could get rid of that and just have a reliable backup system.

 

[sdbardwick]

Thermal throttling due to clogged/dead/unseated HSF?

 

[Mercutio]

Affecting disk access and nothing else? You think? CPU utilization is in the range of 1 - 2% every time I look and commit charge is fixed at only about 450MB, which is pretty much exactly what I'd expect on an essentially idle Windows file server.

I haven't pulled this thing out of its rack yet and in fact it's only been on my plate for the last three days. There's no documentation of any kind (or, I note, indication that anything is licensed) on site.

 

[sdbardwick]

Upon further reflection, thermal throttling does seem unlikely to be the cause. The magnitude of degredation seems out of character, as TM1 only provides 50% duty cycle reduction (I'm assuming Netburst processor). More likely that the SATA controller has issues (like aging capacitors mucking up the signal between the controller and drives, causing huge number of CRC errors and retries).

That said, throttled processors can have an unexpectedly large impact on boxes running software RAID; more than once I have been successful in restoring performance on those setups just with a duster (especially when you can avoid taking the server down until after the normal workday).

 

[Bozo]

It's been awhile since I fooled with an IBM server, but I seem to remember each one had an install disk. You answered a few questions (what OS, what RAID etc) and the setup disk configured the system. When it was finished it prompted you to load the OS.
Could be that you need this disk to set everything up again.

 

[Tea]

If this was my problem to sort, I'd start by switching off all the RAID stuff, taking a spare drive (any drive I had handy, don't care which SATA version), and doing a test install of any handy random Windows OS - XP Home Edition SP3, for example.

This would tell me that either:

1: the hardware is fine: it runs at roughly the speed I'd expect from a Brand X CPU with Y GB of RAM.

2: there is a serious hardware problem here: it runs too slowly for what it is.

Once we know which of thiopse two apply, we are in a position to start looking for good answers. BTW, my bet is that it's software and a reinstall will be the answer ... but where are his licences?

 

[Mercutio]

I wound up installing a two-port PCI Adaptec SATA controller and doing a repair install using that as my disk controller. Had to use a USB floppy to ideal with the install time driver. The setup as it is now might not be ideal, but it's better than whatever the hell is going on with the existing controller, and I didn't have to spend all weekend troubleshooting it.

I spent the whole time I was there thinking how awesome Linux is about dealing with hardware changes.