Very slow data transfers beetwen drives

[egd]

Hi. PC in question is running Windows Server 2003 (no service packs installed) and has a mixture of ATA and SATA drives installed. The motherboard is an ABIT IC7-G which provides 2 ATA channels (up to four devices), 2 SATA channels in the ICH5 chipset and an additional 2 SATA channels on a Silicon Image on-board raid chip/controller. Additionally a FastTrack S150 TX2 raid controller is installed in a PCI slot. The machine currently boots from a striped ATA array (raid0) connected to the Promise controller.

I recently purchased 2x Hitachi Deskstar 7K400 (400GB) drives to be implemented as a mirrored array, either via the Silicon Image on-board controller or via the Promise controller.

Whether cabled to the SI or Promise controller the drives are properly detected, partitioned and formatted NTFS without any hitches.

Problem:
If I then try to copy data any of the other drives in the system to the RAID mirror, the copy operation starts off flying and within a second or two grinds almost to a halt (almost like the OS has to stop and think about where/how it is going to write the data), and then resumes copying very slowly - about the same speed as you get reading from a CD to write to a HDD.

As mentioned above, I've tried this through both the Promise and SI controllers and have the same result.

I've run Hitachi's quick test diagnostics onboth drives and they checkout fine.

Has anyone else come across this problem or have any ideas as to where the problem may lie?

 

[Mercutio]

Are you setting up your arrays with the cards or using the Mirror Volume capacity in 2003 Server?

Whichever you are doing, have you tried the other?

 

[egd]

Are you setting up your arrays with the cards or using the Mirror Volume capacity in 2003 Server?

Whichever you are doing, have you tried the other?

I've tried setting up the arrays with the Promise card and also with the on-board controller.

I am also going to set both drives up independently using the SATA channel of the Silicon Image on-board chip and then try to copy to each just to see if there is a problem with either drive.

Assuming both pass I'll try to implement the W2K3 mirror option - ultimately these two drives are intended to be a RAID mirror for my purposes. I was thinking of going RAID5 using another Promise controller, but I'm a little wary of buying another 3 drives only to find I have the same problem with the RAID5 array.

 

[egd]

Are you setting up your arrays with the cards or using the Mirror Volume capacity in 2003 Server?

Whichever you are doing, have you tried the other?

I've tried setting up the arrays with the Promise card and also with the on-board controller.

I am also going to set both drives up independently using the SATA channel of the Silicon Image on-board chip and then try to copy to each just to see if there is a problem with either drive.

Assuming both pass I'll try to implement the W2K3 mirror option - ultimately these two drives are intended to be a RAID mirror for my purposes. I was thinking of going RAID5 using another Promise controller, but I'm a little wary of buying another 3 drives only to find I have the same problem with the RAID5 array.

 

[Buck]

I would try an installation of 2003 on a single ATA drive (Primary Master) as the only device on the cable. Then I would experiment with a RAID1 on each of the three Serial ATA controllers.

 

[Pradeep]

Have you got the advanced options->UPS backup present->write cache enabled in drive properties for the onboard drives?

 

[egd]

Have you got the advanced options->UPS backup present->write cache enabled in drive properties for the onboard drives?

The two drives in question are running off SATA3 and SATA4 on my motherboard. Those two SATA channels are provided by a Silicon Image 3112 SATA Raid Controller (I have not turned on the raid capability). Selecting properties under either drive attached to these channels does not allow me to change the setting (I'm Admin user, but the driver doesn't allow settings to be changed - for both drives "Optimise for performance" has been selected and greyed out.

I have noticed that it does not matter which SATA drives I connect to these channels, the congestion seems to be with these channels rather than the drives. The reason I say this is that the same issue occurs with other SATA drives when put on the same channels and in all cases the System Event Log shows a multitude of "An error was detected on device \Device\Harddisk# during a paging operation" as soon as you try to perform a copy operation to the drives on these channels.

 

[egd]

I would try an installation of 2003 on a single ATA drive (Primary Master) as the only device on the cable. Then I would experiment with a RAID1 on each of the three Serial ATA controllers.

I've now exhausted all options including going down this path - see my comments re SATA 3 and 4 channels after your post.

 

[P5-133XL]

How many controllers are you using? Most controllers start acting up if you have more than one over the the native ATA. You need to search very hard to have multiple controllers that co-exist well with each other. Try 3ware if you want multiple controllers.

 

[egd]

How many controllers are you using? Most controllers start acting up if you have more than one over the the native ATA. You need to search very hard to have multiple controllers that co-exist well with each other. Try 3ware if you want multiple controllers.

I'm using the on-board ICH5, on-board Silicon Image SiI 3112 SATARaid Controller and a Promise FastTrak S150 TX2plus PCI based controller.

Both on-board controllers are set to SATA only ie I'm not using their RAID capability. The PCI Promise controller only has ATA drives connected at present.

Here is a rundown on what's running in the rig:
M/B: Abit IC7-G
M/B Chipset: Canterwood i875P
BIOS: IC7-G BIOS 28
CPU: Prescott 3.4E
RAM: 2 x Kingmax 512MB DDR
Video Adapter: GeForce FX 5200
OS: Windows 2003 Server

Hanging off this motherboard I have 9 devices as follows:

Cabled to IDE1: (master / slave)
1 MATSHITA DVD-ROM SR-8588 ( Drive Y: )
2 QUANTUM FIREBALL CR13.0A (12GB) ( Drive F: )

Cabled to IDE2: (master / slave)
- (no master installed at present, although with master present the problem remains)
3 Maxtor 6Y200P0 (190GB) ( Drive G: )

Cabled to SATA1
4 Hitachi Deskstar 7K400 (400GB, formatted to (373GB)) ( Drive V: )
Cabled to SATA2
5 Hitachi Deskstar 7K400 (400GB, formatted to (373GB)) ( Drive W: )
Cabled to SATA3
6 Maxtor 6 Y160M0 (153GB) (shows up as a SCSI Disk Device under Windows Device Manager) ( Drive H: )
Cabled to SATA4
7 Maxtor 6 Y160M0 (153GB) (shows up as a SCSI Disk Device under Windows Device Manager) ( Drive I & J: )

Cabled to Promise FastTrak S150 TX2plus PCI based controller and configured as striped array (ATA master / slave):
8 Maxtor 6E040L0 (40GB) ( Drive C: )
9 Maxtor 6E040L0 (40GB) ( Drive D: )


The BIOS is configured to boot the PC from the PCI device (Promise FastTrak S150 TX2plus) ie off the striped array.

This rig is not overclocked (other than being set to Turbo mode).

 

[Bozo]

I would double check the jumpers on the IDE drives to make sure Master/Slave is correct.

Did you load the drivers for the SATA controllers? On my motherboard with the ICH5 and Marvel SATA controllers, drivers must be manually loaded for each. This done during the install of the operating system ( press F6 ). You may be able to load them afterward though. I am running 2003 Server and XP and both need the drivers loaded.

In my BIOS I must enable all controllers to be used at the same time. I also must enable RAID on both controllers although I am not using RAID.
You can select individual / independant drives ( JBOD ) in the the setup software for you controller. This is usually accessed during boot up with a 'ALT + (some letter or number).

Bozo :mrgrn:

 

[time]

AFAIK, your SATA controllers are treated as SCSI devices under Windows.With the server OS, I believe write caching is disabled by default on SCSI drives, but not on ATA drives.

Storage Review has a summary of the 'problem'.

Alternatively, perhaps Windows has some sort of read after write verification enabled? Just a wild idea. From your testing, it doesn't appear to be any kind of RAID/mirroring issue.

 

[P5-133XL]

Could you disable one of the raid controllers and test to see if the transfer problems and error messages still exist; If not, then re-enable it and disable the other, then test again. It is quite possible that the fasttrack is not interacting well with the Sil 3112. My history is that multiple HD controllers generally don't co-exist well.

When I have needed multiple controllers like this, I ended up moving everything to a 3ware controller because I could have multiple 3ware cards and they do coexist well with each other. Other brands, were very problematic and I didn't have the time and money to test to find out which controllers works with which.

 

[egd]

I would double check the jumpers on the IDE drives to make sure Master/Slave is correct.

Did you load the drivers for the SATA controllers?

I've installed the latest Intel and ABIT supplied SI drivers. All devices are recognised etc, there are no problems with ny of the devices being recognised by the OS.

 

[egd]

AFAIK, your SATA controllers are treated as SCSI devices under Windows.With the server OS, I believe write caching is disabled by default on SCSI drives, but not on ATA drives.

Storage Review has a summary of the 'problem'.

Alternatively, perhaps Windows has some sort of read after write verification enabled? Just a wild idea. From your testing, it doesn't appear to be any kind of RAID/mirroring issue.

Thanks for referencing the article - it was the first thing I read and investigated before I made my initial post. SATA3 and 4 are treated as SCSI devices, however I'm not sure the SATA1 and 2 on the ICH5 are treated in the same manner. The more I work this through the more I'm convinced it is some kind of OS idiosyncrasy.

 

[egd]

Could you disable one of the raid controllers and test to see if the transfer problems and error messages still exist; If not, then re-enable it and disable the other, then test again. It is quite possible that the fasttrack is not interacting well with the Sil 3112. My history is that multiple HD controllers generally don't co-exist well.

When I have needed multiple controllers like this, I ended up moving everything to a 3ware controller because I could have multiple 3ware cards and they do coexist well with each other. Other brands, were very problematic and I didn't have the time and money to test to find out which controllers works with which.

I'll give your suggestion a go, but I think the most sensible answer is what you suggest - disable the on board controllers and get a 3ware controller(s) to handle all devices. One question though: does running all drives through a slot in PCI card reduce performance compared with the on-board controllers or does all traffic, irrespective of whether it originates in the Southbridge or a PCI card have to pass though the BUS?