[rommel]

ayone know anything about this and how the performance stacks against say 160/320u scsi?

[Mickey]

I don't see why serial ATA drives in RAID would perform any differently than parallel ATA drives in RAID, since it's unlikely the mechanics have changed in either case. The new interface just increases the available bandwidth; it doesn't mean the drives are actually going to become faster.

[vrythmax]

Initially SATA will provide only the same protocol commands as PATA, however the controllers in some cases may present themselves like SCSI controllers to the OS like some PATA controllers today. If so, they may improve performance over PATA RAID because they would be able to drop the Master/Slave limitations inposed by PATA style drive access. I would bet that while the drives themselves are not faster, SATA RAID will be able to take advantage of independant busses for each drive when presented to the OS as SCSI by the controller.

[docinthebox]

Quote

I would bet that while the drives themselves are not faster, SATA RAID will be able to take advantage of independant busses for each drive when presented to the OS as SCSI by the controller.

So would SATA RAID be any faster than PATA RAID using a separate channel for each drive?

[rommel]

yeah i think that was the point he was trying to illustrate....by using separate buses the I/O will be increased...i was jus twondering how SATA drives themselves compared to scsi....i am running seagate cheetahs now and was looking to go to 320u when i read about serial ata...so anyway.....looks like its scsi 320u

[docinthebox]

What I meant is: anyone running PATA RAID would put their drives on separate channels. I don't think anyone would be silly enough to run, say RAID 0, with one drive as master and another as slave in the same PATA channel. So I bet there would be no performance difference between PATA RAID and SATA RAID.

[Jan Kivar]

SATA would be faster than PATA in RAID0 with three or four SATA HDs. They would outperform 2-channel (two connector) PATA controller with three or four HDs (because 4 versus 2 channels). Performance would be similiar with 3ware card or 4-channel (four connector) model from Promise or Highpoint though.

Too bad that standard 32b/33MHz PCI bus caps the performance with four (or even three) disks.

Cheers,

Jan

[Haversian]

While that is true, it only affects add-on SATA cards, not the channels integrated into the motherboard, which are universally connected directly to the chipset. This limitation of IDE controllers using the PCI bus was recognized a few years ago and corrected. Thus the proliferation of new NB-SB connection schemes, a'la V-Link. The PCI bus was moved from the northbridge to the southbridge, and thus the ATA controllers no longer have to share PCI bus bandwidth.

Your complaint though hits RAID controllers hard. I haven't asked 3ware whether the new cards (native Serial ATA controllers early next year, supporting online capacity expansion!) will be 64/66 or will be 64/33 like the current crop. I imagine they will transition to the faster bus, as 16 SATA channels can blast past 64/33 PCI without even breathing hard. Then of course the problem becomes finding motherboards supporting the faster PCI buses. Asking for faster PCI buses is one sure way to kick yourself out of the enthusiast market and into the workstation/server market, where $400+ motherboards aren't uncommon.

[Jan Kivar]

All current motherboard integrated IDE RAID solutions connect to the PCI bus, not directly to V-Link or HyperTransport or whatever chipset is in use. Standard on-board controllers are integrated to the southbridge, so they can enjoy the greater bandwidth to the northbridge. I think this is not going to change, as RAID chips are purchased from Promise or Highpoint. They'll continue to connect to PCI bus being either SATA or PATA.

But one could run software RAID0 with the on-board SATA connectors (hopefully all new motherboards will incorporate at least four connectors) and get performance increases. Software RAID solution eats up some CPU cycles though.

Cheers,

Jan

[sechs]

So, the answer is, no, they will not perform appreciably better than their PATA counterparts. So, go SCSI!