[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: SCSI vs SATA research

On Fri, 22 Jul 2005, Sean Sweda wrote:

On Jul 22, 2005, at 8:14 PM, Michael C Garrison wrote:

In the hardware meeting I was asked to gather information on SATA vs SCSI and I wanted to provide what I have found so far. Besides my own benchmarks, which will hopefully be ready for the next hardware meeting, this is the information I have found. I'm sending this to the UMCE list as everyone in the hardware group is on this list, and additionally I think others will find this information usefull.

This is interesting stuff, thanks!

I'd like to narrow the focus of the discussion. I think we've already agreed that when capacity is a concern that moving toward SATA is a win, and that a significant number of future purchases will be SATA-based. I also believe we agreed that we want to have as few *types* of hard drives in production as possible. At this point we have effectively standardized on two different drive models: 400GB 7200 RPM SATA and 36GB 15K RPM SCSI. Therefore, it seems to me the questions we need to address are:

I agree.

1) What is the real performance difference between the drives upon which we have already standardized when they are deployed on 1U hardware that meets our requirements.

This is a good question. I have only tested on 200GB drives, but since we received the machines with 400GB drives in, I hope to have a report on this before the next hardware meeting.

2) If the performance of the 400GB SATA drives in 1U configuration is worse than 36GB SCSI drives in the same configuration, is it acceptable for 1U deployment given the price difference.

3) If the performance of the 400GB SATA drives is not acceptable, are we willing to introduce another hard drive type (Western Digital Raptors) as an alternative to SCSI.

I think that, aside from your reasoning above and keeping in mind that we want to keep as few types in production as possible, that this is still something to consider. If high IO is important to a project and disk space is not, these are the drives to go with.

4) If we are considering adding another hard drive type, should we only consider a SATA option? SCSI is moving to a serial architecture as well (SAS), and it shares the same electrical & physical connectors as SATA. This means, theoretically, that we could standardize on two types of drives (one SATA, one SAS) and freely interchange them in servers based on need. Have we even investigated SAS?

I have done a little reading on SAS and in some of the sites I referenced they discuss SAS as well. The CERN powerpoint slides discuss it also. SAS is great on paper, but this year and last, the adoption rates have not met what they projected. Some large vendors believe that, as of right now, SAS presents no benefits over SATA so they are not adopting it.

There are a few reasons for this
- SAS is geared more towards large enterprise technology
- SAS disks are still very expensive
- SATA drives are widely adopted by SAN vendors because of their cost savings.

When SATA and SAS first emerged, the main benefit SAS had over SATA was it had tagged command queuing in its specs. With the adoption of SATA-II (and beyond..), the specs include NCQ (native command queuing) so that advantage is now (mostly) gone. [side note: NCQ is similar to the command quieing or tagged command queuing that SCSI, Fibre Channel, and SAS have but not the same].

My personal opinion is that as the market is right now, SAS is not an alternative to SATA or SCSI.. more so to fibre channel, which is out of our range. In addition, with the current adoption rates (or lack of) and advancements of the market.. unless it changes, I think SAS is not worth it.

Mike Garrison