[JohnR]

I've always understood that SCSI (or SAS - Serial Attached SCSI) is inherently a higher-performance standard than IDE or SATA etc. When you look at the specs for these drives, SCSI drives have 15k RPM speeds, and there is I understand much better buffering and disk request management in SCSI controllers.

Our (AudioCircle) server has a 15k SCSI drive for that reason. Non-standard, and it cost us some money, but it hasn't let us down so far. However I have no comparable benchmark to compare it to. I recently (2+ years later) ordered a desktop machine for work and jumped at the chance to get a 15k RPM SAS drive instead of the standard SATA drive. This machine will be under fairly heavy load, so the incremental cost over SATA seemed well worthwhile.

Anyone with opinions or experience on the choices here?

[machine]

These types tend to be targeted at the server market and are made to a higher standard.

Consumer SATA/IDE are more targeted to a price point/competition based on price.

I try and stick with Seagate units with the five year Warranty.

Barracuda drives (SATA) have a Mean Time Before Failure (MTBF) of 700,000 hrs.
Cheeta drives (SAN) have a MTBF of 1,600,000 hours.

The SATA drives are larger and cheaper.  One strategy is to use RAID with these drives to avoid data loss.

[mgalusha]

Well SAS is Serial Attached SCSI while regular SCSI uses a parallel architecture. One of the advantages that SCSI drives held over IDE/PATA (Parallel ATA) drives was command queuing but now SATA (Serial ATA) drives have this as NCQ - native command queuing. With multiple drives on the bus different drives could be doing different things while two IDE/PATA drives on the same bus had to wait for each other. At least this is the way that I learned it.

I have built many a server and in the past few years some with SATA drives. The servers with the fast SCSI disks have always provided better IO but it's an awfully expensive way to go, especially if you need redundancy, which of course a production quality server must have. We use RAID 0 + 1 for some things where performance counts but at the cost of double the disks. More cost for the disks, heat, power, space but for something like a database that is taking 1K hits/second the cost is worth it.

If you do use SATA, which for a non server environment is likely plenty good try and make sure the NCQ is turned on. This is not always the default. My music server at home uses 4 x 500gb SATA drives and an inexpensive Promise array controller. Be default NCQ was off for all drives. The difference in performance from enabling it was very significant.

mike

[JohnR]

Ah, fair point Mike, I'm thinking of situations in which high concurrent access is assumed/needed. (In my examples, web server, and running a number of virtual machines to simulate a cluster.)

However, the cost for SAS doesn't seem all that much higher these days. In fact, I was surprised...! I felt that getting smaller SCSI/SAS disks, and using slow disks for backup as needed, was a good investment.

Again, just wondering what choices/tradeoffs people have made

Still, I'm wondering how a 15k RPM disk vs a 7.2k RPM disk could be a bad thing? I've never understood why the specs are so different.

[brj]

From my reading, SCSI in any form still wins in a server environment, but has a much harder time justifying its price in a single user envionrment.

By far the most comprehensive source of data and benchmarks that I've found for hard drives is http://www.storagereview.com/.  Excellent site.

(I also check http://silentpcreview.com/ when building personal machines at home, as noise matters to me.)

[jqp]

I think it is a price/performance thing. If you need performance and reliability in a server environment you can justify SCSI.  I also think server operating systems are designed to take advantage of SCSI with higher RPM drives.

If you are at home and need the most space for the buck and frequently adding on/upgrading you will gravitate towards SATA.

I have an old server box I built for home. I bought a motherboard with an Adaptec SCSI contoller on it - I have 2 SCSI CD burners and I think 1 SCSI hard drive. But then I bought ATA drives as I expanded my space, because they were half the price and I am just not really serving data and processing at home - just couldn't justify the cost, and I love SCSI.

Many wanted to see SCSI succeed over ATA even at home but I think the SCSI licensing/manufacturing powers that be wanted too much money for it to be accepted in the consumer market - kind of like USB vs Firewire

[JEaton]

All else being equal, the 15k RPM spindle speed alone will net you nearly a 100% performance increase over a 7200 RPM disk and 50% over a 10k RPM.  On a workstation, even a very busy one, the SCSI interface itself probably doesn't gain you much.  SCSI shines when you have many simultaneous requests being queued, such as on a web server, database server, heavily loaded email server, etc.  For mail and web servers, applications tend to be comprised of many small files, so the greater access speeds of faster rotating disks can make a big performance difference.

One advantage that the newest 7200 RPM SATA drives tend to have, which lets them approach the transfer speeds of many 10k RPM SCSI drives is that they tend to use the latest, highest density disk platters, while the SCSI drives tend to use older, more stable disk technology.  Many of today's 7200 RPM disks have transfer speeds that exceed those of 10k RPM SCSI disks of a couple of years ago.

A lot of large servers and server farms these days have switched to using large SATA drives instead of SCSI.  The cost difference when you're talking about many 10's or 100's of terrabytes of data is just too hard to ignore.

[jqp]

  SCSI shines when you have many simultaneous requests being queued, such as on a web server, database server, heavily loaded email server, etc.  For mail and web servers, applications tend to be comprised of many small files, so the greater access speeds of faster rotating disks can make a big performance difference.

Thats what I was trying to express

  The cost difference when you're talking about many 10's or 100's of terrabytes of data is just too hard to ignore.

And that

[Levi]

I want to commend John for using the best equipment! 

I agree with everyone.  Sata drives with NCQ is very close in performance compared to SCSI counterparts. 

It is that 15K rpm, simultaneous read/write and fast seek time SCSI drives has the advantage over Sata.  HDDs are physically the same, it is the controller chip in the drives and interface that makes the big difference. 

In my own experience SCSI drives are faster using my butt dynanometer.  I installed Windows XP to 10K rpm WD raptor NCQ enabled.  I also installed (same computer) Windows XP in 15K rpm Fujitsu SCSI320 HDD interface.  Switched between drives over a week.  I didn't do a bench mark but I like Windows XP installed on the SCSI320 drive better.  Boot times seems to be similar in speed but when I tried multi-tasking, e.g. ripping/burning DVD and play network games simultaneously...SCSI hdd seems to have the smoothest response.

Servers on the other hand specially database servers, I would only recommend to use the fastest SCSI interface.  That's just me.

Cheers,
Levi

[jqp]

Levi are you talking about the 300GB Velociraptor?

[Levi]

I have the 72gb raptor with the see through window. I find smaller drives quicker than larger drive.

[mizzuno]

Many newer drives are moving towards the 2.5" format due to higher platter density. The smaller platter provide much faster access times. I highly recommend http://www.storagereview.com for analysis on hard drives. While rotational latency and disk access times are important, so too is the RAID striping you are using. RAID 1 giving you disk mirroring (redundancy, good read speeds, ok write speeds) , RAID 0 striping (no redundancy and double the mtbf, great read and write speeds), RAID 5 (high capacity, while tolerating the loss of 1 drive, very good read speeds, slow write speeds), and RAID 10 mirroring and striping (high capacity high redundancy with great read and write speeds) frequently used in database servers. There are other flavors of RAID with greater redundancy and performance tweaks. A combination of how many drives you have the actual RAID controller and the speed of the disks (latency and actual throughput) will determine the performance of your server. RAID levels are generally determined by understanding your ratio of reads-writes and the amount of redundancy you need.

[Levi]

+1.  Agreed, looks like the newer 2.5" drives are indeed faster compare to the 3.5". 

You mileage may vary.

[JEaton]

Many newer drives are moving towards the 2.5" format due to higher platter density.

I think the reason that 2.5" drives are becoming more popular in servers is that you can simply pack more drives in the same space. With RAID 5 and RAID 6 more drive spindles means greater access and transfer speeds.  A one rack-unit (1U) server is limited to four 3.5" drives, but can house as many as ten 2.5" drives.  A 2U server, twelve 3.5" drives, but 24 2.5" drives.