SCSI v. SATA?

Axewerfer

Limp Gawd
Joined
Dec 1, 2006
Messages
202
I'm sure this has been debated countless times in the past, but I have yet to see a single consensus one way or the other, so I'm asking again for a quick summary of both arguments for future reference. What exactly are the differences between SCSI and SATA hard drives? For typical home use (i.e. gaming, etc.), is a 15k rpm SCSI drive network in RAID better than a 10k rpm SATA drive in an identical setup? Is there a difference at all? Why is SCSI so much more expensive? Anything else I should know about the two?
 
Hi,

There are several benchmarks that rate SATA higher than SCSI.

I think that unless you have an application that does random access like crazy across the disk, SATA II will be faster than regular U320-SCSI, or at lest not slower.

You might have a chance using the latest 300GB Seagate Cheetah 15k.5 SAS and a mainboard with onboard SAS controller.

Hans-Jürgen
 
Hi,

There are several benchmarks that rate SATA higher than SCSI.

I think that unless you have an application that does random access like crazy across the disk, SATA II will be faster than regular U320-SCSI, or at lest not slower.

You might have a chance using the latest 300GB Seagate Cheetah 15k.5 SAS and a mainboard with onboard SAS controller.

Hans-Jürgen

Umm, what are you smoking?
 
I'm sure this has been debated countless times in the past, but I have yet to see a single consensus one way or the other, so I'm asking again for a quick summary of both arguments for future reference. What exactly are the differences between SCSI and SATA hard drives? For typical home use (i.e. gaming, etc.), is a 15k rpm SCSI drive network in RAID better than a 10k rpm SATA drive in an identical setup? Is there a difference at all? Why is SCSI so much more expensive? Anything else I should know about the two?

SCSI is a robust platform designed to be run in servers and high end workstations. It's expensive because of the enterprise and critical environments they run in. SCSI (U320) will leave any SATA drive in the dust when it comes to business requirements and server functions, unless if you are worried about acoustics, heat, and power consumption. SCSI's are great in RAID applications. File servers and mission critical environments it's recommended to go the SCSI route.

SATA is more home user or large bulk storage, their reliability isn't as great, so therefore you pay a lot less for them. SATA drives are great for home usage and gaming, support is great for them on the end user spectrum and they are affordable to replace, unlike the SCSI counterparts, they are typically more quiet, use less power, produce less heat, and can be interchanged with many desktop systems.


If you are deciding, I would recommend SATA as the chances of an end user really needing SCSI performance would be slim.
 
Unless this is a server, go with SATA. If it is a server, and you don't know the difference between SCSI and SATA, go with SATA. If it's a server and you DO know the difference, well, you wouldn't have asked now would you?

But paying 300-400 for a 146GB drive when I can get 500GB for under 130, I know what I am using at home.
 
Short answer:
Unless you're building a database server that's going to see some moderately heavy usage, I would recommend SATA hardware over SCSI.

SATA is great for reasonably big, fast, cheap storage. Go with the SATA.
 
A high end 15k rpm scsi drive will be faster than any SATA drive in many, but not all, cases. SCSI and SAS drives are extremely fast, but they're optimized for workloads that are very different from normal desktop operation, so they may not be working up to their full potential loading WoW or whatever.

If you're going to run SCSI, be prepared to buy a serious SCSI controller. If it's not PCIe or PCI-X, you're probably wasting your time. Don't waste time with old SCSI drives either. Unless they're modern (within the last generation or two), modern SATA is faster.

If you still want SCSI for a single user system, plan on $200+ for the controller (though you may find a good one for less) and go with the latest Fujitsu 15k drives (MAX I think). They seem to be the best for a single user load.

With all that said, my next system will probably be running 4x 150 gig Raptors rather than 4x 15k SCSI drives. I'll keep the 15k stuff I have in the system for certain applications, but I plan on doing a lot of virtualization, and getting enough space for all of the OSs I want installed would cost too much using scsi.
 
Short answer:
Unless you're building a database server that's going to see some moderately heavy usage, I would recommend SATA hardware over SCSI.

Agreed. Here at school all our mail directories are currently served from sata arrays :eek: With ~10k users and pretty heavy usage (a few hundred "do i have mail" requests a second, among other things) a pair of Xserve Raid boxes plus some Sun V20z's doing ZFS and AFS stuff on LUNs exported from them handle the load fine. The mail servers are being upgraded/migrated to this, though :D We're getting the sata version of the 6140s because even with the highly random transactions mail servers do, tests show it's still plenty fast to use sata disks. That, and we're trying to move past 100MB quota for user accounts :rolleyes:
 
go with the latest Fujitsu 15k drives (MAX I think). They seem to be the best for a single user load.

the MAX is just the RoHS version of the MAU... so they should be the same speed-wise...

i have a pair of 73gb 15k Fujitsu MAX's in a RAID 0 and they fly man its great...

raid controller is some giant HP PCI-X with like 128mb ram, its U320 i dont remember the model # tho...
 
Thanks for the help! I'm not doing a build in the immediate future, but the science center hosted a 'deconstruction night' a while back. I was comparing an SCSI drive to a SATA,and couldn't help wondering why the former was so heavily built. Can I assume that the casing acts as a heat sink?
 
It's expensive because of the enterprise and critical environments they run in.

Actually, the reason SCSI is expensive is because when the standards first came out, the standard makers decided to charge very high licensing fees and that made SCSI too expensive compared to other technologies like IDE. IDE also put simple controller electronics on the drive itself, making IDE controllers cheaper and easier to make. If the license fees had been more reasonable, SCSI probably would have become more common since the performance improvement would have been within desktop power users' price points.

Now that SCSI has minimal desktop market share compared to the competition, the SCSI makers can't get economies of scale to compete on an equal footing. Hence, they limit themselves to the high end drives at very high spindle speeds.
 
Actually, the reason SCSI is expensive is because when the standards first came out, the standard makers decided to charge very high licensing fees and that made SCSI too expensive compared to other technologies like IDE. IDE also put simple controller electronics on the drive itself, making IDE controllers cheaper and easier to make. If the license fees had been more reasonable, SCSI probably would have become more common since the performance improvement would have been within desktop power users' price points.

Now that SCSI has minimal desktop market share compared to the competition, the SCSI makers can't get economies of scale to compete on an equal footing. Hence, they limit themselves to the high end drives at very high spindle speeds.

Hmm, in reality, SCSI always has been an open non-licensed standard. That's not the reason that SCSI drives and controllers tend to cost more. The real reason is that SCSI devices are typically a lot more intelligent than an ATA device. The SCSI command set has always been very large and robust and oriented toward handling many devices with concurrent outstanding IO's. This is less efficient to implement on ATA devices due to the way backwards compatibility was maintained when moving to faster transfer speeds.

A SCSI controller only need send a single simple command to a target, which is able to act on it and complete the IO separately from the controller. An ATA device is stuck using a set of 'shadow registers' and transferring FIS (Frame Information Structures) back and forth, which requires a lot more bus overhead, even when doing things such as NCQ and FPDMA transfers. Also, the flow control used in SATA is not very efficient and not easily scalable to a point that many drives can be handled by a single controller port (why a port multiplier on a SATA controller may not be able to perform as fast as you'd hope). Additionally, the SCSI command set allows a much more robust method of detecting handling errors and managing devices on the bus compared to what you can do with SATA devices.

What this all boils down to is what most everyone else in this thread has been saying, SCSI is great for heavy multi-user, random IO, such as what you'd find on a busy server. SATA drives work fine in single-user situations, but don't scale very well, due to protocol limitations and how the firmware in these drives are typically optimized for single-user systems.

Also, keep in mind that SCSI drives are typically built to withstand harsher environments than SATA drives. This includes things such as resistance to errors caused by rotational vibration, the ability to work at full load 24/7, and operate at higher temperatures without failures. For this last point, it's possible to make SATA drives that can tolerate this type of environment, but they are available at a cost premium as well, without all the other benefits of SCSI drives.
 
I'm sure this has been debated countless times in the past, but I have yet to see a single consensus one way or the other, so I'm asking again for a quick summary of both arguments for future reference. What exactly are the differences between SCSI and SATA hard drives? For typical home use (i.e. gaming, etc.), is a 15k rpm SCSI drive network in RAID better than a 10k rpm SATA drive in an identical setup? Is there a difference at all? Why is SCSI so much more expensive? Anything else I should know about the two?

SCSI is more expensive because they are built to last longer. If you check the product website, SCSI drives have longer MTBFs.

In addition, SCSI is meant for servers for multiple input-outputs. Therefore, for single user, you might not feel much improvements.

If you really want SCSI, you'll be better off with 15k. If you only want 10k, its better to stick to the Raptors.
 
Hmm, in reality, SCSI always has been an open non-licensed standard. That's not the reason that SCSI drives and controllers tend to cost more. The real reason is that SCSI devices are typically a lot more intelligent than an ATA device. [...]
All this sounds a lot like the information that I have heard over the years. SCSI was (is?) expensive, since the drives were intelligent. To answer the OP's question: there are few reasons for a "average Joe" desktop user to move to SCSI (SAS). Unless you have a very specific workload that benefits tremendously from drives being able to handle multiple outstanding I/Os, I do not think that going SCSI is a financially sound choice.
 
SCSI/SAS does a good job of filling in the moderate usage realm of computing. Your $ per GB jumps drastically as you go from SATA to SCSI/SAS, and even moreso when you go to Fiber Channel.

SATA = Light usage, low I/O load. Optimized for single-user environments.
SCSI/SAS = Moderate usage, moderate I/O load. Optimized for multi-user, single server environments.
FC = Crunchy usage, crunchy I/O load. Optimized for balls-out crunchiness for one or more servers.

That being said, our SANs here do have multiple blades of SATA drives. Then again, the zones that the SATA resides on are used for disk-to-disk backups, and our newest SAN has something like 128GB of cache. There are 15k drives in the SAN for crunchy apps, no idea if they're SCSI, SAS, or even FC though. Hitachi manages all that and I haven't asked our tech about it.

Unfortunately I know jack about SCSI/SAS compared to what I know about FC. When I need crunchy performance I just set up a zone on the SAN and mount it on the server needing crunchiness. :)
 
Actually, the reason SCSI is expensive is because when the standards first came out, the standard makers decided to charge very high licensing fees and that made SCSI too expensive compared to other technologies like IDE. IDE also put simple controller electronics on the drive itself, making IDE controllers cheaper and easier to make. If the license fees had been more reasonable, SCSI probably would have become more common since the performance improvement would have been within desktop power users' price points.

Now that SCSI has minimal desktop market share compared to the competition, the SCSI makers can't get economies of scale to compete on an equal footing. Hence, they limit themselves to the high end drives at very high spindle speeds .

You disagree with me, but then you say this:

"IDE also put simple controller electronics on the drive itself, making IDE controllers cheaper and easier to make"

"Hence, they limit themselves to the high end drives at very high spindle speeds"

Which reinforces my point....


Also, just FYI, the SCSI makers you are so talking about, are the same ones rolling out the IDE and SATA counterparts.

"the SCSI makers can't get economies of scale to compete on an equal footing"
 
You disagree with me

Don't be so sensitive. I'm only pointing out that things were different 25 years ago.

but then you say this:

There isn't just _one_ reason for the price difference. However, licensing fees were a significant part in the old days. I have long since thrown out the issue of Byte magazine that contained the article that discussed these differences when IDE first came out, otherwise I'd give you a specific reference for the license cost. In addition to that article, several writers in the various PC mags in those days were calling for the SCSI folks to drop licensing charges to make it more competitive. Since they didn't, they lost out to IDE in market share. Excepting wider,faster versions of SCSI, IDE wasn't that much slower on the desktop than the basic SCSI versions available.

In the old days, when SCSI drives were the same physical drives as the IDE drives (like some SATA/PATA versions today), the SCSI drives were more expensive and you needed a pricey controller. The only performance improvement came from the faster speed of SCSI command processing and (significantly) channel speed. I had a SCSI hard drive in those days for speed and it was $$$$$ - since it was the same physical drive as the equivalent IDE, the price markup was in the license fee and the onboard electronics.

Also, just FYI, the SCSI makers you are so talking about, are the same ones rolling out the IDE and SATA counterparts.

But they don't produce anywhere near as many SCSI boards for the drives and since they concentrate on the high spindle speeds, they are not making the same parts in the same volumes. Companies like Adaptec don't make the volume of SCSI controllers compared to the volume of IDE controllers on MBs.
 
Back
Top