8 Responses to “SATA vs SCSI?”

1. Ernst Lopes Cardozo Says:
   March 19th, 2007 at 11:14 am

   Tom, I have borrowed your spreadsheet that calculates reliability (TTDL, Time To Data Loss) for various drive and RAID types. I wanted to see what would happen if I (tried to) spend the same amount of money that a SCSI/SAS array costs on a SATA array.

   Of course, you get something like 3 times the useable capacity. Even with RAID6 for the SATA array. Even with smaller RAID6 sets than the RAID5 SAS array. But that was to be expected.

   The TTDL of the SATA RAID6 array is way better than the SAS array, _assuming the drives have the same MTBF_ as per the recent research and a lot of people’s gut feeling / experience.

   The sequential performance is much better, since SATA drives have higher density and we have more to stripe. No surprise really.

   The random IO performance is better too. Not by a factor, but the SATA RAID6 arrays are faster, not slower, than the SAS array. Strength in numbers.

   Bottom line: for the same money, SATA delivers more capacity, more speed and considerably better reliability. Some of the largest storage users seem to have figured that out already.

   OK, this is a theoretical exercise. You need an array and RAID controllers that are up to the task. You need more rack space and a bit more power and hence more cooling. Forget the easy rules, do your own math.

2. Tom Says:
   March 19th, 2007 at 11:52 am

   Ernst, I couldn’t agree with you more. Excellent summary of the points, BTW. Extra credit for using the spreadsheet.

   And I don’t know what’s more ridiculous - (a) drive vendors selling huge SATA drives for $0.02 profit or (b) drive vendors selling tiny SAS drives for the price of my left nut (excuse the vulgarity, but it just fits too well to pass up).

   But I think that’s all about to change. I’ve seen quite a few drive roadmaps under NDA that show SAS drives finally having SATA capacity and SATA drives finally having SAS reliability (e.g., nearline drives). This is starting to cause drives to have a spectrum of prices rather than just too damn low or too damn high - and that makes plenty of sense. Personally I want to see drive vendors stay in business and make a decent profit. It’s silly for a drive vendor to sell SAS drives at obscenely high prices just to balance their business decision that they have to practically give away SATA drives.

   As I’ve said many times, there’s nothing in the SATA spec that says “cheap, unreliably and large”, and nothing in the SAS spec that says “expensive, reliable and small”. And it’s always really bothered me that there are two specs to begin with. I do understand the technical reasons why we need two different command protocols, but these standards should have always been handled by just one committee. I have firsthand knowledge - I was on both committees.

   Oh, well. I’ll stop my whining before I get all worked up. Everything will come out fine in the end.

   TT

3. Dan Says:
   April 23rd, 2007 at 3:06 pm

   Isn’t the current capacity of SAS drives 300 GB not 300 MB? While significantly different, this is not the giant chasm of difference that 300 MB would be. Otherwise, well covered response to a question lacking in some critical details e.g. how critical is uptime how critical is the data etc.

4. Tom Says:
   April 24th, 2007 at 6:21 am

   Oops. Good catch. Fixed.

   Thanks, Dan.

   TT

5. Peter Says:
   May 6th, 2007 at 5:51 am

   One wonders about SATA/SCSI/SAS/cheesecake at times.

   Fundamentally, the hardware appears the same. Compare and contrast drives from one maker for some capacity and different interfaces. You have a rapidly rotating chunk of glass (or aluminium) coated with particularly high grade rust. A brushless motor spins these at the same speed regardless. A voice coil driven head on a spring. The only difference is likely to be the electronics taking orders. Given the SAS/SATA compatibility, that might be even be down to whether its type A or B firmware.

   With all these similarities. Why are there not 1Tb SCSI drives? Conversely, 15k SATA drives.

   Is there some cartel out their fixing prices, maintaining the high-volume, low profit SATA and low-volume, high profit SCSI?

   Conspiracy theory or is there a real deep down hardware difference?

6. Tom Says:
   May 10th, 2007 at 12:25 pm

   Peter, you are completely right. There is a conspiracy. The drive vendors have arbitrarily tied cheap components, slow motors, a poor media to the SATA interface, while the expensive, good parts are all saved for SAS and SCSI. And there is absolutely nothing in the SATA, SAS or SCSI specs that define the preferable hardware behind the interface. Sure, SAS has a few more high-end features, so I guess I can see tying SAS to more expensive hardware, but it never had to be that way. Those same features could have been put into SATA. In fact, not to go too far into the weeds, but why is there a SAS and SATA to begin with? Why couldn’t there be just one interface, still building cheap drives for desktops and expensive drives for servers? The reasons behind that is a much longer story…

   It all comes down to this hole the drive companies have dug where they’re forced to sell their low end drives for just pennies of profit, and then make it up by ripping off their high-end server customers. SAS drives are often 2-4X the price of SATA drives, but are they 2-4X better? Maybe they’re faster than SATA and maybe more reliable, but they’re also a heck of a lot smaller.

   But with that said, there is starting to be some middle ground: nearline drives. These are SATA drives that have borrowed technology and hardware from SAS drives to make a fairly inexpensive drive with higher performance and reliability. Some more information on the Seagate version of nearline drives can be found here.

   OK, now that we’ve brought this conspiracy into the open, we better start watching out for the black helicopters. Oh yes, they’re out there. Trust me.

   TT

7. Greg Says:
   August 15th, 2007 at 6:09 am

   SCSI vs. SATA - real world.

   I’ve performed an intensive study into this and have come to the conclusion that SCSI 320 is still better technology than SATA (including SAS). Here’s why;

   In real life, I’ve seen both raid 5 and raid 1 work with REAL users running on servers. For performance reasons, I only use raid 1. Yes, I lose half the capacity, however, the performance increase on the bus is worth it. I’ve seen near double the overall performance of raid 1 over raid 5 in real life usage. Next is the expandability of SCSI 320, easy to add drives without adding controllers hardware. The only reason for point-to-point is to gain profits for additional drive installations (SATA or SAS) that I can see.

   Yes, I’ve got SATA raid 1 systems out at customers sites- they’re MUCH slower than the SCSI counterpart. My standard load-out on a server is two raid 1 sets (4 drives total) - one set for the OS and one set for the data (all drives 15K). I don’t care how fast your processor is, or how much memory you have - if your drive system is slow, the whole system is slow. Remember - the point of a file server is for central point of storage & sharing of user data (and central point of backup of said data). If users can store data faster on their local HD - you got a problem (yes, it sounds like an expansive system, but my customers demand it). When they see the cost, they choke a little - however, when they see the performance of their system I never hear another word on the price.

   When I build a system, I always use the DPT (now Adaptec) raid controller with battery & ECC memory. Not many controllers ECC all cached data & the bus. Not only that - not many controllers store whats in cache in case of power failure and write the information to the drives when power is restored.

   One of my vendors has sold thousands of that raid controller and says that he only knows of one that was returned for failure (he thinks it was due to the guy who installed it) - I’ve never seen a failure.

   As far as the drive constructions - yes, the basics are the same, however the electronics are completely different. On a SCSI drive, the drive itself holds the brain for communicating with it, on SATA, its the controller.

   If you want more detailed reasons as to why SCSI 320 is still better - look at the actual specifications for SCSI, SATA, & SAS. Look real deep and understand how each one is different than the other, data throughput, overhead of controlling data, overhead of the controller itself to the OS and the interface to the computers bus.

   Now if you want to compare price - total up the cost of each drive (matching as best you can for seek time, RPM, and size as well as warranty), then the cost of the controllers for 15 drives - then you will see that SCSI is not much more expensive than SATA.

8. Tom Says:
   August 23rd, 2007 at 6:56 am

   Hi, Greg. I really appreciate the shout-out for the old DPT controllers. I was with DPT for 15+ years, practically since the beginning, when Adaptec bought us. We were very proud of what we accomplished with such a small team. A lot of those fine folks are still with us here in the Orlando office.

   As far as RAID-1 vs RAID-5, yeah I certainly agree that RAID-5 can be a real dog if you’re doing a lot of short random writes. Performance can easily be half that of a RAID-1.

   Regarding point-to-point, I think that is the one real advantage of SAS (and SATA) over the older parallel technologies. With SCSI a single cable failure will bring down an entire string of drives. That won’t happen with point-to-point – usually, and I’ll get to that. If you have a multi-port RAID card you will have a separate cable per drive. While the cable redundancy is great, I admit that having a bunch of individual cables can be a real pain to deal with.

   Therefore most folks with that many drives will use a backplane with an expander – kind of like a switch. You’d run a single wide 4x cable to the backplane and then individual traces go to each drive. And you’d be correct in pointing out that you’re back to a single point-of-failure in the 4x cable. But this cable is still more fault tolerant because it’s only got two connectors – one at the end attached to the controller and one at the end attached to the drive backplane. With SCSI you have multiple connectors spaced at very specific locations on the cable, and of course you also get to play with the wonderful world of SCSI parallel terminators. Now you can’t tell me you enjoy that, right? I think one of Dante’s circles of Hell had a reference to SCSI termination.

   Regarding performance of SCSI vs SATA, it’s hard to argue with you. Besides the point you make about the intelligence being on the host instead of the drive, most SATA drives are still made with inferior motors, heads, etc,. when compared to modern enterprise drives – be they SCSI or SAS. But some of the SATA nearline drives aren’t half bad. And WD also makes a damn fine SATA drive.

   I encourage you to take another look at SAS drives and controllers. I’d be very surprised if you found SCSI exceeding, or even meeting, the performance of modern SAS. Most of the new controllers, including Adaptec controllers, have CPUs and memory speeds that far outpace that of the older SCSI controllers.

   Thanks for reading, Greg. And if you need any DPT controllers I could probably sell you a few out of my trunk. Don’t tell anyone.

   TT

Leave a Reply

Name (required)

Mail (will not be published) (required)

Website

By submitting a comment, you understand that if your post violates any federal, state or local law, or contains libel, slander, defamation, product disparagement, harassment, obscenity, or indecency, it will be edited or deleted.