FATA drives
Posted in Storage Interconnects & RAID, Advisor - Tom Treadway by Tom TreadwayQuestion to the Storage Advisors, from anonymous: Where do FATA drives fit in?
AFAIK, Seagate is still the only vendor providing FATA drives. They’re basically just cheap, low-end SATA drives with a FC interface. FATA drives are a nitch product directed to one specific Seagate customer - HP. It seems that the only folks that would use FATA drives are those that have already been sucked into the expensive FC religion and their CIO is forcing them to cut costs. It’s like paying for a Turbo Porche but then using low octane fuel.
TT
December 8th, 2005 at 12:51 pm
Comparing the price per GB disk for disk, FATA drives comes in at less then half compared to FC. Although the price per GB goes up when you add up the cost of the EVA array itself, it’s still gives you a substationally lower cost per GB, while at the same time giving you all the goodness that is the EVA when it comes to management.
Customers love it, they can use the same management gui and ease of operation and even mix FC and FATA drives in the same array, while not compromising on reliability an awful lot as long as you keep in mind FATA drives are not meant for heavy I/O 24×7, just as SATA drives.
December 8th, 2005 at 1:06 pm
Charles,
Thanks for your insight on this.
When the FATA drives were originally announced last year I had seen reliability data similar to ATA - an MTBF of ~500K hours, assuming the drive is only used 8 hours a day, 5 days a week. This is less than a 1/10 the MTBF of FC, which seems like a huge drop. But to be fair, if the drives are used in a RAID-5 or RAID-6 array, then that reliability drop could be avoided.
So I just tried to double check my numbers, but I wasn’t able to find the FATA drive on the Seagate website. Is this drive being shipped only to HP now? There was originally talk of Hitachi also doing a FATA drive, but I can’t find that one either.
Any idea where these things are being sold? They seem to have lost a lot of steam since their 2004 launch. You sound like a user (or maybe reseller) of FATA drives. Can you perhaps share some details on your enviroment?
TT
P.S. Mixing FC and FATA drives in the same array? Doesn’t that give bizarre performance characteristics?
December 11th, 2005 at 11:05 am
I work primarily with HP in a reseller organisation, my role is consultancy, design, implementation and to a lesser extent operations and maintenance.
My customers use FATA drives for various tasks:
* Backup devices for backing up to disk.
* Replication targets for replicating branch office data into the data centre.
* Tier 2 storage for archival solutions, such as Symantec Enterprise Vault.
* User data in the form of home directories. This might sound strange, but my experience shows that this load is fairly light. Most is done in the file servers cache.
What is not on FATA:
* Databases
* Exchange
* Other critical apps
As for where FATA drives are sold, I installed a new EVA with FATA drives only last week. It was going to be used as a backup device.
We sell quite a lot of them.
It’s perfectly fine to mix FC and FATA in the same array frame, but you don’t want to create a LUN spreading over both FC and FATA. The EVA implements disk groups, and typically you place all FATA drives in one group, all FC in another. Then a LUN is carved from one group.
February 13th, 2006 at 10:28 am
The one advantage to FATA is that if SBOD monitoring can truly predict imminent drive failures well, FATA in an SBOD array might be a good combination for improved reliability over SATA (although TCO is another matter if you are replacing a lot of “false alarm” failure predictions).
February 17th, 2006 at 6:24 am
But FATA is just an ATA drive with a FC interface. SATA is an ATA drive with a serial interface. They can both do PFA or SMART. So I don’t see how FATA is better than SATA.
In fact, SATA is a much better alternative to FATA because it’s SO much cheaper. SAS initiators, switchs and drives are almost 1/10th the price of FC.
TT
February 25th, 2006 at 12:14 pm
I have to ask the question about two ports to the drive. FATA has a second port, SATA has only 1. Every application I have seen requires that a little drive Dongle card is added that bolts up to the SATA drive so there is an alternative path to the drive. Is HP taking advantage of the second path in their Arrays?
March 24th, 2006 at 11:23 am
That’s right. FATA drives are dual-ported while SATA drives are single-ported. There’s been talk of some of the SATA suppliers (especially the ones that don’t have SAS drives) putting the mux chip (the one you referred to on the dongle) on the drive. It would add cost, but would also be real interesting to replace SAS drives in dual-controller external enclosures.
TT
March 31st, 2006 at 4:57 am
How far can the MTBF figures quoted be offset by RAID 5 or 6 ? Worth doing for backup to disk; e.g. Connected Backup?
April 3rd, 2006 at 12:44 pm
RAID-5 and 6 make a HUGE difference in MTBF. I’ve made several posts on how to calculate the reliability of each RAID level.
http://storageadvisors.adaptec.com/2005/11/01/raid-reliability-calculations/
http://storageadvisors.adaptec.com/2005/11/02/actual-reliability-calculations-for-raid/
http://storageadvisors.adaptec.com/2005/11/03/is-raid-6-made-of-wood/
Let’s go through an example.
Assume that you had 100 customers, each with 16 SATA drives. Each customer could either have (a) a RAID-0 with no protection, (b) a RAID-5 with protection from a single drive failure, or (c) a RAID-6 with protection from a double drive failure. Using typical MTBF and BER rates, you’d have the following MTTDL (Mean Time To Data Loss) for the various scenarios, meaning that one of your customers experienced a failure that rendered their data inaccessible.
RAID-0: 26 days
RAID-5: 68 days
RAID-6: 361 years
As you can see, the benefit of RAID-6 is pretty significant.
TT
May 2nd, 2006 at 6:06 am
=============
excerpted from: HP StorageWorks Enterprise Virtual Array (EVA) Storage Solutions: High-Performance and Low-Cost Disk Drives
=============
“A wide variety of tests were run on sixteen 146-GB 10K disk drives and sixteen FATA disk drives to have a base line performance comparison. The resulting data indicated that the random I/O of the sixteen-member FATA group was approximately 25% less than the random I/O of the146-GB 10K sixteen-member group. The sequential performance of the FATA group was essentially the same as the 146-GB group.”
If FATA drives are half the price of FC, wouldn’t you get better performance (over FC) with twice the number of FATA drives?
May 2nd, 2006 at 5:12 pm
Yep. The random performance is directly proportional to the number of drives. Assuming there is no controller, OS or bus bottleneck, then doubling the number of drives will double overall performance. And as HP points out, FATA (or ATA or SATA) drives are typically about 25% slower for random IO than FC (or SCSI or SAS) drives.
So if you have 50 FC drives at 100 IOPS you’d get 5000 IOPS. And with 100 FATA drives (at the same price as 50 FC drives) at 75 IOPS you’d get 7500 IOPS. That’s a 50% performance boost with random IO. Not bad.
The reason sequential doesn’t increase is due to (a) the drive media rate is roughly the same for low and high-end drives, and probably the biggest readson (b) the IO bus will saturate with just a handful of drives.
Thanks for reading, Casey.
TT
May 3rd, 2006 at 9:38 am
So if we can achieve a 50% performance boost over FC drives, shouldn’t we want to use the FATA drives? (all drives, FATA or FC, would be RAID 6)
I guess I’m asking you for a reason why we shouldn’t use FATA drives. It seems to go against most Storage Guru’s conventional wisdom, but I think we may have a considerable performance advantage by using twice as many FATA drives.
Thanks for the feedback Tom!
May 3rd, 2006 at 12:51 pm
The 50% boost in performance going from FC to FATA and doubling the drive count is only if you stick with the same RAID level. Some folks would argue that FATA requires RAID-6 to overcome the lower MTBF and BER. With 100% random reads FATA RAID-6 would still be 50% faster than a FC RAID-5, but with 100% random writes FATA RAID-6 would be 50% slower.
Also, FATA drives are single sourced from Seagate. There was an announcement from HP that they would be doing FATA drives, but apparently they’re codeveloped with Seagate, and it looks like they’re not shipping yet.
So, if you already have a FC environment, then the discussion of FC or FATA is a good one. But if not, then you (and I don’t mean YOU necessarily - just you in the generic sense) should be looking at SAS vs SATA. FC initiators and switches are much more expensive than SAS. Eventually the SAS and SATA initiators will be free out of the motherboard chipsets. And there are more drive vendors for SAS and SATA.
To answer your orginal question: If you’re already using FC, and SAS is out of the question, then “yes”, you should look at using FATA. But make sure you check the reliability of these drives. If you’re forced to use RAID-6, make sure your access pattern isn’t too heavily slanted towards random writes. If so, you’ll get killed in performance.
Good luck!
TT
May 11th, 2006 at 12:02 pm
Tom,
We aren’t in an FC environment right now and are trying to get into a product that will give us the best performance for DB applications. We have looked at several products and the HP EVA4000 is one we are looking into. What products would you recommend that would be comparable in the SAS / SATA arena?
Thanks Tom!
C
June 26th, 2006 at 11:10 am
Question for Casey:
In your real world experience, are you finding that the failure rates of FATA drives are much higher than FC drives, and is the failure rate usage profile dependant? Thanks.
-Tom O’Toole
June 27th, 2006 at 8:50 am
By the way, the EVA arrays in which FATA drives are used don’t support RAID-6. The redundancy levels are VRAID0, VRAID1, and VRAID5. VRAID1 basically being a RAID10 implementation. The EVAs virtualization has an implementation that’s probably not directly comparable to others, but provides similar failure protection (i.e., VRAID5 = RAID-5 etc….). Since data from each LUN is spread across all disks in a pool, there will be data with different raid levels on the same device.