SATA Patter

By | October 24, 2006

OK, so the above title only works if you pronounce SATA as “sata” and not as “sata”  – may be that would work better with sound Wink

It seems that SATA is causing quite a stir in the storage world.  I recently read a post on a forum written by a fellow storage pro with a client who wanted their tier 1 storage on SATA disk – presumably due to the good cost per gigabyte.  The reaction to his customer was the same as what mine would previously have been – to protest and try to persuade his customer away from this heresy.

After reading the post I started to think that this kind of thing is going to become more and more common.  After all, the people who hold the purse strings in many organisations inevitably prefer the cheapest option when it comes to spending money on IT.  Then a second thought came to mind – why do we automatically react with horror at the thought of using SATA disk as tier 1 storage?  Have we been conditioned by the industry to react this way?

Of course I know that certain applications and situations require the higher performing and more reliable FC disks, however, Im starting to think that SATA could be a viable option for more than just archiving.  Now bear with me while I explain why I think this –  

A short while ago while I was contracting for a company in an outsourced environment who needed to add more capacity to one of their customers storage subsystems.  As I was assessing the requirements and putting together my recommendations I came across a slight catch 22 situation –

•    On the one hand the company had a standard where they only put 73GB 15K spindles into this subsystem.

•    On the other hand they were running out of free disk slots in their existing frames and would require an additional frame installing to meet their capacity needs with 73GB spindles.  However, they also had serious space, heat and power consumption problems in the primary data centre which meant they could not install a new frame for quite some time.  This was not an option – the customer needed the additional capacity “yesterday” – isn’t that always the case!

Anyway I looked at the performance stats for the currently installed disks/array groups and found that all of them, as well as the front end channel ports and cache, were sitting around doing almost nothing other than consume electricity heat the room.  So my natural reaction was to recommend the larger 146GB 10K spindles so that we could install twice as much capacity in the limited amount of physical slots available and not have to worry about performance, and all at a cheaper cost. This wasn’t a problem to get management buy-in.  However, because the cache was coping so well – the backend disks were not even a factor – I actually started to wonder whether SATA disk in a RAID 6 dual parity configuration would be a realistic option??

As it turned out for a couple of technical reasons this wasn’t a serious option in this environment.  But it was enough to make me to start thinking that may be we shouldn’t be dismissing SATA disk so quickly.

MAKING THE CASE –

So what really is the “crack” with SATA disks?  Are they as bad as we (or may be its just me) think?

While I do hear that SATA disks are not built with the same quality materials or to as high a standard as FC disks, I am certain that they are not manufactured with the intention to fail quickly.  I must point out that I can’t comment authoritatively on the disk manufacturing process as Ive never actually seen it, but anybody reading this who owns a disk manufacturing plant is more than welcome to invite the bloggers of rupturemonkey.com to come and view the process.  I think its safe to assume a relatively high build quality for SATA disks designed to be installed in enterprise storage arrays.  Disk manufacturers reputations are on the line here!

I also think that we are often quick to associate SATA disks with cheap commodity desktop computers and forget that we are installing them in high end storage subsystems that are designed to be as disk drive friendly as possible with built in mechanisms designed to extend the life of a disk.  They also won’t be turned on and off all the time, which can cause wear and tear on disk drives.  Then of course they are often installed in specially designed “clean rooms”.  All a far cry from the computer that sits on the company receptionists desk, or your laptop, that gathers dust, gets bumped around and suffers the occasional coffee spill.

Of course SATA is not a replacement for FC disks, but I’ve seen many installations where FC disks are overkill.  There seems to be an ethos that because a subsystem is expensive the disks installed must by default have to be top of the range.  Not always true!  Often true, but not always.

Mackem

Lets also not forget the imminent arrival of a replacement for traditional RAID from a new startup company formed by the bloggers of rupturedmonkey.com  This new technology will reduce the overhead caused by parity and double parity calculations on writes, as well as speed up and smooth out the procedure for recovering from failed disks.  See previous post titled “RAID – Time for a better way

7 thoughts on “SATA Patter

  1. c2olen

    Adding another plus to the SATA drives. When installed in a subsystem, they don’t have to go through a firmware upgrade when upgrading a subsystems microcode. The disk i/o doesn’t need to be suspended for these disks, as it is the case on some subsystems from the big-blue type vendors.

    Although the failure rate of SATA disk drives in our shop ( 1 disk out of 200 fails each week ) proves me wrong, i could also imagine a couple of installations where SATA would suffice.
    The current performance requirements of our shop exceeds SATA capabilities. But for small and medium sized companies, why not?

    Damn, i could swear i had my blueprint for the NFDA here somewhere? How can i replace the RAID technology when i can’t find my blueprints?

  2. snig

    We are using SATA in our shop and have not had a disk failure since we installed them back in March. Roughly 13 TB of the stuff.

    I have file servers, archive data, and even some development databases on the SATA drives. They are running fine and was haven’t seen any performance issues with them at all.

    If at all possible I think users should give them a try before they discount them altogether. I think they’ll find them to be better than what some of the vendors would have them believe.

  3. Liho

    We’re using 30 x 250GB SATA disks in 9570V. No any failure since installation (~ 18 months ago).

  4. c2olen

    Had to look up the word patter. Had no idea what it was supposed to mean in this context. Wikipedia helped me out.
    Although we (the Dutch) can speak/read/write the English language quite well, some words just don’t ring a bell. Patter doesn’t seem to be used often, or am i mistaken on this?

    On the SATA patter comments: I guess we had a bad batch then;-)
    I exaggerated a bit, but from time to time, the failure rate is quite high, and 1 disk a week isn’t uncommon. On many occasions a re-seat of the drive does the trick. So, it isn’t always a real diskfailure. Nevertheless a rebuild of the array is at hand.
    According to the vendor, this is a firmware issue. We’ve done some upgrades in the last months, and the spindels seem to keep running longer now.

  5. mackem

    I think that patter is very much an English term rather than American – although I may be wrong on that. May be its even from my local area in England… who knows 😉

    Re the possibility of firmware being an issue – a while ago I was working with some XP12K arrays and we had lots of failed FC drives which caused a lot of concern for such expensive disks (Seagate disks). After a while it we were told that this was a firmware issue on not the physical disks.

    Also before that I worked with some HP EVA boxes that went through a phase having loads FC drive failures. We were told by our supplier that these were from a known bad batch.

    So it goes to show that even the more expensive and thoroughly tested FC drives can also be unreliable. Also if firmware can be the problem then no matter how well made your disk is, it can look like its failing just because somebody made a mistake when writing the firmware.

  6. Jesse

    Ok, I went to Powerlink (as an EMC freak) to try and look up the MTBF on the SATA vs. the SCSI and the FC drives, only to find powerlink down.

    The biggest concern i can see is that SATA disks are single ported, where as fibrechannel disks are dual. This means that only one storage processor (whatever brand) can have control of the disk at a time. It also means that in failover scenarios data is forced to cross busses in one form or another to reach the port. Whether that be through a hardware/software failover within the enclosure or through the SP, it can mean not as clean a failover.

    As for performance, I’ve found with the 18TB of SATA disks I’ve got running on my Clariion (For Veritas Disk-Cache) as long as I keep the disks within a raid-group on the same storage processor, I’m not noticing any bottlenecks. (Before that, when I took the default of “auto-assign” I was noticing considerable queueing and “wait for disk” on the backup host during the peak backup window, with about 200 jobs running.)

    I’m running 6+1 Raid5 across the board and feel that if the back-end stripe is wide enough the pre-positioning will allow it to fly through writes. Obviously I’d have preferred to go Raid1+0 but couldn’t get the budget for the 90+ drives that I would have needed to make that happen.

  7. snig

    I agree that keeping the RAID groups on a single controller helps performance a great bit. e.g. RAID Group 0 on controller 0 and RAID Group 1 on controller 1. In fact that’s the way HDS will tell you to implement their controllers.

Leave a Reply

Your email address will not be published. Required fields are marked *


*

You can add images to your comment by clicking here.