Choices, SATA and a touch of DMX-4

By | July 25, 2007

Choice is good right?  A core principle of a free society and played an important role in fostering innovation, prosperity and a ton of other good stuff.  And of course we never make the wrong choices do we 😉

Today Im thinking about two things that are at least loosely related –

  • Lack of support for external storage on the DMX
  • Lack of support for ATA disks within a USP

Talk to a Hitachi guy and he will tell you how dangling the carrot of external storage (behind a USP or NSC) in front of potential customers has helped steal many a customer away from the competition.

Talk to an EMC’er and they will tell you how the DMX is wiping the floor with the USP because of its ability to “tier in-the-box” (thanks to the ability to mix and match a wide range of drives within the DMX, oh and lots of them).

In reality each solution has its place, with neither an out-and-out winner over the other.  Like Copy On Write snapshots versus full-block copies – there are cases where one fits better than the other – but neither wins every time.

So my point is this – If EMC are haemorrhaging customers to Hitachi because externalising storage makes sense to some customers, and vice versa, if the USP is losing ground on the DMX because some folks like putting ATA inside their tier 1 array, why not bite the bullet, build it into your kit, and keep hold of your customers.

There is certainly a demand for both, and demand has sparked the odd u-turn or two in the past – EMC adding RAID6 functionality being just one example.

When compared to something like Thin Provisioning, which both vendors are working on, implementing the above features would be a comparative walk in the park.

So if its not that hard to implement, and by doing so you potentially hang on to your customers, why not pinch your nose and take the plunge?

Lets take a closer look at each of the technologies and some of their pros and cons and find out what might be bothering each vendor (I promise not to quote any vendors or their blogs) –

Externally attached storage

Pros.

Allows you to attach lower performing ATA disks while mitigating many of the impacts they would otherwise have on your tier 1 array – like it or not, they bring baggage with them including –

  • longer rebuilds
  • supposedly more frequent rebuilds
  • slower destaging
  • ….. 

However, even when externalising storage behind a tier 1 array, you still have to assign some tier 1 resources such as ports, CPU and cache.

Enables existing lower tier arrays to take advantage of more reliable and scalable functionality found only in the tier 1 array.  Although this can be achieved by throwing ATA directly into a tier 1 frame, by implementing external storage you are able to extend this tier 1 functionality out to your existing kit – protecting and even maximising on existing investments.

Cons

Performance – There is no getting away from it.  If your data is destined for disk on an xternalised array it will take longer to put there and longer to get back.  Its journey will be –

FED-cache-FED-FED-cache-disk instead of FED-cache-disk. 

But if this is a problem for you, don’t do it.  A lot people I’ve seen are hanging SATA disks off the back of a USP, supporting applications where performance is not a huge requirement.

Complexity – similar to performance, by hanging one array off the back of another you are introducing more moving parts, increasing complexity, and increasing the chances of a component failure in the path.  The whole FED-cache-FED-FED…… thingy

Power consumption – In most scenarios it takes more energy to power two arrays than one.

People also ask the question of what happens to your data in the event of a lengthy power out.  One array cannot guarantee that the other has enough, if any, battery power to keep running, so cant safely destage cache to external disks………..  The thing is, with the kind of battery backup you get in a DMX and USP, coupled with the solid UPS + diesel generator type protection your data centre no doubt has, I even wonder if this is a real risk or just one of those made up ones that we use to sell stuff or slate our competitors?  Feel free to educate me.

Internal support for ATA on enterprise arrays

You just need to take a look at the Hitachi implementation of SATA in the AMS to see where Hitachi stands on ATA.  The AMS performs a verify-after-write operation on every write destined for SATA – end of story.  No option to turn this feature off.  Why?  I can think of only one reason, Hitachi does not trust ATA……. yet!

With this in mind, I wouldn’t recommend holding your breath waiting for the USP to support internal ATA – famous last words 😉

The thing is…………. I can’t help but wonder if Hitachi are still testing the water with ATA.  After all, ATA is relatively new and unknown.  They may well have über-techies queueing up and devouring ATA related logs, like kids with the latest Harry Potter adventure.  That’s my hope at least.

Also, I’ve talked about it before but Im gunna mention it again – there will be a world of difference between the reliability of a SATA disk in my laptop and a SATA disk in a purpose built storage array.  My laptop gets bashed, dropped, sprinkled with food and most worryingly, attacked by my 18 month old daughter.  This is a polar opposite to the treatment it will receive when installed in a decent storage array – controlled temperature, air-flow, humidity, vibration……  The storage array is to a disk what a mother’s womb is to an unborn child – the perfect place for it!

So once Hitachi are happy that they can predict the behaviour of ATA inside their kit, may be – HOPEFULLY –
we will see support included in the USP!

Pros

Tiering within a box.  It will often be cheaper and more energy efficient to plug fat and slow disks into your tier 1 array than to run two separate arrays.

Less complexity.  Less moving parts – FED-cache-disk instead of FED-cache-FED-FED-cache-disk.

More predictable in the event of a power out.  Not sure if this is really an issue??

Cons

ATA disks hinder the performance of your tier 1 array.  To continue the womb analogy, while a baby thrives in the womb, many women take a performance hit – tiredness, slowness… and dare I suggest swollen feet and stretch marks (bet you weren’t expecting that when you started reading Embarassed ).  The same can happen to your array – slow rebuilds, slow reads, slower destaging…..  And as far as Im aware, nobody in the storage industry has figured out how to avoid these. 

Interestingly though, not everyone buying a DMX or a USP these days is buying it for sheer feeds and speeds and would live quite happily with the performance hit.

Verdict: While neither will solve 3rd world poverty or global warming, neither is the cause either (well not a huge cause).  I’ve done a bit of external storage, it has worked well and I know of no horror stories.  And while I see that internal support for ATA in a tier1 array may not be the greatest thing for performance, I see how it can be useful.  Certainly nothing that would scare me off, and plenty to encourage me to implement both in my tier 1 arrays.

While I understand where each vendor may be coming from, I wonder if the risks outweigh losing customers?

May be Hitachi and EMC believe they are doing the right thing by removing the choice – therefore protecting us and our data.  After all, although making the right choice can reap great rewards, making the wrong choice can bring a world of hurt!

Hmmmmmmmm…..

Two final things –

Interestingly I see that the DMX-4 is supporting SATA II drives with a bolted on FC-ATA bridge.  Sounds cool and I like how this allows some of the ATA baggage to be offloaded from the DMX to the drive itself.  I’ll be interested to know how much of a real world impact this has.  Oh and what kind of a difference it will make to the price of these drives, SATA disks supposedly being cheap and cheerful.

And finally…. Im not quite sure where I stand on the energy efficiency side of ATA that people are banging on about……  I don’t profess to be an expert and haven’t the time to look into the figures, but surely there is something in the following – SATA disks are slower than FC so they will need to work for longer to complete any given request.  Working longer = consuming more power.  And that’s not even mentioning more frequent and elongated rebuilds (working and consuming power again).  So in reality, is there much of a difference?  Personally I doubt its worth making a song or dance about.

Nigel

13 thoughts on “Choices, SATA and a touch of DMX-4

  1. open systems storage guy

    Sometimes SATA is the best choice- even for an enterprise array. If your enterprise array is hosting storage level copies, backup servers, file servers, or any other workloads that work fine on SATA, it makes sense to take advantage of the low cost per GB of the disks when possible. Of course, it would be cheaper to have a secondary (maybe less available) array that would only do your SATA storage, but that could not serve as a target for storage level copies unless you were to attach it to a virtualization device like the Hitachi controllers or an SVC from IBM.

  2. Ced

    I’m not an expert. Correct me if i’m wrong but an enterprise array never spins down his disks and i’ve got the impression that spinning process sucks more power than moving the arm to read/write the data. So if a SATA drive has a 7200 rpm compared to a 10k or 15k rpm from a FC drive, they should be more energy efficient ? Last but not least, have you any ideas of the average idle/active proportion of disks in a enterprise array ? Ced

  3. Nigel (mackem)

    Ced,

    Thanks for your comments.

    You are of course correct that enterprise arrays do not spin down disks when not in use.  In fact many of them use this free time to sweep/scrub disks.  Re your point that its the spinning and not the actuators that sap the power, you are also correct – when idle (no actuator movement, just spinning) a disk will consume approx 3/4 of what it does when seeking (spinning and moving the heads).  So your point is valid. 

    In my example I was being somewhat flippant, but this is what I was getting at –

    I checked the "typical watts" for an Seagate Cheetah 15K 300GBFC disk and a Seagate Barracuda ES (SATA) and they are as follows –

    Cheetah – 18.8W
    Barracuda – 13W

    However, if it takes the Barracuda twice as long to service each request (7200RPM versus 15000RPM) is there a real advantage.

    Of course this is a way, way oversimplified example and does not take into account that a lot of the time these disks, especially Barracuda, will be idle.

    As for your question re average disk idle/active on enterprise arrays, i have no idea, although my interest is now sparked.  I have worked on quite a few enterprise arrays that sit around doing almost nothing, especially the disks.  Ive always measured this with IOPS and MB/Sec though and would have to scratch my head and see if I knew of a tool that would show idle and active time.  I also imagine anything like that would have to be split into drive types as well as you would expect (hope) that SATA would spend a lot more time idle….

  4. TimC

    I would wager just the opposite is true, and that FC is far, far more power hungry for the reason of usage.  When people implement FC, it’s typically for *high performance*, which to me says they’re going to be beating the crap out of the disks, and it’s something that CANNOT be down (because it’s constantly in use).  On the other hand, sata, in my experience, is more often than not simply a dump where files get put, and are accessed as needed, but most definitely not a constant barrage. 

  5. Ced

    Hi Nigel,

    If we look at your figure, The 'FC' version needs 44% more power. So i agree with  you for the time it need to read/write a block , the SATA is not a real advantage but as i mentioned, the ratio active/idle is important and in this case you could win a lot of energy by using the SATA drive. This pure theory as i would never use SATA on high end arrays  for real prod environment except for VTL.

    To rebound on your article, why using a very expensive EMC box for SATA Drive (except for highly specialised application like VTL) ? i embrass the HDS theory where you install a midrange storage (AMS1000) behind with SATA drive and still get the benefit of lots of cache from the USP-V.

    Just my 50cents.

    Cedric

  6. Nigel (mackem)

    Fair point Tim and I think I somewhat answered your point in my previous comment to Ced when I said "…you would expect (hope) that SATA would spend a lot more time idle…."

    However, what you are saying is not that FC is way way more power hungry than SATA, but rather because we beat the crap out of them and demand more from them causing them to consume more power.  Whereas SATA uses less power partly because we only use them for menial tasks.

    With that in mind, one cannot really call their array much greener than anybody else’s without relegating it to menial tasks.  By saying “our array sucks less power because it can be filled with SATA” is basically saying “our array will suck less power if you only use it for menial stuff”.  A DMX-4 is certainly not for menial stuff.  That said, I like the fact that it can take SATA internally.

  7. Richard

    Interesting piece. I heard a whisper from a very good source that Hitachi are looking to release a SATA disk for the USP-V before the end of the year. Not sure why they are not out there blowing their trumpet if this is true.
    It would for EMC’s hand on the external storage, or leave the HDS stable with a real advantage – being able to offer teh best of both worlds!

  8. Nigel (mackem)

    Richard,

    I hope your sources are correct.  It seems a no brainer to me…….. but then again I have an awful lot of restpect for the guys at the "factory" as they really really know their stuff.

    Also I find it interesting that none of the EMC crowd have chimed in on this one – slamming the notion of external storage.  I wonder if they are looking at implementing external storage in their kit??  Pure speculation of course 😉

    Guess we’ll find out sooner or later

  9. tim1

    I agree the sweetspot for SATA drives in tier 1 arrays is VTL/snapshots; I would imagine the price per GB would be around 1/4 (using 750GB/7.2K) compared to 300GB/10k FC drives.

    On availability in USP-V, I've heard rumours from HDS sales that this was on the roadmap for the last year or so but as RSD (aka the factory) don't publish roadmaps externally it's difficult to tell whether this is real or just sales banter.

    That said, let's hope the USP implementation (if it happens) uses FATA drives, rather than follow the AMS implementation of using SATA drives with one SATA-FC bridge chip for every 14 drives (i.e. slow as a donkey with 3 broken legs).

  10. the storage anarchist (barry burke)

    One important clarification: I don’t think EMC ever actually claimed that the Symmetrix DMX-3 and DMX-4 power advantage was based solely on SATA, nor on comparisons to competitors’ non-SATA configurations. Fact is, both the DMX-3 and the DMX-4 require less power to support the identical configuration of drives, ports and memory vs. the competition. The DMXs also support more drives behind the more efficient logic infrastructure, for greater efficiency advantage in configurations requiring more than 1024/1152 drives.
    Adding SATA drives into the mix does indeed EXTEND the DMX-3&4 power advantage (to as much as 70% less power on a full-up SATA vs. FC comparison), but this admittedly would be a rare system – unless, of course, you consider the DL6000 disk library (full-up with 2400 500GB LC-FC drives). 
    And obviously, SATA drives are indeed more appropriate for less I/O-intensive applications than enterprise-class FC drives – hmm…like backup to disk, for example. And of course, the storage controller software has to be "SATA aware" in order to manage, insulate and optimize the performance differences between different classes of drives (and RAID types, for that matter). That’s what Enginuity 5772 brings to the table.

  11. RacaSAN

    SATA in DMX struck me as  "RAID 6 – the competion has it so we need to cover this too – even if it's not getting used too often" type of thing.

    In a high end system I wouldn't put it in much more than a snapshot. SATA drives are designed for a duty cycle of about 30% max. FC/SCSI are designed for 80%+.

    Expect bad things if you use your SATA beyond this => thus the reason for the existence of RAID6 ? – well that and the ever higher capacity meeting the laws of probability – when are those 1TB drives due (on 4 platters Mr Hitachi)

  12. Jayakumar

    Hi,
    Send me SAN related documents, hardware material description, configuration installation implementation documents video libraries etc.,
    Thanks & Regards,
    Jayakumar.D
    Cell No:+91-9930281590

  13. Helen

    I wanted to make a ccrerotion here. HP didn’t have an OEM relationship with EMC ever. We were a Symm reseller. I was very familiar with that agreement and would love to say a lot more about it but I can not. When all you have is a hammer, every problem looks like a nail and Symmetrix was the only hammer EMC had at the time. Suffice to say, I’d fall out of that bed 20 times a night.I was also part of the team that parterned with Hitachi Japan (not HDS as many like to say) as an OEM and joint development agreement. That relationship has been in place close to 10 years and is as strong as ever.Zilla last I checked, HP’s PC, laptop, and server hardware is doing pretty well without EMC’s business. Whst PC’s and servers will you be sourcing if DELL ends their agreement with EMC?I also remember EMC doing every thing they could to stay close to our UNIX business. My only regret is that our UNIX server team didn’t stop working with EMC as quickly as EMC ran from HP.

Leave a Reply

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


*

You can add images to your comment by clicking here.