IBM’s New Midrange with Easy Tier & External Virtualization

Yes, IBM has announced a new midrange virtualized disk system, the Storwize V7000. A veritable CLARiiON-killer : )

The announce detail should be well covered elsewhere (on www.ibm.com/storage and by Barry Whyte & Tony Pearson for example, if not yet then very shortly) so what you get here is mainly my view on the two major value points of Storwize V7000:

  1. Virtualization of external disk systems
  2. Easy Tier (automated sub-lun data relocation)

I have previously blogged about Easy Tier on the DS8000. It’s great to see IBM cross-pollinating the product set with features – it looks like someone in development has done a great job of getting the various development teams together to share code. What we see with both Storwize V7000 and SVC 6.1 are products built from a code base of SVC 5.1, with a user interface inherited from XIV, and Easy Tier and a robust RAID implementation inherited from DS8000. EMC’s Mark Zilla Twomey has in the past referred to IBM’s separate ivory towers of product development, well Storwize V7000 and SVC6.1 demonstrate that there is actually a very collegial approach taken to product development within IBM and there are no ivory towers.

Key components of Storwize V7000 are:

  • Green shades denote SVC 5.1 software that forms a common base for both SVC 6.1 and Storwize V7000.
  • Blue denotes existing software from the DS8000 DA adapter that has been included in both SVC 6.1 and Storwize V7000
  • Pink denotes new components in SVC 6.1 and Storwize V7000 (management of hardware enclosure and PSU applies to Storwize V7000 only).
  • Note that because they use the same software stack, Storwize V7000 has essentially the same interop list as SVC

The external virtualization story is very tidy i.e. you can effectively use your Storwize V7000 disk system as an SVC. This pretty much covers the simpler, more cost-effective virtualization for smaller customers that I called for in my blog post “One size does not fit all“.

The other big feature of Storwize V7000 is Easy Tier. Easy Tier is about extracting value from SSDs. Some folks think that SSDs are going to replace all other storage in the next few years, but the science doesn’t support that, and looking out 5 years it’s clear that spinning disks will continue to provide the bulk of enterprise disk capacity, with SATA (or more likely it’s SAS-fronted sibling, often referred to as ‘Nearline SAS’) leading the charge in reducing cost per TB.

So the challenge is how to extract maximum value from these expensive little puppies we call SSDs, plus take advantage of the lower cost per TB of large SAS and NL-SAS/SATA drives without compromising performance.

Automatic tiering that is easy and cost-effective for the midrange is the killer app that Storwize V7000 delivers on. There’s not much value in a tiering system that is complex or requires multiple elements of charegable add-on software. IBM’s Easy Tier will tune your system, automagically migrating 256MiB extents to and from SSD as required (from SAS or Nearline SAS/SATA) giving preference to small random reads (the workload most suited to SSDs).

Let me summarise from the Easy Tier Redpiece:

I/O operation Typical SSD relative to HDD
Random Read 100 times faster
Random Write 40 times faster
Sequential Read or Write 2 times faster

Easy Tier will only move random I/Os to SSD. For every extent, Easy Tier will look at the following:

  • I/O size, and whether it is a read or a write and random or sequential
  • I/O rate and latency

Easy Tier will calculate the ‘temperature’ of each extent and will migrate the hot extents from the HDD to the SSD, based on highest temperature first.

The extent relocation algorithms factor in the cost of moving an extent into the decision to relocate. Volumes can also be excluded from Easy Tier, so that several apps may be served from a storage pool and only your performance critical ones tagged to reap the gains from SSDs/Easy Tier.

Storwize V7000 has inherited thin provisioning from it’s SVC code-base, and thin provisioning on this architecture provides excellent performance and is fully supported in an Easy Tier environment – something some other vendors might struggle with.

Also, if you don’t have any SSDs, you can still turn Easy Tier on and it will give you some conservative estimates on what improvements it could make to your performance if you did add SSDs.

Storwize V7000’s Easy Tier function is smart, easy and free (of course you probably want to buy at least three SSDs).

I hope that Easy Tier and Storwize V7000 will become synonymous in the midrange. Other vendors may hold their SSD value-extraction systems back by charging large for them, or by making them complicated, or using them as relatively basic caches, but IBM has made it easy and affordable and smart, and I love it when someone makes smarter technology affordable so that it can be widely adopted.

Also worth noting is that the following features are included in the base Storwize V7000 product, with no additional licences required:

  1. Easy Tier
  2. High performance thin provisioning for tier1 apps with no performance caveats
  3. Local volume mirroring (e.g. between internal disk and an externally virtualised disk system)
  4. Flashcopy (snapshot+writable snaps+clones+instant restore of snapped volumes)

This I like, because it unlocks innovation – the features get deployed in real life rather than staying on the brochure as happens with some other vendors midrange disk systems.

Storwize V7000 delivers fantastic performance, function and ease of use, bringing to an end the easy run our competitors have had in the midrange space up to now. With EMC delaying CX5 until 2011, and EVA on the back foot after the Left Hand and 3PAR acquisitions, Storwize V7000 will be hard to beat.

About these ads

18 Responses

  1. Jim,
    good article,
    I’m still looking for some docs about v7000 easy tier, architecture design, and capabilities on the IBM site without success at the moment, :-(

    can you help me to understand better the V7000? here you are some questions:

    From your article it seems that V7000 writes data in the low tiers and then promote them up to the SSDs if needed, correct?
    is it a 3 tiers tiering or only two?
    the “promotion mechanism” is due to the nature of SSDs used? (MLC instead of SLC).

    The 256MB extent is very huge if you think about 8k DB blocks, can I tune this parameter?

    is the system wide striped or it is based on old separated raid groups ?

    Are snapshots (space efficient flash copy) pointer based or copy-on-write?

    I can’t find nothing about replicas. does the system support, at least, snapshot based replicas or async replicas?

    thanks a lot in advance,
    Enrico

  2. So many questions : ) It’s 11pm Thursday night in NZ, so here are my quickfire answers:
    – I will post some doc links over the next day or so
    – Easy Tier is a 2 tier approach per pool. I have conflicting info as to whether one of those tiers needs to be SSD, so assume it does for now.
    – High write cycle Enterprise-MLC SSDs
    – 256MiB default is based on default extent size. You can set extent size smaller if you want, but internal testing shows that 256MiB default works well for most workloads including OLTP database.
    – Each pool is usually striped across multiple RAID groups – the RAID implementation is taken from DS8000 code, and the striping is pure SVC. Pools were known as managed disk groups in the old terminology. RAID options are RAID1/10, RAID5 and RAID6.
    – Snaps are COW – same as SVC
    – Replication (Metro Mirror & Global Mirror) is the same as SVC. i.e. sync continuous, and async continuous. Snap-based replication is planned for a future release.

  3. Jim,
    thank you for the prompt answer.
    I’ll look forward for the docs.
    Enrico

  4. [...] the V7000 is all SVC. Much of the software is directly derived from SVC 5.1 (green stuff in IBM’s diagram), while [...]

  5. The only disappointment for me is no 10 GbE iSCSI so far.

    • If you can flood 4 x 1Gbps ports, then I wonder if iSCSI is really the right choice for that environment. And the reverse question, can vendors offering 10Gbps ports process the packets at line speed?

      • I agree 100%, was talking more from a marketing PoV rather than technical.

        It is like 120 disk mid-range systems with 4 or 8 x 8Gbps FC ports.

      • What a nonsense again, it really speaks volumes about your (in)experience with iSCSI… since when iSCSI is not limited by the actual physical connection? How is FC8 is better than, say, 2x 10GbE ports?

        Serving up dozens of hard-loaded VMs over 4x 1Gb ports is pretty idiotic idea, a rather dead one on this Storwize as it is now if IBM does not upgrade it to 10Gb quickly (I heard some rumors the chipset is there, it’s just validation that needs to be done before they enable it in a firmware upgrade?)

        FYI reaching 700-800MB/s *sustained* speed on iSCSI boxes like an Equallogic one is very common and makes administrators’ life very easy when they have to move data around or serve up stuff for NAS heads via iSCSI.

  6. Buddha, can you answer one question?

    How come that this products doesn’t even mention Storwize real-time compression?

    IBM bought Storwize to get access to their real-time compression algorithms, and the first storage product that has got Storwize name in it has absolutely nothing to do with the Storwize original product?

    Did IBM just like the name Storwize for marketing purposes or what?! :)

    • Yep – you got it. As best I can tell (it’s not entirely clear) IBM plans to use the Storwize brand in the same generic way that HP uses ‘StorageWorks’, so I expect to see future products using Storwize naming. For the full story on the use of the Storwize brand, see Steve Kenniston’s (Storage Alchemist) blog post.

  7. Hi Jim,

    Disappointing really, granularilty at 256MB default is huge. Compellent start at 2MB and move down from there, now that’s true tiering. Also snapshot COW is very old school, IBM need to catchup. The snapshot used by StorageTek with their Iceberg disk arrays, (oemed by IBM) were using better snapshot technology with log structured disk array file systems 20 years ago. Come on IBM you can do better than this!!.

  8. p.s. I use MB on purpose

  9. Gents,

    EMC is using 1GB slices to tier, so 256MB seems like a deal.

    All this functionality seems to hit performance, V7000 only achieves 56K SPC-1 IOPS

    http://www.storageperformance.org/benchmark_results_files/SPC-1/IBM/A00097_IBM_Storwize-V7000/a00097_IBM_Storwize-V7000_2-node_SPC1_executive-summary.pdf

    So what is the benefit unless the storage subsystem has been under sized from day one and the only option is to move data in and out of SSD ?

    Rgds
    John

    • You make two suggestions that seem strange to me.
      1) Performance: 56K IOPS is a lot for a midrange system by anyone’s count, and about the same per drive as the relatively low-function DS5300 so I’m not sure why you would suggest it isn’t much. Most midrange requirements I encounter in real life are around 10K IOPS.
      2) Under-sized: The idea is to deliberately use fewer spinning disks, and boost the performance with SSDs. It’s not a question of fixing an under-sizing, it’s a smarter sizing. It’s what caches have been doing for decades, only now we are adding a solid state read/write cache layer.

  10. “2.High performance thin provisioning for tier1 apps with no performance caveats”

    According to the IBM V7000 Redbook, Thin Provisioned Volumes are clearly not recommended for High Performance or write intensive workloads

    Thin-provisioned volumes require additional I/O operation to read and write
    metadata to the internal disks or to the back-end storage which will also add additional
    load to the IBM Storwize V7000 nodes. We therefore do not recommend the use of these
    volumes for high performance applications, or any workload with a high write I/O
    component.

    • Yes I saw that comment in the redbook recently. However, having co-authored two redbooks myself I know that there is a lot of leeway given for personal opinion in them – they are not quite the authoritative documents people sometimes mistake them for : )
      There is definitely a performance hit, and when thin provisioning (referred to at the time as space-effiicient vdisks) was originally released there were “try it and see” type caveats applied, but since then, experience in the field has shown that the overhead is not really noticeable in real life. To demonstrate how it performs in an OLTP environment, back in 2008 we ran an SPC-1 benchmark (274,997 SPC-1 IOPS) with thin provisioning turned on http://www.storageperformance.org/benchmark_results_files/a00072_IBM_SVC4.3_SPC1_executive-summary.pdf so that now we can advise people they are fine to run it in their high performance environments if they want.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

Join 817 other followers

%d bloggers like this: