Comparing the new NetApp AFF8000 series with EMC XtremIO 4.0

NetApp has just announced the AFF (All-Flash FAS) series of flash optimised arrays which has been tuned exclusively for Flash (HDDs are not supported). Included with the new models is an updated version of Clustered Data ONTAP (8.3.1) which adds Inline Compression with near-zero impact on system performance. At the same time EMC has announced XtremIO 4.0, this is a key release as it adds double disk protection, non-disruptive cluster expansion and native replication (with support for RecoverPoint).

So how do these products compare? In many ways you would assume that this would be an easy win for XtremIO as it is a “ground-up” flash design, what is interesting is that in many ways the products are remarkably similar – the All-Flash Array market is really hotting up so let’s explore these products in more detail:

Performance

Comparing the performance of the arrays is difficult as you really need to look at specific workloads and get the vendor to model them and also clarify the exact conditions that were used (i.e. for NetApp is the array using Always on De-dupe and is it a real world solution that has been used for some time – clean systems will always give higher numbers).

Based on specification sheet numbers XtremIO is rated at 150K random IOPS (70% Reads, 8K blocks @0.5 ms avg latency),  whereas an AFF8060 is rated at around 160K random IOPS (80% Writes, 12K blocks @~1ms latency).

So both provide extreme performance, but amazingly the AFF is substantially faster with a write heavy workload – if the XtremIO solution was tested with a VDI workload (80% writes and 12K blocks) its IOPS would drop significantly.

Bottom line – both will massively outperform any non-All-Flash Array

Storage Efficiency

Both products have the same core capabilities (Compression, De-duplication, Zero Write Elimination, Thin Provisioning and Cloning), but XtremIO has several advantages:

  1. Inline rather than post-process de-duplication
  2. Global rather than volume-based de-duplication
  3. 100% metadata caching

AFF needs inline global de-duplication ASAP – I am sure NetApp are working on it as it already exists in FlashRay and adding more RAM to the AFF would remove any metadata caching disadvantages.

How much difference the metadata caching makes in the real-world is going to be hard to quantify, but as the AFF is an All-Flash Array it does mean the metadata that is not in RAM will be read from ultra-fast SSDs and the AFF8080 has 256 GB of RAM so clearly you could have a pretty large amount of the metadata cached.

Do all workloads need to have their metadata 100% cached? – I think not, and this is why it is more difficult to have a scale-up XtremIO system as at some point you would need more cache. Ideally I would like to see both products giving you the ability to pin the metadata for specific volumes in RAM, but with the option for the less performance sensitive volumes to have just a percentage of their metadata cached.

Block size on AFF is 4K and XtremIO is now 8K (it was 4K in the initial release) so you would expect AFF to get better de-duplication ratios, but this does come at the expense of more metadata and also it uses volume rather than global de-duplication which is likely to be less efficient.

Bottom line – both are good, but XtremIO has several architectural advantages

Availability

Both products have double disk protection with algorithms that are designed to provide high-speed writes (unlike traditional RAID 6). XtremIO does not have dedicated hot-spares, instead it uses hot-space – AFF needs to move away from dedicated hot-spares ASAP (NetApp’s E-Series already can with the Dynamic Disk Pools features).

In the very unlikely event of the loss of a pair of XtremIO controllers this would result in the entire cluster going down – this is not the case with AFF.

Bottom line – both are good, with each having small advantages over the other

Integrated Data Protection

AFF has a mature and proven set of capabilities (including snapshot replication and vaulting, and application consistent replica management).

Bottom line – no contest here, as XtremIO has to rely on external data protection solutions

Flexibility

This is where the AFF really comes into its own:

  • Scale-up as well as scale-out
  • Non-disruptively upgrade an array or migrate to a new array
  • Select from four different controller models
  • Build mixed clusters using AFFs and hybrid arrays
  • Eliminate silos as a single cluster can support
    • Server Virtualisation, Virtual Desktops, Databases and File storage
  • Support for
    • SAN and NAS protocols
    • VMware Virtual Volumes
    • Hyper-V with SMB 3.0

Bottom line – no contest here, as AFF is far more flexible

Conclusion

EMC XtremIO is a great All-Flash Array, built to deliver extreme performance and efficiency. The new NetApp AFF8000 series does a good job of getting close to matching these capabilities, but really comes into its own when you look at the broader set of use cases it is able to support and its greater flexibility.

It looks to me like NetApp and EMC are well placed to dominate the All-Flash Array market moving forward – do you agree or do you think there are others who are equally well positioned?

Related Posts

  1. Introducing EMC’s all-flash array – XtremIO
  2. Is there another storage platform as feature rich as NetApp FAS?
  3. VMware VVOLs on NetApp FAS is now available to deploy
  4. Should NetApp “kill off” FlashRay?
  5. Affordability means that 2015 will be the year that the All-Flash Array (AFA) goes mainstream
  6. Does the All-Flash Array really make sense?
Mark Burgess has worked in IT since 1984, starting as a programmer on DEC VAX systems, then moving into PC software development using Clipper and FoxPro. From here he moved into network administration using Novell NetWare, which kicked-off his interest in storage. In 1999 he co-founded SNS, a consultancy firm initially focused on Novell technologies, but overtime Virtualisation and Storage. Mark writes a popular blog and is a frequent contributor to Twitter and other popular Virtualisation and Storage blog sites.
twittergoogle_pluslinkedinmailtwittergoogle_pluslinkedinmailby feather

About Mark Burgess

Mark Burgess has worked in IT since 1984, starting as a programmer on DEC VAX systems, then moving into PC software development using Clipper and FoxPro. From here he moved into network administration using Novell NetWare, which kicked-off his interest in storage. In 1999 he co-founded SNS, a consultancy firm initially focused on Novell technologies, but overtime Virtualisation and Storage. Mark writes a popular blog and is a frequent contributor to Twitter and other popular Virtualisation and Storage blog sites.

7 thoughts on “Comparing the new NetApp AFF8000 series with EMC XtremIO 4.0

    • Hi Dan,

      Thanks for the comment.

      ADP does not change the fact that FAS/AFF is not able to spread the hot-space over multiple drives and therefore it will have idle drives/partitions.

      ADP is just away of sharing the Root Partition with data, parity and spare drives.

      If you look at XtremIO and DDP on E-Series they have no concept of a hot-spare drives (partitioned or dedicated) they just spread the capacity across all the drives. If there is a drive failure the hot-space is then used to reconstruct the lost data which will be much faster than with FAS or any conventional RAID solution (i.e. VNX).

      Best regards
      Mark

  1. Great post, thank you. Just wanna add two major facts about AFF.
    You can seemlessly grow by adding SSD shelfs only, not the case with XtremeIO where you need to add full bricks.
    AFF is supporting cloning on every level – Volume, LUN, file *and* sub-lun which means basically you can clone files on top of luns (e.g VMFS). The cloning becomes very handy for virtualized enviroments – copy your VDI’s instantenously with immediate space savings – and without peaks anywhere. Another benefit, it provides very easy restores too.

    • Hi,

      Thanks for the comment.

      As you point out today you cannot clone below the LUN level with XtremIO – it will be interesting to see if this changes when XtremIO support VVOLs or if a VVOL will map directly onto a LUN.

      Regards
      Mark

    • 2 chr1g1
      Have you notice that in cDOT 8.2 we have major improvement with RAID-DP, so now we can add in to an Aggregate 2 times less drives then average RAID group?

      For example if we have an Aggregate with 2 RAID-DP groups each 14+2 drives, we can add to the Aggregate:
      a) 2/4/6/8/10/12 drives to each existing RAID group
      b) 6+2 drives into new RAID group

      “All RAID groups in an aggregate should have a similar number of disks. The RAID groups do not have to be exactly the same size, but you should avoid having any RAID group that is less than one half the size of other RAID groups in the same aggregate when possible” Page 93.

      Clustered Data ONTAP® 8.2
      Physical Storage Management Guide
      https://library.netapp.com/ecm/ecm_download_file/ECMP1368404

  2. Hi. I think hot-space vs hot-spare is good idea in case of traditional HDD. But in case of SSD, where we have limited number of re-writes, we can get in trouble, as at time when we will need this hot-space, entire SSD drive might not be usable.
    So I logically suggest dedicated SSD drive is better from protection perspective.

  3. Pingback: Netapp Tape Library | oldmanhealth.com

Leave a Reply