Mark III Systems Blog

NEW SVC DH8 Nodes (as if the CG8 nodes weren’t enough)


1 million IOPS?

File IBM's announcement of the upcoming SAN Volume Controller DH8 generation of nodes under the category of "more horsepower than you'll ever possibly need" for traditional I/O, which our clients have found is a really nice insurance policy within the realm of the end-user user experience (with a wide variety of different storage system backends).

As in past releases, one of the nice things of piggybacking on Intel's roadmap is that SVC is able to leverage processor enhancements by Intel and turn them into substantial performance improvements.  The May announcement of SVC v7.3 and the new DH8 nodes is no different with the DH8 nodes featuring the latest Ivy Bridge v2 processors and giving users the new option of including up to 2 Intel QuickAssist chips for compression.

Additionally, there are a bunch of other announcements out there on some of the functionality enhancements and changes in SVC form factor (now 2U, but without the usual UPS), among other things, so I'm going to focus my thoughts below primarily on performance improvements.  To a certain extent, this is a conversation in the weeds about details don't really matter, but it sort of illustrates a point about how performance is never and really never will be a concern with SVC, if very reasonable planning is done.

The following are some performance numbers for a single SVC node pair (I/O Group) from last week's IBM Technical Edge conference.  These numbers were all run with an IBM FlashSystems backend, in an effort to see where the SVC nodes themselves maxed out at (i.e. the disk backend was not the limiting factor).  In short, the DH8 announcement takes the CG8 node's really high maximum and pushes it to an unprecedented level, especially considering you can build an 8-node SVC cluster today with virtually any combination of SVC node generations.


[caption id="" align="aligncenter" width="414"] SVC DH8 vs. CG8 Performance Comparison[/caption]


What's the net of this post?

First, performance with SVC is not and will not be a concern, if you responsibly plan ahead of time.  Even if you don't, these huge performance maximums give you an incredible margin of error.

Second, the substantial horsepower in the DH8 nodes (in combination with the Intel-based compression accelerator) makes it fairly easy for anyone to adopt Real-time Compression without much thought.

Third, the DH8 nodes provide a certain degree of investment protection down the road, in case IBM is planning on releasing any new game-changing type features with the SVC (similar to that of the level of Real-time Compression).  If you invest in the DH8 now, I would feel pretty good that you should be able to accommodate any new features that may come down the road for at least 5 years.