EMC Symmetrix was introduced in 1990 (as a Mainframe storage) and it has quickly become one of the most successful enterprise Tier1 storage ever (Now the VMAX). It’s big (now supporting up to 3200 disks and 2048GB of cache) but its architecture is outdated and so are its features. What was good enough to sustain Mainframe and large Unix server workloads is no longer when having to manage thousands (or tens of thousands) of VMs.
More and more… and more
It’s not only the number of VMs (and consequently, of the physical servers) it’s much more than that.
In a proper high demanding cloud environment, requests for resources rise and fall continuously: VMs get started, stopped, cloned, zapped away at any time depending on the needs of the moment (e.g. Ephemeral instances). Efficient and strong APIs, ease of use, rapid provisioning/deprovisiong and integration with any kind of Cloud platform is fundamental.
Strong QoS has to be implemented at the core of the system, the problem of “the noisy neighbor” has to be managed carefully when the environment gets very large and the workloads are totally different from each other. Even in this case, ease of use, efficiency and configurability of the QoS mechanism has to be rock solid.
Scalability is another issue. It’s not only about total capacity but also about IOPS and, even more important today, predictable and low latency! Scale-up design has its limits in the maximum IOPS or disks in a box and, every time you add a new box, migrations and reconfigurations lead to a potential service disruption.
Availability and resiliency are still priority in any storage request, even more in complex tier 1 cloud environments, but you know… is RAID enough?
Then there are other important features, like multi protocol support (I think more for legacy compatibility than for the future), replication (SRDF has done its time, hasn’t it?), and so on… the list is very long.
VMAX clearly doesn’t have the necessary characteristics to run a private cloud of substantial size and it’s very costly when it comes time to configure it with Flash. I’m not thinking only about TCA, it’s the TCO that scares me the most!
Thinking about the alternative
After Storage field day 5 and the clear (natural) shift in Solidfire’s strategy, now more focused on the private cloud enterprise than in the past,
I think that Solidfire Storage System could be considered the VMAX of the Cloud era. You can find an interesting comparative presentation between different AFA architectures at the bottom of this post.
When compared to other solutions, even when you think about EMC XtremIO, Solidfire is faster, more reliable and much more scalable than the competition. Its set of features is designed from the ground up to solve problems at scale without compromises in terms of performance or availability.
Why it matters
Solidfire really looks like a next generation storage enterprise for the cloud era. It’s confirmed as a really interesting product and positioned at the high end, as is VMAX.
In fact, while I can compare Solidfire to VMAX, I can’t do the same with other AFA vendors. For example, I find many more similarities between Pure storage and Clariion (VNX2 now), but not the VMAX… and, in this case, I could say that Pure looks very like the next midrange storage enterprise for the cloud era (it even looks similar in the form factor, 😉 ).
What About XtremIO? Well, I’ve just recently written about it. It’s immature and at the moment and it looks neither like the next VMAX nor the next VNX. But you know, we are at the beginning of its development and EMC is consistently improving it.
Disclaimer: I was invited to this meeting by GestaltIT and they paid for travel and accommodation, I have not been compensated for my time and am not obliged to blog. Furthermore, the content is not reviewed, approved or published by any other person than the Juku team.
Great overview video. The architecture summary slide is @1:55 is a little disingenuous for the all flash arrays i.e They DON’T have a Full Enterprise Feature Set, saying these features are on the roadmap is should be a given, but how far out are they and how robust are they likely to be ? .
Some certainly have more of a full feature set than others right now 🙂
That said, the segmentation was really meant to point out that the hardware-centric flash approaches are unlikely to ever have a full feature set, given the massive investment in software required.