vMax to the Max

I was honoured to participate in Storage Field Day 14 as a delegate and the first presentation up was from the Dell EMC High End storage team covering the state of the vMax, specifically looking at the All-Flash versions.

One of the first things that struck me in the lobby of the Dell EMC’s (very nice) new briefing center where they have a vMax 250F and an XtremIO cluster is just how small “big-iron” has become. The core of the vMax 250F is a quarter rack, upgradable to a half rack with two V-Bricks, scaling to 1PB with 64 front end ports (FC or iSCSI). Moving up to the 950F pushes the limits up to 4PB.

I don’t have a lot personal history with the vMax/Symmetrix line-up but the memories I do have are all about rows of racks rather that just part of a rack. The 950F only takes up 4 racks fully populated with 16Tb SSDs to get to 4PB. I remember a client having issues with the cabling constraints that required they all be side-by-side racks and their vMax was in a spot where they had filled in 4 side-by-side and there was no room to add racks so they had to rearrange their server room in order to upgrade the system. The good news is that if you need to spread out modern vMax systems, the V-Brick interconnect on the 950F (but not the 250F) will let you install them up to 25 meters apart.

The vMax’s role is all about solid, proven dependability and availability. Part of this is the integrated Active/Active array option over SRDF allowing us to build highly available metro-clusters (in VMware speak). When the conversation brought up the idea of using a vPlex for the Metro clustering, it was quickly pointed out that the vMax can easily saturate even the largest vPlex configuration so there’s really no point in going down that path.

Dimishing latency returns

That got me to thinking about the long term impact of this new generation of insanely performing All-Flash systems being used in Metro-cluster configurations for availability. One of the value propositions of this new generations of storage systems is the flat latency performance in and around a millisecond or maybe two for almost all transactions. Which is awesome, but if this is also coupled with a high availability requirement that involves synchronous replication over distances that systematically add 2-8ms to every write transaction I wonder when we start seeing diminishing returns on the IOPS/$ metrics. Our application availability architecture requirements are hobbling the overall performance capabilities of these new high end storage systems.

Granted, this type of configuration does provide very predictable performance where I can validate my application’s latency requirements against the replication link’s RTT and the storage system’s write latency as they have become fixed values for all practical purposes. Which may be the most important factor for storage admins (we’re not quite dead yet!) with being able to ensure predictable, consistent performance at all times.


I’d just like to point out to the FC purists out there that the biggest, baddest system from Dell EMC can be configured with 10GbE iSCSI, so yes, you can (and should) take iSCSI seriously. Given the available performance of this system, I was a little surprised to see that currently Ethernet is capped at 10GbE and FC at 16Gbps, but I expect that’s just the usual qualification cycle at play.


Tech Field Day events are organized by Gestalt IT and sponsored by vendors who thus indirectly cover our travel costs.

The vendors sponsoring Tech Field Day events don’t ask for, nor are they promised any kind of consideration in the writing of blog posts. All opinions (and mistakes) expressed here are entirely my own, and please send me a note if there are factual errors.