Pure Accelerate: FlashArray Gets Synchronous Replication

Synchronous replication is a key feature that many enterprises depend on for data protection as it offers the ability to ensure data is 100% copied to a remote site in the event of a disaster. On flash storage, sync replication may seem a little at odds with the idea of providing high performance, but the performance impact trade-off is essential where RPO of zero is needed.

So finally Pure Storage has brought synchronous replication to the FlashBlade product line with a new feature called ActiveCluster. In comparison to traditional replication where one volume is active and the other passive (typically read/write and read-only), ActiveCluster presents two fully active volumes out of two FlashArray systems. Each volume is capable of servicing I/O and thus be implemented in configurations such as a VMware metro cluster.

Active/Active

Prior to the official announcement, bloggers were pre-briefed on ActiveCluster (as part of a Storage Field Day Extra event), offering a chance to discuss some of the detail of the implementation. Managing multiple concurrent writes to two copies of a volume isn’t a trivial task and there’s a risk of a “split brain” scenario if the network is interrupted. For this reason, Pure offers a cloud-based arbiter (called the Pure1 Cloud Mediator) to manage network disruption scenarios. However even with this in place, an application needs to be cluster-aware. If not there’s the obvious corruption risk, with “latest writer” taking priority. Incidentally if you watch the first of the SFD videos (link) you can see me ask about the way in which I/O updates are applied in the event of a non-cluster aware update. What I was trying to get to was the point that either one array or the other has to be the leader, because one has to make the “latest writer” decision.

A Single View of Data

Having true active/active means on platforms like VMware vSphere, a datastore truly looks like a single entity. As a result, when moving virtual machines around the infrastructure, only the active VM in memory needs to be moved (vMotion rather than Storage vMotion). As far as any cluster member is concerned, the array pair looks like a single array, just with more paths in place. Naturally it makes sense to favour the nearest path to the host and there are configuration options within Purity to do that for a replicated volume or group – in Pure’s terms called a Pod. The array simply advertises the shortest links from array to host via ALUA, once configured by the administrator (see the ActiveCluster demo by Larry Touchette for more details – link).

The Architect’s View

I have two views on the implementation of ActiveCluster. My technical head says this is a great implementation, sorely needed by customers and doing active/active is much more interesting than traditional replication. I would however like to have the ability to run active/passive for environments where I don’t have a cluster aware setup – just in case.

My more cynical side says that the decision to create an active/active cluster could equally be about matching and jumping the competition. Dell EMC VMAX needs VPLEX to add active/active to SRDF. XtremIO X2 now has native replication but not active/active – again VPLEX would be needed. HPE 3PAR has metro cluster support but the replication isn’t active/active. ActiveCluster addresses customer needs and provides sales/marketing a leap over the vendors most likely to be in a bakeoff with Pure. Now, none of this is a bad thing, in fact, if that was part of the plan, it’s just good product strategy sense and worth applauding.

You can listen to me talking to Ivan Iannaccone, Director of Product Management at Pure on the Storage Unpacked podcast. We chat about the latest software updates, including ActiveCluster.

Active/Active

A Single View of Data

The Architect’s View

Further Reading