HPE Nimble – synchronous replication (finally)

UPDATE: 28-05-2019 I posted an article describing setup and test of peer persistence and replication. Check that as well.

Nimble was always on par with 3PAR in the midrange storage category, it was safe to say it can perform under sub milisecond latency, can squeeze 300k IOPS for sure. This is something which is the top most feature of this product and the largest advantage of the platform, next to the well known deduplication and compression capabilities. This latter can have a huge impact on useable storage since it does that really well but more importantly inline. (HPE can be self confident since majority of their installed base do report to Infosight, they can get exact real life measurements about the achieveable dedup/compression ratio. I’ll explain Infosight later.)

Is was all good up to this point, but had one quite important issue. Rephrasing it, not issue but lack of feature.

Client: Is the replication possible?
Me: Sure it is, asyncronous replication is supported.
Client: But our two 3PAR storage can do sync.
Me: If you need syncronous replication you might want to keep using 3PAR.

As opposed the Nimble pricing is pretty much in the ballpark of a similar 3PAR system, but it’s features list a bit shorter and clients who could afford such a storage could not take it seriously as a one-to-one replacement option of their 3PAR – or other vendor – storage systems. There are many clients using Peer Persistence and VMware Metro Cluster. Nimble cannot and will not deliver the same what a 3PAR does. Simply summarized, if anyone needs full in depth configurable parametres. So if someone is keen to set RAID6+2 for a CPG and set AO up, Nimble is not for them. Period.

But then for who?

Nimble puts the easy usage in focus. So you need 22TB of storage? Here it is. Application that will run there? VMware? There you go. Dedup? Okay, it will be deduplicated. You cannot twist and turn it, no need to understand the trio of CPG-Virtual Volume-chunklet.

Who needs this?

For anyone who wants storage, fast storage without the complexities attached.

Some weeks ago it was offically – and finally – announced the arrival of syncronous replication in NimbleOS. It will be available somewhere in 19Q1. Important to mention this does not mean that Nimble systems cannot deliver 99,99999% availability – Infosight again there to proove it. There are many cases where the storage systems itself can do such high availability, but other datacenter components, like cooling, power lines can’t. It might be fair to state that the storage will not crash or die in any event, but it is still one device – so as Synergy – and components in it can be dupliated but all of those are in some rack units, concentrated to the same squaremetres. Some customers required to rule out even this factor and in order to do that sync capability was a must. Async was always possible, so to some extent resilient data centers could be created in the past based on that.

There are still some unclear areas around sync replication. It requires the same type of model on both sites, but my question would be what if the target site has a more beefy model – is that an issue or not.

Replication will be supported over IP only and as usual it comes without buying additional licenses. This is a huge plus, I go nuts if I see grey checkboxes that are not clickable only if you pay. Already procured and running Nimble systems will get the feature too.

In current readyness VMware, MSSQL+Windows are supported, but they have sketched plans to support Oracle as soon as possible. No surprise if solution need a wuorum/witness which must – capitals – reside at third location. Not much information is available if this is an appliance or something as an installable component for Linux/Windows – this latter one would be similar to the arbiter of Simplivity.

Replication can be configured per volume basis, I don’t even understand why would it be in any other way.

In a sync partnership, host on both sides can use the same volumes and the paths for the storage connectivity is controlled by ALUA integration, setting the active path to the volume which holds the ownership of that volume. If failure happens will put the previously “unused” paths into active state.

Transparent failover is possible though the surviving storage can bring up the IQN/WWN of the failed storage.

I hope and also believe in our small and belowed county will see more HPE Nimble systems in operations as up to this date Nimble could be excluded from procurements due to missing features. The best and mind blowing feature would be if 3PAR could replicate to Nimble and vice versa – at least in async mode.