VPLEX Geo = Awesomesauce.

So – VPLEX Geo is now GA.   For the first time, it’s possible to have active-active geographically dispersed storage models across Async distances (~50ms).   It’s going to take some time for this to “sink in”.

The VPLEX Geo announcement is an the GA (and literally available now) of VPLEX GeoSynchronicity 5.0 (the software) and new VPLEX hardware (bigger, faster, stronger).

So – what does this mean in a nutshell?

1. VPLEX now has a 3rd site witness option.
2. The VPLEX Metro (synchronous) use cases with VMware have been hardened (partition behavior works much better now, and has been developed hand in hand with VMware folks working on the next version of vSphere).   BTW – stretched vSphere clusters over async distances are, as of right now, still not supported though VPLEX Geo is out.   Stay tuned for more on the topic.
3. There’s a new class of options – where certain workloads can be active-active across asynchronous distances.

BUT – beyond that (which on its own is awesome and extends the “category creator” nature of VPLEX), there’s a lot more.

The core technology (distributed, coherent transactional cache) behind VPLEX is what we think of as a “foundational technology”.   While it’s first manifestation is as a product (VPLEX) which virtualizes external storage and makes it active/active across geographic boundaries – it has MANY interesting applications.   BTW – if you look at the Project Lightning blog post, you can imagine how the technology inside VPLEX could also help there.

But – here’s another FASCINATING idea of how it could be used.   This is a real Proof-of-Concept developed by the EMC CTO office in conjunction with the VMware vmkernel team.   It’s a response to an interesting customer challenge…

This customer runs large HPC workloads, and like many in their position, one frustration is that sometimes a job runs for several days, and then, before it finishes, a new job comes that MUST run immediately, making the days of prior work wasted.

They asked: 1) Could we capture distributed system state as “one big consistency group” – both the compute, memory, and dataset all together; 2) could we “push those workloads to a public cloud”.   We said “sure, let us work on it”.   Then, they added a tough twist: “we don’t want to ACTUALLY move our dataset to the cloud provider – but some of them would suck over a WAN – can you help with that too”

So…  After developing a custom vSphere vmkernel that enabled capturing system state across many VMs with exact synchronicity and a dramatic acceleration of the snapshot and resumption process… we integrated in something REALLY cool – the ability to “push out” a virtual VPLEX (vVPLEX if you will) instance which presents the storage to the VMs as they are running in the cloud – without the dataset ever actually really existing at the service provider using VPLEX’s AccessAnywhere approach.

Check it out:

Download the high-rez version here in MOV and WMV.

So – here’s an interesting question dear readers… This customized vmkernel would expose this “consistency group accelerated snapshot via an API” and could be used for all sorts of stuff.   Right now, this may/may not ever make it in, the main question being “is this something customers would like, and partners would leverage, or is it just relegated to HPC?”  I can think of MANY use cases, a ton of which cross all sorts of use cases.

But – would you want this to become a general vSphere capability?

Comments

[Andrew Fidel]

Chad, so you're saying that they run a VSA VPLEX node which federates the VM data requests back to their actual VPLEX controllers? If so that's really, really cool but I'm unsure how applicable it would be to commercial workloads as most workloads that would require the kind of heavy lifting such an option would provide would also need storage bandwidth that would make it unworkable (at least the ones I'm familiar with). I guess one option would be to have a data only DR site where some/all of the compute capability is provided by the cloud provider, but then why not have them do storage for you as well?

[50mu]

VPLEX rocks ! #yam

[Tim Kossack]

Hi Chad,

any chance at least some of the functionality of vplex will find its way into the vnx plattform (especially since "VPLEX Metro" sounds a lot like "metro cluster" where you don't need a totally separate hardware for...;-)?
especially for smbs the cost of two different systems in order being able to run globally, maybe cloud-enabled dispersed infrastructures is simply to high.
is it right to assume that vplex and vnx are identical hardware-wise?
will this replace mirrorview etc. in the long run - maybe you could eloborate a bit more in detail which role vplex plays in the emc-ecosystem...

[Tim Kossack]

Hi Chad,

will at least some of the vplex-funktionality find it's way into the VNX(e)-series?
for SMBs like us, two different systems are simply not affordable...

[Mark Burgess]

Hi Chad,

I attended all the EMC World sessions on VPLEX and there were a few outstanding questions regarding VMware integration.

It was agreed you were the best person to help!!!

The questions are as follows:

1. Will vSphere 5 resolve the problem of VMware HA not restarting the VMs when the cluster is lost and the hosts cannot see the storage, you referred to this in the past as VM guest behaviour is “indeterminate” (i.e. using the new storage heartbeat feature)?

2. What stretched HA cluster sizes will be supported in vSphere 5 (i.e. will it be bigger than the current 8)?

3. With vSphere 5 and VPLEX 5 are there any remaining limitations with stretched VMware HA clusters (clearly in the past you have not recommended this solution)?

4. When will vMotion using Geo be supported?

5. When will stretched HA clusters be supported with Geo?

6. Are there any plans for VMware to support round-robin multi-pathing with VPLEX?

7. What are the plans for VPLEX to support vCenter SRM and do we really need it - it would add considerable cost?

8. What are the timescales for VSI based VPLEX provisioning?

9. What are the plans to have a VPLEX GUI process (i.e. VSI) that will provide disaster avoidance and vMotion all VMs to the other site using DRS to balance the load - this would also need to change the preferred VPLEX site for the LUN?

10. What are the timescales for VPLEX to support VAAI?

11. What are EMC's plans for supporting the VMware HA Application Monitoring APIs and is this relevant to VPLEX?

It's amazing how you can go to EMC World and this trigger even more questions!!!

Many thanks
Mark