50mu – about storage

The mother of all arrays has just been given an upgrade!

Well ok, maybe EMC did not produce the mother, since it’s fair to say IBM 3390 disk subsystem came first, but since the first Symmetrix came out in the early 90s with as much as a dozen or two disks, EMC has come a long way. They set the standard when it came to enterprise storage arrays. And it wasn’t just size that mattered back then: performance was and is still the number one objective for the Symms. After the “dark ages” (roughly before the year 2000) things got serious with the DMX series in 2003. The number of disks went up and loads of cache had to make sure that performance was guaranteed. DMX1, DMX2, DMX3/4 were quite a success.

And then there was VMAX

In 2009 a new array, based on industry standard components (intel CPUs) and less proprietary and more interchangeable disk enclosures, as seen on the Clariions – the DMX3/4s had similar DAEs as well, by the way, but were still based on PowerPC CPUs) and an explosion of cache, so it seemed. The maximum amount of cache a DMX4 could have was 256GB, but VMAX doubled that, to begin with, and nowadays it’s up to 2TB when using 8 engines in a VMAX 40k!

And with VMAX also came FLASH. Hugely expensive in the beginning, but as it goes with most (non) volatile memory / storage, prices went down rapidly and the Tier 0 was born. And if that wasn’t enough, the scalability went through the roof as well. The DMX-series allowed 2400 disks, but VMAX upped that to 3200, but that’s not enough, as we will learn soon enough.

VMAX3

EMC redefines the description of this new VMAX3 to not just being a large storage array, but being “industry’s first Enterprise Data Service Platform“. But what do they mean by that?

First of all it runs the HYPERMAX operating system and EMC explains this as an OS that allows services that traditionally run outside the array to now run inside the box.

Storage provisioning can now be done with a simple click by only defining the service levels once and assigning a service level to provisioned storage. But haven’t we heard the single click myth before? There’s no such thing as a single click provisioning system, right? So how’s this new array any different from the existing arrays? The single click points to the effort needed to change the current service level into another service level. Simply choose the service you want by selecting it, and that’s the single click they’re talking about.

Hypermax is a significant enhancement of Enginuity and has its own embedded storage hypervisor. This allows various data services to run inside the box. Think about management consoles, file gateways, cloud gateways, data mobility solutions and data protection solutions that can now run on the array itself, instead of having to implement separate data movers or management servers. All of this can now run using the array’s HA and virtual matrix to ensure availability.

The big advantages are obvious as you don’t have to buy as many boxes as before, which costs less floor space, the power requirements go down and cooling therefore too and so APEX and OPEX can decrease significantly. Also, since separate boxes can now be virtualized to run inside the array itself the latency between these separate boxes can go down as all these services are now integrated closely together and run on the same hardware. I must admit I was a bit sceptical when I first heard this during EMC World 2012, but it seems EMC has finally made that promised step.

EMC claims this new VMAX3 to be THE most powerful system on the market and with power usually comes complexity in management. As said before you can now categorize different workloads into different categories which will then allocate the suitable resources for that typical workload. For example you can have sub-millisecond latencies, which will allocate the data on FLASH, 1-10 ms to “regular” storage and for example 10-20 ms to the lowest tier in the box (wasn’t that called FAST or FAST VP before?). When these so-called service levels are in danger the system is supposed to send you an alert, so you can then decide whether or not you have to add additional resources like FLASH, rotating disks or port, CPUs, cache or whatever variable you can think of. In the end YOU decide whether you want to optimize on $/GB or $/IOps.

AFA – the All Flash Array

All three new models can be configured as an all flash array. The data density can be as high as 600TB per floor tile with a maximum of 4 PB of FLASH in a single array.

The VMAX3 now has:

• Up to 384 CPU cores
• Up to 3 x IOps
• Up to 3 x bandwidth
• 2 x logical devices
• > 2 x Front end ports
• 6 x IOps per floor tile
• 2 x TB per floor tile
• Up to 5760 drives
• Up to 256 ports
• Up to 3.2 million IOps
• Up to 2 engines + drives in a single cabinet (the VMAX 10K only had 1 engine + drives in the same cabinet)

So density AND performance have gone up, which is incredible since the smaller rotating drives came with a lower number or rotations per minute (15 k versus 10 k). A full 15 drive 15 k DAE was able to perform at 15 x 180 IOps = 2700 IOps. The newer 25 drive DAE (except the 300 GB drives) were able to do 25 x 140 = 3500 IOps. So when comparing full DAEs the smaller form factor delivers a better performance per DAE and therefore per floor tile! But six times as many as before? That can only be done using FLASH. So this comparison with the older arrays is somewhat unfair, since even those could be installed with FLASH. But then again, the VMAX3 can house more FLASH, so that’s where the 6 x must come from. I wonder if the back-end is capable of handling these amounts of FLASH as well. In the DMX-series (or even CX4/VNX for that matter) EMC always warned us about having too many SSDs on a single bus and the best practices pointed in directions where spreading them as wide as possible was – sort of – the big message to be read between the lines. The XtremIO box was a more FLASH-optimized system, so I wonder… does the VMAX3 have a FLASH-optimized back-end as well? It IS the industry standard monster of storage arrays, so one would expect this to be true: The virtual matrix is based on infiniband, the back-end is based on SAS.

EMC claims that the 200K model will probably fit most of the storage customers today and if you really need performance and/or scalability the 400K will take you to a level no other brand supposedly has to offer nowadays.

When talking snapshots, the new VMAX can take as much as 1024 snaps PER SOURCE, which is the leading benchmark nowadays, so EMC will match that criterium. And just like with the VNX snaps, you don’t have to create a dedicated pool to contain copies of overwritten data, all snaps are now pointer based and since the snapshot service will simply run as a different thread, servicing the production data is not impacted by keeping up the running snapshot.

EMC ProtectPoint

Another big step forward is this newly introduced feature in the world of backups: ProtectPoint. Until now backup environments had dedicated resources: backup server, media agents / storage nodes, tape drives / disk arrays / storage nodes and all data that was being backed up had to go through that back-up environment. And during that backup the only real way to ensure data was consistent, was to shut down the db server during that back-up. Another way was to temporarily suspend the db services, take a snapshot and backup-up the data from the snapshot.

With ProtectPoint you first of all could get rid of that additional backup-up environment, since data is being transported directly from the array to the backup-up target, without any help from an external backup environment! Secondly: during the back-up the db server only needs to be in back-up freeze mode so a snapshot can be made, BUT: still no external back-up environment is needed and data will flow directly from the snapshot to the back-up target. Data Domain is supported as target and application integration is provided as well, so data consistency is guaranteed.