The future of VMAX is flash and SASy
EMC going 2-tier
EMC World NetApp's Tom Georgens was right. You only need two storage array tiers and EMC is going that way, to a VMAX with Flash and SAS disks.
VMAX was launched in 2009 and the time is getting ripe for a refresh of the technology. El Reg talked to Brian Gallager, president of EMC's Enterprise Storage Division, and the subject was VMAX 2 (our term).
We should expect to see SAS drives in VMAX, with Gallagher saying: "You will. I'll guarantee it. ... But it's not about SS; it's about FAST, FAST, FAST." FAST is EMC's automated data moving technology for tiered arrays. Flash is going to play a huge role in VMAX and the destination is a 2-tier array with SAS-interface flash and "big, fat, SAS data tubs." That's SAS data tubs, not SATA ones. NetApp, Gallagher said, is following EMC's lead.
Gallagher expects that in the medium term we'll need three storage tiers in VMAX but, eventually it will net out at just two. Some of the big, fat, SAS data tubs will be spun down to save energy, the ones holding old and cold data. The data on the spun-down tubs will be deduplicated, possibly by Data Domain, possibly by Avamar; Gallagher did not say which. Customers may be offered a choice anyway; Avamar for source dedupe and Data Domain for in-line dedupe.
We asked about write-once; read-many (WORM) drives and the response was: "I don't see that in the short-term."
We will see VMAX SPC benchmarks, hitherto untrod ground for EMC. Gallagher thinks the SPC-1 benchmark does a good job as an envelope test but does not reflect real-life VMAX and other high-end array deployments well, what with snapshots, replication and so forth.
He said: "EMC has recently re-invigorated the SPC relationship. We will see VMAX SPC benchmarks." That says to us that EMC is well aware of the 380,800 or so SPC-1 IOPS recorded by an IBM 6-node SVC/dual-DS8700 array system and the recent TMS RamSan-630 one that topped 400,000 SPC-1 IOPS. We speculate about something that will put EMC in the SPC-1 stratosphere, with a result greater than 500,000 IOPS, but that's pure guesswork on our part.
EMC is working with others and the SPC to encourage the development of a benchmark that does take real-life, high-end array deployments into account. We might call it the SPC-3.
We asked Gallagher about a SAS backplane possibility: "We think backplanes are a thing of the past." Today it's Fibre Channel (FCAL and FCESW) and SATA wires coming out of the DAEs (Disk Array Enclosures). Tomorrow, so to speak, SAS wires will come out of the DAEs and maybe SAS will go right through to the engines or possibly InfiniBand could be used. Gallager did not say which and maybe he couldn't as the choice has not been made.
FCOE (Fibre Channel over Ethernet) support has just been announced for VMAX. Gallagher expects high-end customers may elect to segregate their storage networks from general Ethernet traffic, but with mid-range customers being more inclined to pursue network convergence. EMC isn't evangelical about this; it will do what customers want. Personally Gallagher thinks FCoE "is the right way to go," and the transition will take time.
Other things we would have liked to ask, but couldn't because his efficient time-keeper cut us off, included the thinking about an entry-level VMAX, a sort of VNXe-style VMAXe, but that will have to wait for another opportunity.
The future for VMAX is flash and SAS with SAS flash and big, fat, SAS data tubs. The drive count will almost certainly go down, we think, because flash replaces so many hard drives and the data tubs will be capacious. VMAX flash and SAS: it's got a ring to it. ®
post inline correction
Data Domain is inline - NOT post process...
give it a rest
SAS is an interface not a tier. Enterprise Flash disks are usually based on SAS anyway with a handful on FCAL
You can have different tiers of disk with the same interface, larger slower ones that are cheaper than smaller faster ones.
Tieiring is only as effective as the mechanism that moves it and EMC have yet to prove that FAST actually is fast. It certainly isn't easy to use which defeats the purpose of using a technology that is meant to be all-encompassing.
Maintaining the suitable ratio of flash to required non-flash capacity is also not cheap nor practical (yet). Much like EMC's cache to disk ratio - they are happy to charge you a truckload for it but the benefits are limited.
I look forward to an EMC SPC result on VMAX. It is impossible to hide the complexity it takes (and cost) to produce a big number. I am sure they will achieve a reasonable number, just the cost/IOP and commands per usable TB of SPC testing will be massive - and that is something they don't want the world to see.
Try write about things you understand, more fact, less "guesswork". If readers wanted guesswork they'd go to a psychic!
Also, yet another netapp reference, so at least its consistent. Thanks for taking the time to vacate Netapp's back passage to attend and EMC jolly..