QuantumLeap

EMC CTO AMA
Curious about EMC’s Quantum Leap? Join EMC CTO and others for a no-holds-barred Ask Me Anything!
John Furrier
Q1: What is the main announcement today and what does "all-in" mean for the #allflash array products/solutions?
John Roese
it means that we think that flash is now ready for broad enterprise adoption across transactional workloads for the first time
Matt McDonough
@emcdssd is announcing two things (1) new product category Rack-Scale Flash and (2) our first product the DSSD D5
John Furrier
love the leap year theme
Chris Ratcliffe
We have committed to move our entire primary storage portfolio to All Flash Arrays. The first out the gate is VMAX All Flash.
Rodrigo Gazzaneo
@theICToptimist it seems that the economics for the All Flash Data Center are finally there
Susy Jorge Maestre
"finally here"? Other companies have been doing this for some time now. Couldn't it be argued that EMC is very late to the game?
Chris Ratcliffe
It certainly is for primary storage. We think secondary and data protection etc will remain mostly HDD or hybrid for the near future.
John Furrier
Susy great question. What's different
Peter Herdman-Grant
hi @ceratcliffe VMAX could already be an AFA, what's different after today's announcement?
John Roese
Flash has been used tactically for targeted workloads but now with broader options and availability plus innovation the use cases expand dramatically. Remember we were very early to put flash in storage systems a long time ago
Chris Ratcliffe
@DBAStorage In the past these were essentially standard VMAX running all SSD's. We've now released VMAX All Flash which has all the great VMAX features but includes significant code changes to optimize the architecture for All Flash.
Susy Jorge Maestre
@theICToptimist That's not even a little true. Flash arrays are "primary storage". It's not tactical. Solid Fire, 3PAR, & Pure storage are all touting use cases that are very generic and not tactical.
Rodrigo Gazzaneo
the difference is All Flash is moving from focused use cases to broader use cases.
Chris Ratcliffe
@DBAStorage We've also extended the Xpect More Program to VMAX and created a simplified packaging and licensing model.
Susy Jorge Maestre
@theICToptimist Also, lot's of people had flash in 2008. Again, 3PAR, HDS, IBM, etc. Saying that you had it back then isn't exactly a unique differentiator. I'd love to know why all-flash VMAX & why now/
John Furrier
let me ask this as separate question Susy. Great point
Clodoaldo Barrera
@ceratcliffe Do you expect the primary impact of DSSD as a component within VMAX or as the storage for Big Data applications (HDFS, Cloudera,...)
Peter Herdman-Grant
@ceratcliffe nothing wrong with repackaging and making the new AFA offerings clear, the part I missed earlier was "significant code changes to optimize the architecture for All Flash", thanks
Susy Jorge Maestre
@vGazza Again, this isn't true. Today Pure Storage talks about broad use cases all day long. So does HP & IBM & everyone else. I guess what I'm asking is, why is EMC only talking about it now and why the change in direction from XtremIO?
Rodrigo Gazzaneo
no change. It's a portfolio play because each technology is optimal for a need.
Peter Herdman-Grant
I remain surprised that we have not seen XtremIO, and now DSSD, made an integral part of VMAX, is there a reason for that?
Rodrigo Gazzaneo
If you only have a hammer you call every challenge a nail before you hit it.
Rodrigo Gazzaneo
It's not only about the media. It's about architecture and functionality. That's why the portfolio matters.
Peter Herdman-Grant
I read this blog today http://bit.ly/1QnyF6... VMAX is industrial-strength storage being made #AllFlash, it's actually that simple
Susy Jorge Maestre
I could make the argument that the only one with a "hammer" is EMC. Consider IBM, Hitachi, HP, and SolidFire all solve a broad set of problems, all address general purpose needs within a single product set. Why not EMC?
Chris Ratcliffe
There is no change in direction for XtremIO - it's still the #1 AFA in the market with over 39% share. VMAX addresses other use cases that no one else can 6 9's availability, multi-pb scale, mainframe, file, block etc all in a single scale up & out array
Susy Jorge Maestre
@ceratcliffe Chris, this isn't true. HP 3PAR makes the same claim. Market share speaks more to EMC's salesmanship than to the quality of the product. Please stay on topic: If XtremIO is the answer then why not incorporate everything there?
Rodrigo Gazzaneo
because architecture matters. An army needs snipers (@EMCXtremIO) and machine guns (#VMAXAllFlash).
Chris Ratcliffe
Both are true. IDC & Gartner list EMC as #1 in AFA. No other AFA has the combination of scale and features as VMAX (esp. Mainframe). EMC customers are smart enough to buy what works for them not what a salesperson tells them to.
Susy Jorge Maestre
@ceratcliffe Marketing claims isn't cutting it anymore. HPE 3PAR & HDS both have a portfolio of scale & features equivalent if not superior to VMAX. EMC customers have been buying more HP apparently than anything else if we believe IDC overall.
John Furrier
Q2: What is the big leap quantum that is with this announcement? What is the real impact to customers?
John Roese
Lots of quantum leaps in the announcement but one of the biggest is DSSD... super high performance for the most demanding new workloads out there.
Matt McDonough
for @emcdssd it is about game changing performance...in 5U 10M IOPS, 100GB/s, 100 microseconds latency and 144TB of raw storage. Imagine what you can do with this performance for your data intensive analytical workloads!
Susy Jorge Maestre
I'm still confused. Flash alone has already delivered more IOps than most people actually need. EMC now has 4 flash arrays technically. It's very confusing. Performance is an issue flash already solved. Please help me navigate throught his.
John Furrier
EMC folks: the trend that all storage for non-archive will be flash by 2020 according to EMC; is this a mkt shift plus performance trend?
Matt McDonough
Susy today all flash arrays are delivering lots of performance for many existing workloads, but if you take something like Oracle there still are a lot of workarounds such as lots of indexes, extra copies, materialized views, and silos.
Rodrigo Gazzaneo
it's about architecture, not only the media point of view. Not every AFA is similar.
Matt McDonough
With @emcdssd and Oracle in 1/5 the rack space and at 1/3 TCO and 1/3 the latency we can help customers eliminate many of these workarounds and derivative objects with Oracle.
John Roese
Yes, both because flash systems are maturing and are optimized for these but also because the HDD market is heading towards sequential IO models optimized for big data, archival and write once workloads. Seeing media bifricate in next few years.
Matt McDonough
Also, the need for more performance only further expands as you address next generation applications. There is a big need to help customers running Hadoop deliver Enterprise Analytics.
Stuart Miniman
Wikibon's @dfloyer articulated the importance of Flash as Memory Extension (FaME) which EMC DSSD fits into http://wikibon.com/f...
Susy Jorge Maestre
@vGazza Thank you Rodrigo but everyone says "it's about architecture". If that were true then VMAX couldn't be an all-flash array but you announced that today. I feel like you're tap dancing around my question.
Phil Dunn
@mjmcdonough Actually, I believe the trend in analytics is moving to in-memory, where latencies are practically eliminated, bandwidths much greater. w/10x compression, can handle ~160TB in todays SPARC M7 systems for example @ >1TB/s bandwidths
Susy Jorge Maestre
@mjmcdonough Considering that Hadoop treats disk as "JBOD" and is 100% more DRAM efficient than any storage array, I would argue that your use case is waaaaay off.
Susy Jorge Maestre
@mjmcdonough In fact, the nature of Hadoop relies on triple mirroring of the array and it can deliver sub ms response time from 15K SAS drives. The issue is the cost because of the triple parity. I'm not sure how DSSD solves that problem.
Rodrigo Gazzaneo
#VMAXAllFlash is different. Best platform for hi-end enterprise apps that compression-friendly.
Matt Schneider
It's a VMAX with All Flash, some tuning to leverage it yes, but mostly it's easy to consume product with sizing and scaling models built around it, just that is a BIG deal if you need mission critical storage in a mixed workload datacenter
Rodrigo Gazzaneo
#VMAXAllFlash brings #SDRF to AFA. Nothing similar in the AFA space.
Matt McDonough
@Phil_Dunn1 good comments. Big advantage with @emcdssd and in memory is now you have a near limitless memory pool with the fastest persistent storage out there. Very helpful for in memory DBs that use persistent storage as part of the working set of data.
Rodrigo Gazzaneo
but still, there are workloads that are more "dedup"-friendly. @EMCXtremIO will crunch it like nothing else.
Phil Dunn
So what about pricing? I have yet to see any mention during launch? Even ballpark #'s would be nice.
Matt McDonough
Susy, we've worked on a Hadoop plug in with our partner @cloudera that enables customers to run an application like HBase on top of HDFS unmodified and only write once to storage while still maintaining data locality and driving 10X the performance
John Furrier
@Phil_Dunn1 EMC said it will be available in March
Phil Dunn
You mean tomorrow?
Matt Schneider
@Phil_Dunn1 partners can price today, go ask your VAR or EMC rep
Matt McDonough
@Phil_Dunn1 we will be GA in March, but we expect our ASPs to be in line with purpose built AFAs from a usable $/GB basis. Will learn a lot more in the coming months.
Susy Jorge Maestre
@vGazza Hitachi and HP 3PAR all have AFA with equivalent replication scenarios to SRDF. In fact 3PAR claims to be 6 Nines without needing replication. How do you compare?
Rodrigo Gazzaneo
We are confident #SRDF is the most advanced replication solution in the market and happy to have a conversation.
Phil Dunn
@mjmcdonough So it seems that DSSD basically is a bunch of NVMe-accessed SSD modules connected through PCI-E switching inside a box? How does it scale across many servers? How does it compare to EDR Infiniband which will be out soon too?
Susy Jorge Maestre
@vGazza If it is, then why do you have Recover Point? Why not SRDF for everything?
Rodrigo Gazzaneo
Because not all workloads demand the scale and service levels of #SRDF
Rodrigo Gazzaneo
Because not all workloads run on Mainframe, for instance.
Rodrigo Gazzaneo
Portfolio is a strength. A single product can not span all use cases like a strong portfolio.
Matt McDonough
@Phil_Dunn1 today DSSD connects to 48 servers redundantly. We have the world's largest PCIe NVMe fabric internal to the D5 (two of them for redundancy). Every server has a direct connection to each flash module with no CPU in the data path.
Matt Schneider
Asking why to provide choice to customers seems like an oxymoron, the plethora of options solves different requirements and allows customizing the storage to fit your environment... while many storage requires changing your environment