RedefinePossible

Good question. And does the internal cluster fabric matter for storage performance?

If the network didn't matter why are all/most AFA's FC?

@silvestri_73 Nobody said the network doesn't matter

latency matter to applications. And if they are real time apps (IMDB driven) then you want the shortest data path between your app and the data. This is the DSSD server side play.

raw NAND access is 60uS ... so why does it take 1mS to access a fast AFA ?

Answer is ... a bit of network latency but a whole bunch of storage protocol gunk.

lots of other processing (dedupe, compression, read/write integrity checking etc)

@jburton Fair enough, for VMAX, XtremIO, and VNX flash, we believe Fibre Channel is still the best network. 700-800ns latency.

If you were at EMC and ran products what would you do with this portfolio ?

standardize the interfaces. Unisphere is a good first step. Coming up with a standardized CLI would be great

Tough question - I'd look to harmonise the features/functionality where possible and simplify. VMAX & VNX could move to a single platform for instance.

I would also focus on (native) data mobility - ability to move objects/files/LUNS between platforms to make ViPR truly useful.

Only reason VNX and VMAX haven't merged is customer install base. If you could bring Flare and Enginuity together, or hide it behind an interface, would be easier.

@jburton You have to give customers a way to manage the silo'd "stovepipes" - we hear that is the biggest complaint

Continue listening to clients as well as use innovation to find better ways to solve the same problem we are already solving today

@jburton great crowdsourcing data; std mgt will be winner for #winning across the portfolio of EMC

Well done @EMCcorp @jburton - thx for sharing the data heon Crowdchat #crowdsourcing conversations

@CrowdSpotsAwesomeOrg

Crowd goes crazy #EMC #megalaunch #crowdchat

:) http://www.cartoonsc...

like QoS

ScaleIO does - it was designed for that environment

VMAX did have backend physical I/O prioritisation as a feature - not front-end host QoS.

True QoS almost impossible to implement with spinning disk where the I/O response time is unpredictable.

@chrismevans Agreed. SolidFire has a solution on all-flash. Not sure about the new HP array.

#VMAX has Host Ports I/O Limits too

and #VMAX has dynamic cache partitioning as well

@vGazza Yes, however those features are focused on workload balancing to optimise the array throughput, not to implement a QoS service level.

@chrismevans Host Ports I/O Limits can actually limit bandwidth and throughput per Storage Group

@chrismevans thus reducing "noisy neighbor" by enforcing predictability to service multiple tenants

@vGazza But that's done as a physical workaround; needs to be turned on it's head - deliver what the app needs not simply throttle the pathways.

@chrismevans if you are xSP and sell $/IOPS it would be nice to be able to model the performance available for each consumer, I guess

I say yes;

Drive cost out, help us manage data growth -- from IT

from biz guys (who something don't bring IT to the meeting) - "we need to become a s/w company"

@jburton does "become a sw company" refer to creating new offerings fast?

So is data growth still viewed as a problem more than an opportunity in your view?

in a single x-brick ?

20TB raw

scale-out to 120TB

In a single arrday. IBM tops at 1Tbyte, HP's new array is clsoe to 2 TB & SolidFire is at 2.8 TB.

Scale out was my next question. Both are important. Some applications need large scale up capabilities.

XIO shares memory - it is not a loosely coupled cluster (like ScaleIO) ... many folks will argue over whether this is a scale-out or scale-up design!!

we now support over 20 different 3rd party arrays - big hump was supporting Cinder southbound out of ViPR controller. EMC arrays and some 3rd party (NTAP, HDS) we did native drivers for

follow up - does ViPR need some kind of data migration feature to move between disparate platforms, to be more truly "software defined"?

Yes EMC is very innovative and that comes from @jburton from Day 1 when he joined EMC he pushed the envelope that's why he's President now cc @furrier

Well done @EMCcorp @jburton - thx for sharing the data heon Crowdchat #crowdsourcing conversations

depends on what you're trying to do.

not concerned at all - gaps are worse than overlaps

I love that angle. Key is to manage the overlap with a layer that customers can have a singular view

And not confuse the market as IBM discovered in the 1980s.

Hopefully this is where ViPR is going. Multiple tools with a single management solution to put the data where it needs to be

Specificially - XIO is an "array" - data services... latencies in the mS. DSSD is server-side flash... latencies < 50uS.

VipR is 2 things - control plane ... a cloaking layer to present common north bound interfaces to things like VMW and Openstack...

and a new storage system (Object, HDFS, block) designed to run against commodity drives

@jburton Is there any overlap between DSSD and XtremSF PCIe flash cards?