[August 30th 7:30pm PST - Updated with some minor fixes]
“Project X” (XtremIO) is FACE-MELTING + MIND BLOWING. I know that I use funny turns of phrase (“chadisms”) – but wow, check this out, and then tell me if you think I’m being hyperbolic.
- Insane performance and latency. Think 250K IOps per X-Brick, delivering a consistent 500 microsecond (that’s 0.5 ms) latency.
- Insane in-line data deduplication. Think that every single IO gets deduped on ingest. For EMC Data Domain customers – this is how Data Domain behaves. It’s not a post-process (put data in, and dedupe it on some scheduled basis), but happens in real time. In this case, however, it happens while maintaining that crazy 500 microsecond response time. This means that a) you don’t actually need to write it (so effective IOps and Bandwidth can be insane), and for use cases with high degrees of commonality, like general purpose VMware, vCloud Director, and VDI use cases. In VDI example – it means that regardless of persistent/non-persistent deployment models – you can be simple, and space efficient.
- Insane snapshots and thin behavior. In fact, like the dedupe – this is just “how it works”.
- X-Bricks which are (like all EMC hardware) leveraging commodity x86 hardware – so the magic is software.
- Insanely… all that works in a scale-out fashion.
Chuck talks here about some of the hardware in the platform.
Want to know more? See some killer demos? Read on!
Here are the demonstrations we’re showing at VMworld. The config is the one below (this one details the vCD use case) – the servers were the only ones we we could get on short notice, and unfortunately they were not even close to being able to load the 4 X-bricks :-)
- Demo: Creating a OLTP workload on 4 X-Bricks – and getting to ~500K IOps, with sub-millisecond response times.
- Demo: Creating a 1000 VM “cloud” out of 500GB vApps using vCloud Director (with huge acceleration through flash, inline dedupe, and also a great VAAI implementation), and then generating a nutso load.
- Demo: Creating 500 persistent and 500 non-persistent 30GB VMware View instances – and consuming only a few hundred GB because of inline dedupe.
Check out the demos (product of a joint XtremIO and EMCer effort – thanks guys!) below:
Also – VMware will be sharing their own testing results (thank to Itzik and Garrett – vSpecialists who have been working with VMware on this project) for their and EMC’s own internal VMware View use. This will be discussed in EUC-EUC1190: VMware View 5.1 Reference Architecture, Monday Aug 27 11:00-12:00 PM and Thursday, Aug 30 12:30-1:30 PM – and the session is FILLED with the testing results.
Here’s a quote from Donal Geary at VMware Cork who completed the testing:
“Many thanks to you Garrett and Itzik, for coming to Cork and the long hours you put in, getting this POC off the ground. The results are outstanding and I am hoping that XtremIO will play a big part in our VDI expansion in the coming months
XtremIO… Mind = Blown”
So – was I right? FACE-MELTING? MIND-BLOWING?
So… before you get too excited (or have I done that already) :-)
- Yes, XtremIO is that cool, and that is all real – but not GA yet. I’ve had analysts and customers respond “it seems very mature”. It is, and does everything we push at it. But, it’s important to know that “mature” is relative to their development cycle. EMC acquired XtremIO relatively early compared with the acquisitions we do. Why? Well – customer feedback was a huge part. Another huge part is the engineering team is very strong. Another part is that their technology – particularly that they STARTED with a presumption of scale-out was really unique. We’re using time to carefully identifying customers that are a fit, and making sure that we spend time to get the hardware right (merging with EMC’s hardware team – go back and watch Chad’s World Episode 12 – the IB SLIC is for the next gen Isilon and XtremIO hardware), hardening and testing the availability (both HA and NDU) to get it up to “EMC Logo” expectations our customers will have. Those availability things take time – because (remember I’ve said this before) storage is persistent – so if you have a bad day, it’s a really bad day. All that said – if you’re an all-flash array startup – how do you stack up to those demos and capabilities? … Oh and I really, really hope for your sake you’ve started your engineering presuming that scale-out needs to be built-in from day one… right?
- When it is GA - is it right for you? That will vary customer to customer. While XtremIO is as sexy as Isilon (both examples of how you architect when you can start from fundamentally different assumptions) – the mass market continues to be “swiss army knife” storage aka VNX. Don’t get me wrong – VNX is growing, Isilon is growing like it’s on fire, and we expect XtremIO to be… very popular :-)
So… Hybrid array configurations (Flash + magnetic media with something like EMC FAST) coupled with a sprinkling of server-side flash will remain the dominant storage model for a while (for reasons I discuss in this post here).
Still – it shows the power in flash-land of starting with a blank piece of paper. Cool, eh? What do you think? Do you have a fit for this sort of technology?