EMC, you big tease! At last, the specs for million-IOPS VNX2 Getting tired of EMC's teaser campaign for its coming VNX refresh in September? Relief is at hand: slide decks, blogs and PDFs seen by El Reg's storage desk confirm that EMC's VNX2 is a 7-product line-up optimised for multi-core CPUs and flash with new operating software. VNX arrays were getting slow because the software …
Standard upgrade of hardware. Almost all vendors doing multicore and dedupe. No big deal.
Isilon seems far advanced technology than VNX. Clariion to VNX to VNX2 was advancement in CPU and SAS technology. No major change in Flare OS or DART. EMC seems to zero in on Network Appliance but customers are all looking outside these two vendors anyway.
ZFS does dedupe, but it is not that good. For instance Tegile and Greenbyte use OpenSolaris + ZFS with a rewritten proprietary dedup engine to power their enterprise storage servers:
http://www.theregister.co.uk/2012/06/01/tegile_zebi/
http://www.theregister.co.uk/2012/10/12/greenbytes_chairman/
Still no unification? I guess you are referring to the seperate CPU's and memory and optimised operating systems for file and block all managed under Unisphere as a single array?
So you have other vendor arrays which have 2 CPU's and so much memory total for all block, file, replication and advanced features (compression, dedupe etc), where as VNX has a degree of scale out (upto 8 x-blades) all dedicated to file serving only which lets the backend block element to get on with what it does best.
Who care's how much seperate hardware is behind the scenes (higher resilience as well) as long as its managed as a single unified device.
"The VNX controlling software is now called its Operating Environment (OE) and it, unlike the older Flare"
The VNX code has always been called OE and hilariously, had cx5 in the file name. With the separation of the x-blades still I'm hoping you have further information cause from what it looks like I don't see anything new
Active/Active, but not if you use pool luns, terrifies me. They're having trespass issues with the controllers. Do not be the first, second, or the 100th to enable this. When EMC released thin provisioning on the CX there were serious loss of data corruption issues. This screams to me why it's been delayed and rather than fix it they think they can fix it in code later.
Also, can't wait to see the price to get 4.2TB of flash cache from 200GB drives.
The article also misses one important part that all VNX will move to capacity pricing and frame licensing will only be available on the VNXe at this stage.
We priced up upgrading our VNX controllers but were told we wold need to purchase capacity licenses to match the existing drives. Hope this goes away as that is almost the same rice as an entire now system.
So for me the thing that stands out with all of the VNXe specs is the continued specification of 2TB maximum LUN size. In all of my (increasingly) meagre experience, the LUN is the unit of storage presented to the server by the array.
So what I obviously don't understand, in this day and age, why is 2TB still the maximum? MBR vs GPT for some reason? Or am I completely clueless (it wouldn't be the first time)?
I have SMB customers with larger shared datastores than 2TB - let alone larger enterprises (in which a 10TB fileserver is the small one in the branch office). Is there some unit of storage that aggregates 2TB LUNs together at the array level, or is one expected to present 5x 2TB LUNs to the server and aggregate them there (with mdadm / Storage Spaces / etc)?
I think it boils down to performance. There may be some I/0 advantages to multiple LUNs on the backend if they live on different disks or are owned by different controllers. When drawing from a larger Lun versus smaller LUNs when it comes to virtualization the bigger the store the more vm's need to access that Lun which causes higher queuing greater wait times for access and disk contention for example . Tack on legacy support for smaller LUNs backup performance and time and I think EMC is just reacting to opportunity cost of going bigger.
EMC CX4, EMC VNX-Gen 1 all had been Multicore. Atleast advertised in the past but looks that it may start working with VNX2.
Have a look at Isilon, Tegile, Tintri, Nimble. They all are a better MCX arrays, does better dedupe,compression and costs less. All of them does scale-out which VNX does not.
Nothing new and So many limitations suggest its a beta release.
- Active - Active only without thin luns. Who uses system without thin luns anyways.
- There are other vendors providing multi million IOPs with flash what's new here. I heard skyera claiming 5M IOPs.
- Read only 1 Meeeelion IOPS with r0 and r1_0 on SSDs. Real world workload stays questionable!!!
- Dedup is only promised and not provided at this time. Since last October there were leaked promises about a 1MIOPs it took a year since then that suggests me another year for dedup :)
- I think with data services enabled it wouldn't even get 1/5th of Meelion IOPs.
If VNX2 does represent a huge rewrite of FLARE, then I advise all EMC customers and prospects to hit the REJECT button until some poor suckers try it out for at least 12 to 18 months. Massive rewrites of very old code bases have a 100% track record of having extremely costly data loss and outage bugs. My advice - don't touch it, despite the fanatical EMC sales force and channel representations. Only consider it in September 2014, not 2013.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
