VMware View on VMAXe or big VNX?

Even though VMAX(e) doesn't have FAST cache, it does have FAST VP, and its FAST VP is VERY granular (~7.6MB chunks).

Set it up much like you would a VNX that didn't have fast cache - use FAST VP to store most of the data, with perhaps a dedicated set of SSD LUNs for the Linked Clones.

If I'm understanding correctly that the VMAXe is 100% virtually provisioned, then I wouldn't be able to create a dedicated RAID group of EFD's.   Correct?

I am willing to trust the fact that FAST VP is much more granular on VMAXe and the moving around of bits is more dynamic.   I'm still struggling to see any alternative to designing around the lack of FAST Cache without having to add a lot more disk.

Even though its virtually provisioned, you can still create a LUN that isn't allowed to cross tiers.  Its wont be on a specific raid group, but it WOULD be bound to exclusively SSDs.

You dont need to design 'around' FAST Cache, because the VMAXe simply has more than enough.  A 4 engine system has 512GB of cache, which is more useful than fast cache because its faster (by at least an order of magnitude).

To some degree though, you're right.  A VMAX(e) based VDI environment IS going to cost significantly more per desktop than a VNX based one.   For most scenarios where you are talking about these numbers of sessions, its probably better to consider scaling out with multiple VNXs (maybe 2 5700s) than scaling out by going VMAX.

Dan,

What's your VDI number? On EMC WP,  the reference infrastructure is to use 2 mirrored 100GB EFD as FAST Cache in VNX5300 to suppport up to 2000 users. The available FAST Cache capacity is only around 93MB. VMAXe would be qualified enough on the hardware perspective. I heard a VMAXe is currently tested for 5,000 concurrent VDI users.

Krichter, please go to the below URL for WP.

http://powerlink.emc.com/km/appmanager/km/secureDesktop?_nfpb=true&_pageLabel=image7b&internalId=0b0140668044f523&_irrt=true

must be restricted to employee/partners, i can't get to it either.

When trying to access that URL I get the error:

"

The content you are trying to access is unavailable. This content may have expired, the link may be incorrect or you may not be entitled to access it with your account.   Click here to return to the Powerlink home page.

We apologize for any inconvenience."

   -Alex

DeLF wrote: When trying to access that URL I get the error: " The content you are trying to access is unavailable. This content may have expired, the link may be incorrect or you may not be entitled to access it with your account.   Click here to return to the Powerlink home page. We apologize for any inconvenience."    -Alex

You need to have a powerlink account to access these EMC WP

Your consideration of using FAST /w EFD to absorb boot storm is great.

Please look at your user profile carefully, if they are able to be cataloged into several type (similar OS / App), Linked Clone mode on EFD would be wonderful.

I do and I was logged in trying to access the link. Still didnt work.

    -Alex

I ended up building out a model that showed the 2,000 user mark was right about where it made financial sense to consider VMAXe in this particular scenario.  The further under 2,000 users, the more cost prohibitive it was just because the VMAXe has a higher initial cost of entry.  However, it scales considerably better and the large global cache and very fine granularity of FAST VP on Symmetrix did make a signficant difference at scale.  There is a reference architecture for 1,000 View users on VMAX /SE, and my requirements were a little under the ones documented there.   I had about double the EFD in my config for the additional performance requirements to support 2,000.  

Slowrider, how did you use your EFD? store linked clone gold copy?

I didn't get that detailed yet to determine if I would try and pin the gold copies to the Flash tier or "trust the system" to auto-tier the right chunks, but there's certainly enough Flash there to do it.   The 1000 user ref architecture on VMAX had 8 EFD's and served out at peak of 37,000 IOPS during the boot storm and the desktops were booted in 15 mins.  The config I put together had 16 EFD's.    My IOPS requirements were less than what is documented in the 1000 user reference architecture (7 IOPS per desktop steady-state), but I added the extra EFD to have more IOPS for boot storms (even though there won't be 2000 booting at once, probably not even 1000).   Most of the capacity came from SATA.   Again, it made sense at 2,000 users, not so much sense (financially) if they start at 500 and slowly scale to 2,000.

The more I dug into it the more I was impressed by FAST VP on Symmetrix.  One could make the argument the VMAXe would actually be easier to manage than the VNX at scale, if you can truly get away with one virtually provisioned pool rather than dealing with multiple pools/raid groups on VNX, and FAST Cache.  One could also sleep easier at night with no concerns about what happens if you encounter a (rare) situation of FAST Cache getting disabled or needing to be flushed in preparation for an expansion.   I also get better than 5 9's availability and things like non-disruptive upgrade (that doesn't require trespassing LUNs between SP's when they reboot).