Using Pure Flash or Cache for VDI

Over the last couple of days of Storage Field Day the conversation of Flash vs Cache has been discussed multiple times. Flash vs Cache is an interesting topic for VDI. Do you want to put your whole work load into Flash or use Flash as a Cache and balance the workload with traditional hard drives?

For the purpose of this article I am only listing the vendors that were at Storage Field Day.

Below is list of vendors that are using Pure Flash for their Storage Arrays:

Pure Storage
Nimbus Data
Violin Memory

Below is a list of vendors that are using Flash as a Cache

Nimble Storage

The vendors that are offering a end to end solution with Flash are trying to bring down the cost of Flash by using techniques like duplication, commodity hardware, build your own drives and will talk about power savings. The Flash as Cache camp talk overall cheaper cost per GB, need for cheap disk and that sequential IO are still better on spinning disk.

If you’re after an clear winner for Flash vs Cache it’s just not the simple. The feature sets between all the different vendors vary quit a lot and have different value propositions. I think it’s important to break down what you need for a VDI solution and make your decision based on that.

Replication – You need the ability get user data and golden desktop images offsite and protected. This doesn’t have to fast disk all.

Need for Speed
– Your replica’s and linked clones need to be fast. Today’s end users are getting SSD in their laptops. Comparing people’s 5 year old computers to VDI are coming to a close. Your virtual desktop needs to deliver the best performance, consistently.

User Data – profile data, user documents, shortcuts and other users errata. Doesn’t need to be on fast disk unless your making use of redirection. If you’re copying data onto the desktop from a repository you don’t want this to be the bottle neck.

The Trash – Page files, swap files and temp files. They take up lots of space so either you need lots of disk or way to dedupe the data.

Applications – An array providing SMB\CIFS share can go along way for distributing your applications to the desktops. This data\IO will land on the linked clones for the most part but an active non-persistent environment can cause a heavy load on your distribution method of choice.

Over the three days at Storage Field Day I cam really close at changing my stance on which makes the best option. Both Pure Storage and Nimbus have some good products but I still think you need disk. If you where only going to go with one array vendor for VDI I would have to go with Flash as a Cache option. To have only one array vendor in your overall solution can go along way with troubleshooting and managing your environment.

User data is going to continue to grow and I believe more of unstructured, hard to dedupe data will be apart of that make up. Also lots of data will be at rest and never be touched after it’s created, I believe this lends well to a flash as cache scenario. Having the disk in the system also helps for replication if you want to use the standby array for other uses during the day. The replicated data can sit on the disk while other systems can use the flash.

All of the Full SSD vendors of their own unique value proposition like Nimbus with there ultra low cost drives and full feature set of offerings and Pure with their ultra safe no virtual machine never UN-aligned again and dedupe upfront features but I still think you need the spinning rust.


  1. Interesting question and the answer goes alot deeper. It really depends on what you are trying to accomplish beyond VDI. What if you just have 500 VDI users and want to consolidate that on a single storage system. That is a very different environemnt that 30,000 VDI users.

    As ever their is no simple answer but in general vendors like Nimble and Starboard Storage http://www.starboardstorage.com (Who I work for) are providing a cost effective means for customers to increase performance by capitalizing on SSD as cache. In the case of Starboard Storage we provide a system that offers the ability to consolidate mixed workloads of unstrustured, virtualized and structured data. In general if customers are looking for consolidation the cache route is going to be the best for them.

  2. Some of this certainly comes down to cost. If the cost differential between all SSD and flash as cache is small then why wouldn’t you go all SSD?

    • Well of course if the cost difference was small you would. The fact is though that most of the business we sell to have 10’s to 100’s of TB of data as we consolidate mixed workloads including unstructured data and they are looking for <$1000/TB fully managed as they grow. They want a balance of $/GB and $/IOP. Not sure where you get 100TB of all SSD Array for under $100k. That is why an intelligent caching solution that can scale with HDD like the Starboard Storage AC72 makes sense.

    • dlessner says:

      If cost was the same I totally agree with that, lost of the SSD arrays don’t come with all the features you might be looking for. Also if SSD folks can make the drives cheaper, why not so the same for hard drives. I know TinTri has plans to do just this, they will eventually even dedupe the hard drive data.

      I was really impressed with what Nimbus had to offer, I think they were a disk company before for 6-7 years so they had lots of IP to work with.

  3. Thank you, really good information here.

    Flash or Cache will always come down to the Cost 🙂

    Nice blog.

    • Deduplication is not a panacea. Mostly people look at it for backup workloads and VMs. With other data it is much less effective. The reality is that with snapshots and pointer based architectures you do not duplicate the backup and virtual machine data in the first place. You just have pure clean space. No need to invest in deduplication. Also if you use SSD cache you can use larger HDD drives behind it. A 3TB 7200 RPM drive is 5 times bigger than a 600GB 15K drive. You are therefore solving the space issues with low cost hardware that has efficient ways of avoiding common duplicates. That will work for all workloads and win everytime over deduplication when deduplication efficiency vary’s greatly by the data on the system.

      Traditional ssytems are terribly inefficient and pushed deduplication as a desirable feature but new storage systems are not. All SSD vendors have to do it to get theoretical $/GB numbers anywhere near what is viable to most customers. Deduplication is not bad, it is just not really that important except in backup systems.

      Don’t believe the hype. If a system gives you all the space you need dmore for less then deduplication is an expensive curiosity.


  1. […] Using Pure Flash or Cache for VDI (Dwayne Lessner) […]

Speak Your Mind