local flash storage for VDI

[dani_bosch]

Hello,

According to the following white paper:

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns944/whitepaper_C22-726648.pdf

a good alternative when implementing VDI is using local mlc flash cards in the UCS blades, containing highly-accessed assets, like Master Image...what I don't understand is, in case we have more than one blade (99% of the cases) how do I share this storage across the blades? In case I don't share storage, I'll have to copy the same master images across all the flash cards in all the blades....rather stupid.

The white paper doesn't specify how did they face this issue...

[Robert Burns]

Dani,

This comes down to the way FusionIO drives operate.  They will cache a copy of the master image upon first access, then all subsequent image deployment will be exterememly fast.  Even the initial master image deployment will be far quicker due to the IODrive's caching enginer & agent interface.  All blades would need the FusionIO adapters, but your Master Images would still reside on the backend shared storage.  The IODrive technology will greatly speed up access & deployment of images.

Regards,

Robert

[dani_bosch]

So again, bottom line is, I'll have to purchase a flash card for each and every single blade....which will be rather costly in an implementation of let's say 1000 VDI's in 10 blades....I'll have to purchase 10 flash cards and worst thing is that all of them will have the same content....so if my Master Images need let's say 100 GB, I'll need 100GB times 10 blades...

I think it makes much more sense to have this flash memory on the shared storage, thus needing purely the 100GB mentioned before...

Don't you think so?

[Robert Burns]

I had the same thought originally when I was learning about FusionIO technology, but the way in which they implement caching has a tremendous impact on the bottom line performance of each host.   It's far more efficient than simple shared SSD storage could ever offer.

You could have SSD flash HDDs on your backend shared storage array, but you'd still be limited by the shared caching capabilities of the array, storage network infrastructure, storage protocols and adaptors.  The main benefit of FusionIO is blazing fast data transfer & caching that are far lower latency than traditional SSD.  You will be hard to reach anywhere near the IOPS fusionIO can provide vs. shared SSDs.

Here's an internesting overview of FusioIO's implementaiton of Flash and how it differs from traditional SSD.  The point of the video is that FusionIO implements NAND flash (SSD) if a far more efficent way than anyone else.

https://www.youtube.com/watch?v=w-_Hr5f7QHw

If I find any other useful resources I'll update this thread.

Robert

[dani_bosch]

I understand that from an IOPS perspective this technology is more powerful than shared flash storage array....but I still think that's not a business case for VDI, since having a flash card in every single blade must be insanely expensive...in my opinion price/performance ratio for VDI would be better achieved in shared flash storage arrays.

Critical databases could definitely be a business case for this,as they don't need dozens of blades....but not VDI, in my opinion.

Don't you think so?

[dani_bosch]

Robert, would you agree with last thought?

Thanks,