Infinio – Why cache to disk when you can cache to RAM 2

Way back in April I had the opportunity to meet with a start-up called Infinio to participate in a review of their UI.  I had no idea going in what the product was about but was quickly floored by their vision and potential.  Infinio will have their first public demo Thursday at Tech Field Day 9 and will also be a sponsor for the #vBrownBag at VMworld 2013 in San Diego on August 25th – 29th.

Infinio will allow you to create a read cache directly in RAM via a software appliance that will run on your ESXi host like any other VM.  Unlike caching to SSD or flash there is no additional hardware or infrastructure to buy (unless your host are light on RAM) like you would with a flash based array.  Since the VMware community is so awesome, I was curious about RAM availability in most host – enter  The metrics they collect support the idea that most hosts will have more than enough RAM to support the VM.  On average, host are running just under 12 VMs with an average of 115GB of RAM.  I pinged William Lam on Twitter to see if they can add the average amount of RAM per VM, but with 115GB per host, I suspect you have some idle capacity.


Since this is RAM based, you will only be caching read requests since there is no hardware/battery backup like you would have with a Write cache.  For me, every millisecond I can save, every piece of IO I can offload from datastores hosted on my NAS helps provide a more responsive VM environment.

A couple of launch items to note; Infinio will only support NFS at launch.  If you are grappling between iSCSI and NFS for your datastores, add Infinio to the list of ‘pros’ for selecting NFS.  While one of the huge benefits I see is the speed of your RAM, SSD support will also be added in the future so you can add even a single SSD drive to your host to cache even more requests (lets face it even SSD is better than spinning disk for caching, eliminating disk swapping).

While the folks at #TFD9 will be getting a live demo, I am scheduled to get this hands on in my lab next week.  I had hoped to get an early preview of the software and installation process, but my recent job change negatively affected my schedule and access to a lab environment.  Once I am able to catch up with the team at Infinio I will have a follow up post to let you know how they are doing.  From the early work I did see, the product should be a win for shops running their datastores via NFS by the ease of installation, use and increased performance many workloads will see.

What do you think about a product that can cache to RAM?  Are you as excited as I am about its potential?  What questions do you have about Infinio?

  • What advantages would this have over something like ILIO from Atlantis Computing, where they simply build out a ram disk in memory and entirely run VMs from it?

    • Peter Smith

      Hi Rick,

      Apologies for the delayed response. There are some very powerful differences that make Infinio shine:

      -Infinio installs without reboots, plugins, or extra hardware. We’re a software-only solution with some secret sauce that allows us to insert in an active I/O stream with zero disruption to VM workloads. Just download our installer, click through the installer and we take care of the rest.

      -Following on the previous point, you can deploy to a single test host or your entire datacenter in around 30 minutes. No manual steps. No downtime. No hassle.

      -Infinio accelerates all workloads, not just VDI. I was getting 10k IOPS offload on a datastore hosting a compile farm and test automation workload. Saves a bundle on storage.

      -Speaking of money. We’re much lower cost, at just $499 per socket. This includes your first year of support.

      -Our dashboard wraps up all the value/performance metrics for compelling proof of ROI.

      -The big trick is how we pool memory into a globally deduplicated content-based cache that’s visible to all ESX hosts in a cluster. When we detect that content has already been seen in the cluster, we only store it once; furthermore, when a different VM requests the same content (even from a different host in the cluster) we serve the content from the globally visible cache instead of going to disk. This strategy dramatically increases cache hits.

      We GA in Q4 2013. Shoot me an email at if you want to trial our beta.

      Peter Smith
      Director of Product Management
      Infinio Systems, Inc.