Initial Musing about Coho Data Scale Out Networking

Initial Musing about Coho Data Scale Out Networking

March 7, 2014 2 By Eric Shanks

COHOLogo2

I was fortunate enough to have spent some time at the Coho Data headquarters this week for the announcement that their new product, DataStream 1000, is now generally available.

The announcement was made at the GestaltIT Virtualization Field Day 3, which was streamed live and the recordings can be found online.

 

 

All travel expenses and incidentals were paid for by Gestalt IT to attend Virtual Field Day 3. This was the only compensation given.

If you’re not familiar with them already, Coho Data has developed a very flexible scale-out architecture for a storage platform.  If you’re running out of storage or performance, you add a new node and move on.  Each one of these Datastream chassis has two controllers called “microarrays” that manage the data.

Each microarray runs a hypervisor that virtualizes access to:

  • two Intel 910 Series 800GB PCIe Flash Cards
  • six 3TB SATA disks
  • two 10Gb NICs
  • two Intel Xeon E5-2620 Processors

MicroArray

coho-switch-sdn

 

As with any new storage platform the performance is always something people want to hear about.  I’ll show you the advertised performance metrics with the caveat that I’ve not performed any tests on this myself.  What I do like about this graphic is that it shows what the scale out looks like.

CohoScaleOut

 

My focus is to look at how the NFS microarrays scale out with a single IP Address.  The thing that REALLY caught my attention during Andrew Warfield‘s presentation was that each of the 10Gb Nics have the same IP Address.  What!???

If we had two datastream 1000’s there would be 8 10Gb NICS that all have the same IP address, as shown below in my crude diagram.

COHO1

It’ took a bit for me to understand exactly how this solution was made to work, but the great thing about Tech Field Day is that you are around a variety of great minds in different disciplines and Tom Hollingsworth was able to fill in some blanks for me.  Coho is using an Arista 7050 switch which is an OpenFlow switch.  The magic here is that this switch is a Software Defined Switch in the sense that the data plane is taking instructions from a control plane that can manage the layer 2 traffic.

Let’s walk through a quick example.

The host in the diagram below will submit an NFS request over the network to the storage array at the 1.1.1.1 IP address.  The switch knows that one of the ports with a 1.1.1.1 IP address needs to get the traffic, but NOT all of them.  The Control Plane of the switch can decide which port is least heavily utilized and push an entry into the TCAM of the switch where then the data plane will forward the layer 2 frame out the desired port.  This is how Coho can have a single IP Address across all of their devices so adding a new chassis to the configuration is no big deal.

COHO2

 

Coho Data has a really unique way of handling storage and seems as though they can do it very quickly by utilizing PCIe Flash as well as low cost disks.  It’s also very helpful to be able to right size your environment and be able to scale to the size of your requirements.

Please tune in and keep an eye on this company.  It’s a really interesting design and a bit different solution than you’ve seen from other vendors.

If you want to read more, please check out the following related posts, or contact Coho Data.  I’m sure they’d love to talk to you.

Tech Field Day – Coho Data Presentation at Virtualization Field Day 3

 

Chris Wahl – Coho Data Unveils Hybrid Flash Storage Combined With Software-Defined Networking

Eric Wright – Tech Field Day VFD3 – Coho Data – Coho-ly Moly this is a cool product!

Jeff Wilson – Kicking it with Coho