IDC has been tracking the Open Compute movement very closely since it's inception 18 months ago due to the involvement of very large IT consumers such as:
- Goldman Sachs
Leading up to the Open Compute Summit there was lots of speculation regarding what would be discussed since the agenda wasn't released until about a week before the event.
The only hint of an emphasis on storage as during registration when there as an option to attend a technical storage track.
Even days before summit, no formal agenda of the technical tracks was published building to a sense of excitement as to what was to come.
It wasn't until I was actually at the conference did I really get the sense that storage was going to be a major theme with the Knox project transitioning to Open Vault.
Disaggregation of Hardware Provides Intelligence within the Open Vault
A key theme of the Open Compute Summit IV was the desegregation of infrastructure hardware and technologies. Open Vault was no exception with the ability to choose between simple SAS expansion or compute board in the SAS expansion slot in the rear of the chassis.
What was more interesting was the size and amount of real estate the specification provided for the board. Probably less than 50 percent of the board was being utilised from any implementation, leaving lots of room for future expansion and interesting applications.
During the storage technical track, Calxeda presented for about an hour and outlined several different connectivity options for their implementation:
- 4 x 10Gbe
- 2 x 1Gbe
- 4 x QSFP
More interesting than the plethora of configurations was Calxeda's idea of a 2D Torus fabric that networked together 40 Open Vault chassis without a dedicated networking switch. Not only does that reduce networking costs, but due to the Calxeda design the networking fabric doesn't access compute resource to switch networking traffic.
SSD & Open Vault
The Open Compute Summit's vendor area was relatively small with less than 20 vendors. I was particularly interested at the Wiwynn booth showed an Open Vault configuration with 60 SSD drives!
Open Vault specification is for 30 drives (with 2 trays at 15 drives per tray ), so how is this possible? For serviceability, Open Vault does not require any kind of form be attached to the drive before insertion into chassis. Simply open the drive container in the chassis, drop in drive, close container (which pushes the drive into the SAS connector), & voilÃ , you're finished.
There is extra space in the drive containers, so Wiwynn was showcasing an adapter that could hold two SSDs and plug into the SAS adapter. Therefore it is possible to house 60 SSD drive in an Open Vault unit, but it is unlikely to be a practical application.
Three drive containers share a 6GB/s SAS connection, so (in the Wiwynn configuration) with 6 SSD per 6GB/s SAS channel would surely be saturated quickly. However, it does open the door for future design iterations to include SSD either through simply increasing SAS channels or even hybrid SSD/HDD implementations.
Knox & Open Vault is Real
Beyond the hype & paper specification the Knox / Open Vault design is definitely real & ready to ship. Several vendors were showcasing their Open Vault product such as Hyve & Wiwynn.
The Open Vault design provides for the SAS expander card to be swapped for a board with more intelligence, which two option were being showcased: Intel & Calxeda. It seemed that Calxeda was receiving lots of press for being the 1st ARM based option providing competition to Intel in this nascent market.
Opportunity: Software, software, software
An interesting question came to mind while at the Open Compute Summit. Could someone become a storage array vendor without ever developing or shipping hardware? It is totally conceivable that the customer could purchase whatever iteration of Open Vault & drive hardware from one vendor and then download the intelligent array software from another vendor.
Furthermore, let's say the client wasn't happy with one software vendor, they could potentially change vendors by installing another vendor's storage software onto the platform. Even from a Proof of Concept (POC) perspective it could alleviate many vendor and client pains.
The client knows they want Open Vault hardware, so they purchase the hardware independently. Then the client could POC the software at relatively no cost to the client or vendor. Disaggregating hardware technology is the first step, but disaggregating hardware and software purchasing would be market changing.
Posted by Dan Iacono, Research Director, IDC