OCP on Steroids, or Just Done Right?
Last week I got to tour Penguin Computing’s facility. As one approaches one gets the distinct impression that something is different. Hayward is all empty buildings and parking lots, but as I rounded the final corner it became evident that I either had to double park or walk a mile. There was excitement here, lots of coming and going, much goings on.
On the tour I was shown many exciting things. One of their clients asked them to push both efficiency and density to a whole new level. Enter the 1-OU, liquid-cooled, OCP-compliant server!
I say compliant because it fits into a OCP rack, holds our core philosophy at heart, but it has some significant differences. One big one is power management. The PDB is intelligent and can orchestrate load in order to avoid a thermal event locally, as well as be orchestrated remotely to avoid exceeding designed rack kW draw. This fine-tuned capability allows their clients to run right up to the edge and still have the knowledge that the system will not fail.
Other differences are Penguin has chosen much more normal motherboards. Meaning you can kiss the SOL management goodbye and use a normal enterprise style management tool for provisioning, monitoring and debug. Whereas that deviates from the raw OCP method, in that there are some components on the motherboard that are hardly ever used, it has the upside of enabling any enterprise to use Penguin Computing OCP sleds (liquid cooled or not) in their Object stores, Private or hybrid clouds, etc.
Penguin Computing optimizes the customer relationship too, acting a bit like an ODM, but providing full-stack solutions tailored to the needs of the client, a breath of fresh air in this day and age of manufacturers telling us what we want. Penguin Computing is truly a collaborative environment to work in. I predict they will need to add another facility soon, as the long-pole in lead time is the burn-in and validation testing. They have 2 aisles dedicated to just shaking out the misbehaving units, and by the look of the assembly line, I am betting they will be needing 2 more inside a year.
It is my recommendation that if you are planning on doing a high efficiency cluster deployment (you already did the math on the savings that brings, if not call me.) that Penguin Computing be on the list of potential partners to make your deployment a success.