There has been a lot of activity over the past couple of months with respect to the acceptance of hardware accelerators within NFV. I can still recall back this time last year when accelerators were a four letter word within NFV. Now, we’re talking about creating entire shelves of them. For me, this notion of disaggregation is much more important that what others think about when they talk of other initiatives like OCP (Open Compute Project).
The reason for this is simple… we need accelerators in our various platforms. But we also need to be able to get better amortization of such additional hardware. Simply burdening every server blade with such technology is too costly. This drives up CAPEX considerably. And it simply contributes to the ongoing problem of over provisioning.
Instead, with disaggregation, we will some day have the ability to dynamically bind an accelerator to an application on an as needed basis. And this could be reconciled back to an customer SLA or some kind of equivalent business justification. And this is important. When we can rationalize and properly account for an additional business expense, we can understand how to ensure that we’re properly charging for it. So, those that truly need the functionality, not just everyone, can pay for the capability.
I’m happy to say that this subject was accepted into section 4.4.2 of the Compute Domain GS document, last month in the Cambridge face to face INF meeting.
Now comes the challenge of creating an ecosystem of accelerator shelves, which can be used as NFVI (NFV Infrastructure) gear. This is the fun part of NFV!