So for the longest time we have always preached that we do not process switch frames because obviously it takes the cpu more time to decide what to do with it as opposed to standards based code running on chips that know specifically what to do with that data and do it quickly. As we move away from standards and towards "....oh dont worry, we have an API", are we potentially putting ourselves into bottleneck positions until regular processors beef up? Otherwise, why are we not just using asic-less API driven switches with todays processors to network traffic? Is this an academic topic being discussed right now? In the past, the vendors would wait, argue and bicker for the features they want in the protocol and as soon as it was ratified, they started making chips for it and selling their products against competitors. But now it just seems like everyone makes an API and haphazardly makes statements like "oh we have an API, you don't have to worry about that" thinking this is a cure all. Am I wrong in my concern of having too many functions performed by an API and expecting the same performance that you may be used to?
No comments:
Post a Comment