I'm genuinely curious if I'm missing something, but I don't see the point of 25 GbE because 40 GbE was already established as a mature technology by the time they rolled it out. I can purchase 40 GbE ConnectX-3 gear all day long for around $40/card. Likewise, 40 GbE switching equipment is equally as cheap, I just picked up a Mellanox SX1036 for around $200.
The sequence of progression doesn't make sense ether, 1, 2.5, 5, 10, 25, 40, 50, 100, 200, 400. What about all the gaps between 50 and 400? To me it seems like it would have made more sense to do periodic multiples of 40... 80, 120, 160, 200, 240, 280, 320, 360, 400. This would have allowed for linear scaling of fanout cables. i.g 80 GbE = 2x 40 GbE, 160 GbE = 2x 80 GbE or 4x 40 GbE.
The transition between 50 and 100 is too large, the Linux software stack is struggling to push packets faster then 70 Gb/s... the ability to push packets in software vs hardware are highly divergent with regards to realized speed. If we would have set 80 GbE as the next target instead of 100 GbE the hardware vendors could have pushed out new equipment out the door faster and cheaper.
This all strikes me as not being well thought out in advanced.
No comments:
Post a Comment