Bechtolsheim races Arista to zero latency
Every microsecond counts for Bechtolsheim's bolshy switches
This is where he says low, low latency is a big advantage. The 7124 has substantially less latency than the 5-microsecond Nexus 5000, the 15-microsecond Nexus 7000 and the 20-plus microsecond latency of the Catalyst 6500. Cisco is the network behemoth Arista has to beat, and it aims to whip Cisco to death with superior speeds and feeds.
Bechtolsheim has a latency roadmap slide bearing a single arrow pointing downwards - no numbers or timescale though.
Naturally, Arista has a virtualisation strategy. The idea is that administration should be unified at the vCenter level, with the network reacting in real time to the server farm virtualisation administrator's control inputs.
There is a VMware virtual machine instance of Arista's EOS operating system, vEOS, which provides topology discovery and virtualisation. There is also VMTracer which has direct integration with VMware APIs and provides control and data linkage with VMware - all versions of VMware.
Arista wants to plug into the heart of the virtualised data centre and out-perform everybody and provide the best cost/efficacy in the networking business. Today it is focusing on 10gigE, although: "Almost no server has 10 gig on the motherboard." The computer scientist continues: "That should go away a year from now. There will be a much more rapid conversion to 10 gig when that happens. By 2014, 90 per cent of servers should have 10 gig."
Forty gig is on its way. Bechtolsheim said: "We use merchant silicon and every vendor has 40 gig on their next chip, with 100 gig after that."
Bechtolsheim is a low-latency man in everything he does. He spent half his presentation slot talking at high speed about the slides, getting through a slide a minute in some sections of his pitch. He's done workstations, he's done storage, Arista is ready to be bought - well, it appears that way - and the mercurial Bechtolsheim is probably already involved in his next venture. ®
Sponsored: Hyper-scale data management