4 ms·
> packets must arrive within ~40us (that's us, not ms) of their allotted time, something very difficult to reach in software. (I will be talking about this at
by paulmd 9y ago
> packets must arrive within ~40us (that's us, not ms) of their allotted time, something very difficult to reach in software. (I will be talking about this at demuxed next week...)
Maybe in Ethernet-land, but this is trivially achievable (actually exceeded by more than an order of magnitude) with off-the-shelf Infiniband gear that is available for peanuts these days. Actually it's close to two orders of magnitude if you want to shell out for the latest gear.
https://en.wikipedia.org/wiki/InfiniBand#Performance https://en.wikipedia.org/wiki/InfiniBand#Performance
- amelius 9y agoBut does it actually provide a real-time guarantee?
- paulmd 9y agoYes. Adapter latency is guaranteed to be 0.5-5 usec depending on generation (later revisions are faster+lower latency). Switches add about 170 nsec per hop, plus whatever speed-of-light delays your cables add. 1 usec end-to-end is realistic for modern gear. http://www.hpcadvisorycouncil.com/events/2014/swiss-workshop/presos/Day_1/1_Mellanox.pdf http://www.hpcadvisorycouncil.com/events/2014/swiss-workshop... Of course with IP-over-IB you also need to factor in whatever latency is added from your network stack, but Infiniband also supports RDMA so IB-aware applications can essentially bypass this for high-performance or hard-real-time usecases. Basically, Infiniband takes the network out of the equation here. The bottleneck then becomes how precisely you can time the software.
- kierank 9y agoYes, easily possible with Infiniband but quite complex with Ethernet.