> Don't forget they are absurdly under-buffered (16MB per card, compared > to 256MB for 6708), and you can easily cause head of line blocking with > certain traffic profiles. If you want to run anywhere close to line rate > on them you need to monitor for drops or overruns and be prepared to > play the port shuffle game to find an arrangement that works. Passing a > lot of traffic within the same fabric channel (from port 1<->2, or > 3<->4) is the biggest sin, it will start dropping at 7 Gbps.
Well that's wonderfully comforting. Though I really probably only need two ports anyway - ring-in and ring-out. Maybe not so bad. I'd consider a 720-VS-10G head if I had some confidence that those two ports on the sup were actually connected to the fabric. I don't really need to run line rate - this is more about latency and burst capacity than sustained throughput. I have loads that burst from 0 to 500Mb/sec (then back) in nothing flat, and multiple of those may run through the wire at the same time. Or not. Someone pointed out that the X2 and SFP+ xcvrs don't have much punch, and I'm going to be shooting 20-30km through passive MUXes. So that might matter. (This is a bit of a roll-yer-own local metro NYC ring, which I'm doing because I can get the wave for not much more than I'd pay for the switched gig.) _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
