On 12/8/11 7:58 AM, Jon Lewis wrote: > On Thu, 8 Dec 2011, Gert Doering wrote: > >> The best choice? Don't use 6148-GE-TX modules. They are fundamentally >> broken (8 ports share one ASIC with a single-GE uplink, one port that's >> "full" will block out the other 7 ports, ...). It's even worse if >> you use them for 100M links, because a saturated 100M link will eat >> all the buffers from the other 7 ports on the same ASIC, causing RTT >> jumps on these other ports. > > Just to clarify, as I understand it, this (shared buffers) is an issue > with the 6148-GE-TX, but not with the 6148A-GE-TX, which according to > cisco documentation has much larger buffers and they're per port, not > shared by the ports in each 8 port group. The 6148A-GE-TX is still 8:1 > oversubscribed, so it's a poor choice if you have a need for lots of > 1000baseT ports handling much traffic, but at least it has nice per-port > buffers. I suspect if most of the ports are used as 100baseT, and you > have the occasional 1000baseT port that might carry just a little more > than 100mbit/s, it should do fine. >
And the 6148A supports jumbo frames, if that matters. But yeah, it has 2.6MB per port buffers instead of 1MB shared across 8 ports. ~Seth _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
