On 14/03/2011 03:18, George B. wrote:
bit is how TCP was designed to deal with congestion. Having several seconds worth of traffic sitting in various buffers along the path means that by the time you learn that a path is congested, it might be too late to really do anything about it.
Exactly - but we're not talking about several seconds worth of buffering on these switches. 2M per chassis on a 24 port 10G switch means 1.6ms headroom across 24 high speed ports. That's not a lot.
The only switch which comes close to this is the Force 10 S60, which provides 1.25G of shared buffer space for 24x1G + 4x10G. That would equate to 10 seconds worth of buffering on a single port, assuming pathological lab conditions and lab configuration (i.e. much less in real life). However it's worth pointing out that F10 specifically aim it at streaming and other applications where latency is much less of an issue than packet drops.
IOW, choose your equipment to match your requirements. > Google "bufferbloat". He hasn't done the world's best
job in actually communicating it in a way most people can understand, but he is dead on the money. Bigger buffers is actually probably a bad thing.
We're talking apples and oranges here. Big buffers on a decent quality switch are not the same as ridiculous buffers on a trashy CPE device.
On 14/03/2011 03:27, George B. wrote:
Oh, about price ... the 7124's are about US$13,000 list. Don't expect as much of a discount off list as they price their stuff pretty low anyway. The ER optics are about $6K I think, SR optics are under $600
Last I heard, Arista vendor-locks their transceiver ports - although that was a couple of years ago and maybe things have changed since then. This means that if you're using third party SFP+ (where you might expect to pay e.g. €120 per SR transceiver in small quantities), you will need to get them vendor coded.
Nick _______________________________________________ foundry-nsp mailing list [email protected] http://puck.nether.net/mailman/listinfo/foundry-nsp
