[c-nsp] ES20 throughput in the weeds?

[Jason Lixfeld]

We've got an 7600-ES20-GE3CXL HW 1.2 FW 12.2(33r)SRB SW 12.2(33)SRC4 in a 7609/Sup720 3BXL chassis.  We ran some performance tests for a customer, and we were quite appalled by the results.  We're using Exfo test sets to run RFC2544 patterns between two ports.

We're using 7 frame sizes; 64, 128, 256, 512, 1024, 1280, 1518.

When we look at the throughput results, we see this:

Frame Size	TX-to-RX - Layer 1-2-3 (Mbps)
64		449.197861
128		538.181818
256		560.97561
512		974.358974
1024		956.043956
1280		1000
1518		994.825356

Now if we do that same test on another card, say a WS-X6724-SFP, we get very different results:

Frame Size	TX-to-RX - Layer 1-2-3 (Mbps)
64		1000
128		1000
256		1000
512		1000
1024		1000
1280		1000
1518		1000

I've tried the same test on a Juniper EX4200, and an ME3400E-12CS and all the results are identical to the WS-X6724-SFP.  The ES20 seems to be the anomaly here.

I know that processing smaller packets is generally much more taxing than processing large packets, so I didn't really expect to see line rate on any of the tests that were run, but considering the 6724 reported what seem to be line rate results vs. the ES20, I can't help but to wonder whether I've got a bad card or something.