network speed on t2000
in [Solaris]
|
From: jdh13 on 8 Feb 2007 11:15 Casper H.S. Dik wrote: > jdh13 <jdh13(a)free.fr> writes: > >> and: >> t2000-root% ifconfig e1000g3 mtu 9600 >> ifconfig: setifmtu: SIOCSLIFMTU: e1000g3: Invalid argument >> so if i modify e1000g.conf, i have to reboot! >> why this driver is such configured? > > update_drv -f e1000g did not cause the changes to be picked up? > t2000-root% update_drv -f e1000g Cannot unload module: e1000g Will be unloaded upon reboot. Forcing update of e1000g.conf. it seems it don't work
From: Rick Jones on 8 Feb 2007 14:12
jdh13 <jdh13(a)free.fr> wrote: > hi all, > when transfering data between two t2000, i noticed a bad speed, so i did > some tests: > t2000_A e1000g0 --- switch HP 2824 --- e1000g0 t2000_B > t2000_A e1000g3 ---------------------- e1000g3 t2000_B > i'm using netperf, and in all cases, i got around 300 Mbit/s > if i run 3 netperf in parallel, each one displays around 100 Mbit/s > Has anyone noticed it on this kind of machine? All are Solaris 10 U3 > with latest patches I can say that I have a lot of recentish experience with netperf performance on the T2000 (U2 mostly for the core, a little with U3 and 10G), but a strict interpretation of Clause 5f of the Solaris EULA suggests I am not at liberty to discuss it in public without Sun's prior consent. I will though ask some questions: *) Do you know to which strands interrupts from the NIC(s) are going? *) Do you know on which strands the netperf/netserver processes are running? *) Did you add any --enable-mumble options when you did the configure step? *) What command line options to netperf are you using? *) Have you also run netperf TCP_RR tests? *) Have you noticed some of the network related tunes Sun listed in their SPECweb2005 disclosures on the T2000? rick jones -- oxymoron n, Hummer H2 with California Save Our Coasts and Oceans plates these opinions are mine, all mine; HP might not want them anyway... :) feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH... |