From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Wed Jun 15 2011 - 09:37:40 CDT
Well I have a TCP version, and yes tcp congestion control highspeed made a
big deference in scaling behavior on my cluster. Maybe try out. What is the
ping time of your cluster?
--- c03.cluster ping statistics ---
36 packets transmitted, 36 received, 0% packet loss, time 35000ms
rtt min/avg/max/mdev = 0.074/0.089/0.117/0.019 ms
Von: Thomas Albers [mailto:talbers_at_binghamton.edu]
Gesendet: Mittwoch, 15. Juni 2011 16:10
An: Norman Geist
Betreff: Re: AW: namd-l: NAMD crashes with OpenMPI/OpenMX
> Which TCP congestion control
> algorithm do you use (reno or cubic) -> try highspeed (sysctl -a/-w),
Cubic, but should this matter at all, considering that the precompiled
binary uses UDP?
> what kind of gigabit network adapter do you use, is it onboard and if,
> bus connection does it have (pci or pcie) or is it a plugged pcie nic?
Intel PCI-E card w/ 82574 chip.
> Another thing is the capacity of the switch, have you tried to run two
> jobs simultaneously on 12 cores for example?
According to the datasheet the capacity of the switch is 16 Gbit/s,
and testing reveals that send and receive rates are close to
theoretical, even with all for computers sending and receiving
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:24:05 CST