[ofa-general] MT25204 versus MT25208

[Sagi Rotem]

You can use the ib_write_bw instead ,this newer test has more options.
Test generally cant "guess" automatically the right port, what would you
like to use as defaults if both ports are up ? If you have more than 1
card ? Etc ...
So for the single case of cable connected only to port 2 u can use the
flag Dotan suggested.


-----Original Message-----
From: general-bounces at lists.openfabrics.org
[mailto:general-bounces at lists.openfabrics.org] On Behalf Of Dotan Barak
Sent: Thursday, January 17, 2008 5:56 PM
To: Bart Van Assche
Cc: Openib-General
Subject: Re: [ofa-general] MT25204 versus MT25208

Bart Van Assche wrote:
> On Jan 17, 2008 4:38 PM, Dotan Barak <dotanb at dev.mellanox.co.il>
wrote:
>   
>> Bart Van Assche wrote:
>>     
>>> Four hosts are connected through an InfiniBand network, two with a
>>> MT25204 HCA and two with a MT25208 HCA. All four hosts communicate 
>>> fine via IPoIB, and iperf shows the expected results. Until now I 
>>> only succeeded to run ib_rdma_bw between the MT25204 interfaces. 
>>> When I run ib_rdma_bw on one of the MT25208 interfaces a strange 
>>> error message is printed. Any idea how I can fix this ?
>>>
>>> MT25208 to MT25204:
>>> root at 192.168.102.5:~# ib_rdma_bw
>>> 6468: | port=18515 | ib_port=1 | size=65536 | tx_depth=100 | 
>>> iters=1000 | duplex=0 | cma=0 | server read: Success
>>> 6468:pp_server_exch_dest: 0/45 Couldn't read remote address 
>>> root at 192.168.102.100:~# ib_rdma_bw 192.168.102.5
>>> 6466: | port=18515 | ib_port=1 | size=65536 | tx_depth=100 | 
>>> iters=1000 | duplex=0 | cma=0 |
>>> 6466:main: Local lid 0x0 detected. Is an SM running?
>>>
>>>       
>> What is the output of ibv_devinfo in both of those devices?
>>
>> (maybe you need to specify the IB port value or the device name?)
>>     
>
> Thanks, after specifying -i 2 as an argument to ib_rdma_bw the test 
> runs fine.
Happy to hear this ...
> Would it be possible to let the ib_rdma_bw program determine this 
> parameter automatically ? In my setup the InfiniBand cable is 
> connected to the second port of the MT25208 interfaces.
>   
I think that for every IB application/test one should specify the exact
device name + IB port.
(this is what I'm doing in our nightly regression)

I'm not the maintainer of the test, but do you suggest that this
benchmark will work on the first IB port which is in ACTIVE state?


thanks
Dotan
_______________________________________________
general mailing list
general at lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit
http://openib.org/mailman/listinfo/openib-general