[ofw] incompatibility issues between WinOF & (Linux) OFED 1.1

[Tzachi Dar]

Hi,

WinIF is supposed to be compatible with (Linux) OFED 1.1!

Please note however, that ib_send_bw is just a utility that works over
these two stacks and is not compatible between the two versions of the
stacks. This is why your test didn't work.
iperf is an example for an application that works both for windows and
Linux. (works over ipoib and SDP)

In order to have your application working between windows and Linux you
can do one of the following:
1) Write your program directly over the verbs.
2) use IPOIB.
3) Use SDP. In that case you will have to use OFED 1.2.

If you can tell us what your application is trying to do, I can try and
give more help.
(In other words, are you just starting with your application, or is it
already written? Do you care for BW or latency? What is the size of the
messages that you are trying to send? Does your program need to scale to
a big number of computers or not?)

Thanks
Tzachi

> -----Original Message-----
> From: ofw-bounces at lists.openfabrics.org 
> [mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of Ham Nguyen
> Sent: Saturday, May 26, 2007 5:56 AM
> To: ofw at lists.openfabrics.org
> Subject: [ofw] incompatibility issues between WinOF & (Linux) OFED 1.1
> 
> Hi,
> 
> 	?: Does WinOF suppose to be compatible with (Linux) OFED 1.1?
> 
> 	I ran the ib_send_bw test, and found out that the 2 stacks
> 	aren't compatible, so would someone please shed some lights
> 	on this incompatibility, please?
> 
> 	Here're the test results:
> 
>     1.	Linux as client & server: works.
> 	This includes:
> 	   a.	One single Linux running as both server & client.
> 		(This is a silly, but sanity test.)
> 	   b.	Two different Linux running as server & client.
> 
>     2.	Windows as client & server: works.
> 
>     3.	Linux as server.  Windows client failed with:
> 		Completion wth error at client:
> 		Failed status 9: wr_id 1 syndrom 0x81
> 		scnt=1000, ccnt=0
> 
> 	(Linux server does report "remote address ..." so there
> 	seems to be some connection.)
> 
>     4.	Windows as server.  Linux client failed with:
> 		Completion wth error at client:
> 		Failed status 12: wr_id 1 syndrom 0x81
> 		scnt=300, ccnt=0
> 
> 	(Again, Windows server reports connection from client.)
> 
>     5.	If I specify '-c UC' or '-c UD', then the client reports
> 	that it completed 1000 iterations together with the
> 	calculated peak/average bandwidths.  However, both client
> 	and server seem hang, and requiring a Control+C to terminate
> 	the test.
> 
>     6.	*IMPORTANT*: in all cases, the LID of the 
> Windows host is
> 	always shown in reverse-byte order!  E.g. Windows vstat
> 	report a Windows port LID is 0x1234, then:
> 	   a.	Linux reports   "remote address: LID 0x3412"
> 	   b.	Windows reports "local  address: LID 0x3412"
> 	   c.	In case server & client are both Windows,
> 		then both report LIDs in reverse-byte order.
> 	So is it an indication some (Windows?) code didn't do
> 	a host-to-net conversion for LID?
> 
>     7.	I also dual-boot a Windows host as Linux, then
> 	ib_send_bw ran flawlessly.  That proves no problem
> 	with hardware, HCA, fabric, cabling, etc., so the
> 	problem is purely software.
> 
>     8.	Versions:
> 	   a.	Linux: OFED 1.1
> 	   b.	Windows: 1.0.0.614 downloaded from
> 	http://openib.tc.cornell.edu/downloads/binaries/1.0.0.614/
> 
> 
> 	So I'd appreciate anyone who can shed some lights on this issue.
> 
> 
> thanks
> ham
> _______________________________________________
> ofw mailing list
> ofw at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw
>