[ofa-general] Compiling SRPT

Vu Pham vu at mellanox.com
Wed Jul 11 16:57:29 PDT 2007


Stanley Sufficool wrote:
> Do you have any reservations that the WinIB (Mellanox) SRP initiators
> will not work with SRPT? 
>   
There are two version of SRPT: ofed/gen2 srpt and ibgold srpt
You are working with ofed/gen2 srpt now

WinIB srp initiator works well with ibgold srpt
I quickly test WinIB srp intiator with ofed/gen2 srpt. It sees the 
target but not its lun - some debugs are required

ibgold srpt only work with suse/sles 9, rhel 4. It does not work with 
rhel 5 or sles10 or vanilla kernel > 2.6.11

Do you have any restriction on kernel, version of IB driver/srpt driver 
on the target machine?

> If there is any doubt, I need to know so that I can fall back to iSCSI
> over IPoIB (iSIPIB??? ;) )  . This has lots more overhead, but it's a
> sure bet until this can be worked out.
>
> On Tue, 2007-07-10 at 09:55 -0700, Vu Pham wrote:
>
>   
>>> Added a new wiki page based on Vu Pham's readme and issues with recent 
>>> kernels. I hope to keep it current as I get our targets up and running.
>>>
>>>       
>> Thanks for doing this.
>> Please use the latest readme from this link - 
>> http://mellanox.com/pdf/products/software/Gen2_SRPT_README.txt
>>
>>
>>     
>>> http://wiki.openfabrics.org/tiki-index.php?page=SRPT+Installation 
>>> <https://wiki.openfabrics.org/tiki-index.php?page=SRPT+Installation>
>>>
>>> WinIB initiators --> Gentoo Linux SRP Target.
>>>
>>>       
>> I mainly test linux initiators with gen2 srp-target. I have 
>> not tested win srp initiator with the target.
>>
>>     
>>> Anything wrong with the above approach, I would be interested in a best 
>>> practices if there is one. I saw a CentOS target post, is this more 
>>> stable or better performing?
>>>       
>> There is no difference when you run the same srp target / 
>> scst codes in CentOS or RH/SuSe linux distributions. The 
>> storage back-end will determine the performance
>>
>> -vu
>>
>>     
>>> Thanks.
>>>
>>> On Mon, 2007-07-09 at 09:55 -0700, Vu Pham wrote:
>>>       
>>>> Stanley Sufficool wrote:
>>>>         
>>>>>   Compiling on kernel 2.6.21-rc6 from kernel.org Torvald's branch
>>>>>
>>>>> Got the latest srpt from the git repository on OpenFabrics and had the 
>>>>> following issues.
>>>>>
>>>>> ib_srpt.c    Line 1997, missing second argument, should be?   
>>>>> sdev->scst_tgt = scst_register(tp, NULL);
>>>>>
>>>>>           
>>>> Yes. You need the change if you test with top of scst svn 
>>>> trunk (or from version 0.9.6-pre2)
>>>> If you test with scst before 0.9.6-pre2 (ie. version <= 
>>>> 0.9.6-pre1) you don't need the second argument for 
>>>> scst_register()
>>>>
>>>>
>>>>         
>>>>> SCST was built successfully after fixing an issue in scst_vdisk.c 
>>>>> (missing #include <linux/sched.h>)
>>>>>           
>>>> I tested with 2.6.20.x - I have not tested with 2.6.21-rcXX 
>>>> - you should send the patch to scst devel
>>>>
>>>>         
>>>>> Just thought this would be nice to have documented, took me half a day 
>>>>> to track down as a novice in C programming.
>>>>>
>>>>>           
>>>> there is *lean and mean* srpt's README in srpt_inc
>>>> SCST also has some document
>>>> You can add some wiki/notes for the problems in openfabrics 
>>>> wiki page https://wiki.openfabrics.org/tiki-index.php
>>>>
>>>> -vu
>>>>
>>>>         
>>>>> ------------------------------------------------------------------------
>>>>>
>>>>> _______________________________________________
>>>>> general mailing list
>>>>> general at lists.openfabrics.org <mailto: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
>>>>>           
>
>   




More information about the general mailing list