[rds-devel] Re: Updates to rds-stress

Richard Frank richard.frank at oracle.com
Fri Nov 23 12:57:17 PST 2007


OK - I will apply and test.

Olaf Kirch wrote:
> On Thursday 22 November 2007 17:58, Olaf Kirch wrote:
>   
>> I noticed :-) The code that allocates the buffers for RDMA just allocates
>> them for one thread (rdma_buf[id])...
>>
>> Let me finish merging this, and then lets debug this
>>     
>
> Here it comes...
>
> The code compiles, and it still seems to work for the non-RDMA case.
> Which is not that bad :-)
>
> What does not work yet is RDMA. Using the tree from ofed 1.3 plus some
> version of Rick's barrier patch, plus some patch of mine to make
> it all compile on RHEL5, I manage to freeze the machine every time I
> run rds-stress with RDMA :-)
>
> It appears it is getting stuck in some endless loop in rds_rdma_drop_keys...
>
> The tarball below contains the following patches:
>
> 0001 Create net/rds.h
> 0002 rds-stress: tolerate EINTR
> 0003 Add types and constants related to RDS zerocopy interface
> 0004 rds-stress: add macro for run-time tracing
> 0005 rds-stress: use pfhack.c to discover PF_RDS and SOL_RDS
> 0006 rds-stress: move header building out of send_packet
> 0007 rds-stress: use sendmsg instead of sendto
> 0008 rds-stress: change the way we build the message pattern
> 0009 rds-stress: add support for RDMA interface
>
> Vlad, I think it is safe to merge all patches from this tarball except
> the last one, which adds RDMA support. The RDMA patch itself may still
> see a couple of revisions while I try to get things working.
>
> Olaf
>   



More information about the rds-devel mailing list