perftest question

perftest question

[ib-x2spCj9RiN0z5UmgcLIfJQ]

Just downloaded a snapshot of the perftests.  In the README under  
"Test Descriptions" it says:

rdma_bw.c 	streaming BW test with RDMA write transactions

and

write_bw.c 	BW test with RDMA write transactions

What is the difference between rdma_bw and write_bw?  I see that the  
rdma_bw description contains "streaming" where as write_bw description  
does not...  but I am not sure I understand what that means?

E

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perftest question

[Ido Shamai]

Those are actually the same tests, from an IB point of view.

the difference between them is that ib_write_bw passes the arguments that need to be known to other side ( LID or GID, Qp num, PSN, RKey and the address of the writing buffer) before post_sending,

through TCP connection (using sockets) and the rdma_bw uses the rdma_cm module , which sends those arguments through the IB wire and the SM.

rdma_cm is not updated with all the features ib_write_bw have. I will work on this test shortly

Ido


On 5/17/2011 10:19 PM, ib-x2spCj9RiN0z5UmgcLIfJQ@public.gmane.org wrote:
> Just downloaded a snapshot of the perftests.  In the README under "Test Descriptions" it says:
>
> rdma_bw.c     streaming BW test with RDMA write transactions
>
> and
>
> write_bw.c     BW test with RDMA write transactions
>
> What is the difference between rdma_bw and write_bw?  I see that the rdma_bw description contains "streaming" where as write_bw description does not...  but I am not sure I understand what that means?
>
> E
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perftest question

[ib-x2spCj9RiN0z5UmgcLIfJQ]

Ok.  Thanks for the information.

Is there any particular advantage of using CM over TCP for this  
information (LID or GID, Qp num, PSN, RKey and the address of the  
writing buffer)?

Is ib_write_bw an example of RDMA write / RD?

E


Quoting Ido Shamai <idos-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>:

> Those are actually the same tests, from an IB point of view.
>
> the difference between them is that ib_write_bw passes the arguments
> that need to be known to other side ( LID or GID, Qp num, PSN, RKey and
> the address of the writing buffer) before post_sending,
>
> through TCP connection (using sockets) and the rdma_bw uses the rdma_cm
> module , which sends those arguments through the IB wire and the SM.
>
> rdma_cm is not updated with all the features ib_write_bw have. I will
> work on this test shortly
>
> Ido
>
>
> On 5/17/2011 10:19 PM, ib-x2spCj9RiN0z5UmgcLIfJQ@public.gmane.org wrote:
>> Just downloaded a snapshot of the perftests.  In the README under   
>> "Test Descriptions" it says:
>>
>> rdma_bw.c     streaming BW test with RDMA write transactions
>>
>> and
>>
>> write_bw.c     BW test with RDMA write transactions
>>
>> What is the difference between rdma_bw and write_bw?  I see that   
>> the rdma_bw description contains "streaming" where as write_bw   
>> description does not...  but I am not sure I understand what that   
>> means?
>>
>> E
>>
>> -- 
>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>



--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perftest question

[Steve Wise]

On 05/17/2011 02:41 PM, Ido Shamai wrote:
> Those are actually the same tests, from an IB point of view.
>
> the difference between them is that ib_write_bw passes the arguments that need to be known to other side ( LID or GID, 
> Qp num, PSN, RKey and the address of the writing buffer) before post_sending,
>
> through TCP connection (using sockets) and the rdma_bw uses the rdma_cm module , which sends those arguments through 
> the IB wire and the SM.
>
> rdma_cm is not updated with all the features ib_write_bw have. I will work on this test shortly
>
> Ido
>

In addition, rdma_bw and rdma_lat work over iWARP when used with the --cma option.  The ib_* versions do not since they 
don't have a option to use the rdma_cm, which is required for setting up iWARP connections.

Steve.


>
> On 5/17/2011 10:19 PM, ib-x2spCj9RiN0z5UmgcLIfJQ@public.gmane.org wrote:
>> Just downloaded a snapshot of the perftests.  In the README under "Test Descriptions" it says:
>>
>> rdma_bw.c     streaming BW test with RDMA write transactions
>>
>> and
>>
>> write_bw.c     BW test with RDMA write transactions
>>
>> What is the difference between rdma_bw and write_bw?  I see that the rdma_bw description contains "streaming" where 
>> as write_bw description does not...  but I am not sure I understand what that means?
>>
>> E
>>
>> -- 
>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: perftest question

[Ido Shamai]

On 5/17/2011 10:53 PM, ib-x2spCj9RiN0z5UmgcLIfJQ@public.gmane.org wrote:
> Ok.  Thanks for the information.
>
> Is there any particular advantage of using CM over TCP for this information (LID or GID, Qp num, PSN, RKey and the address of the writing buffer)?
>
> Is ib_write_bw an example of RDMA write / RD?
>
> E
>
Except of the iWarp support Steve mentioned I think the rdma_cm has an advantage because it does not required an additional Ethernet connection

to the machines. the test has all it's need.

ib_write_bw is an excellent example of RDMA write with a lot of feature. it can run on RC and UC (no RD).

If you want to view a code of RDMA write, look at ib_write_bw.

Ido

>
> Quoting Ido Shamai <idos-LDSdmyG8hGV8YrgS2mwiifqBs+8SCbDb@public.gmane.org>:
>
>> Those are actually the same tests, from an IB point of view.
>>
>> the difference between them is that ib_write_bw passes the arguments
>> that need to be known to other side ( LID or GID, Qp num, PSN, RKey and
>> the address of the writing buffer) before post_sending,
>>
>> through TCP connection (using sockets) and the rdma_bw uses the rdma_cm
>> module , which sends those arguments through the IB wire and the SM.
>>
>> rdma_cm is not updated with all the features ib_write_bw have. I will
>> work on this test shortly
>>
>> Ido
>>
>>
>> On 5/17/2011 10:19 PM, ib-x2spCj9RiN0z5UmgcLIfJQ@public.gmane.org wrote:
>>> Just downloaded a snapshot of the perftests.  In the README under  "Test Descriptions" it says:
>>>
>>> rdma_bw.c     streaming BW test with RDMA write transactions
>>>
>>> and
>>>
>>> write_bw.c     BW test with RDMA write transactions
>>>
>>> What is the difference between rdma_bw and write_bw?  I see that  the rdma_bw description contains "streaming" where as write_bw  description does not...  but I am not sure I understand what that  means?
>>>
>>> E
>>>
>>> -- 
>>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>
>
>
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html