MVICH and VI-GM

Hi,

I would like to know if somebody have been tried compiling MVICH (MPI over 
VIA, from NERSC) with VI-GM implementation.

Myricom people: have you tried? Is it possible?

Tkanks in advance for any information.

Best Regards,

===
Leonardo Alves de Paula e Silva     lapys at inf dot ufrgs dot br
Msc. Student                       Institute of Informatics/UFRGS
GPPD                                 http://www-gppd.inf.ufrgs.br

Address:
Av. Bento Gonçalves, 9500 Bloco IV    PO Box 15064      91501-970
Phone: +55 51 3316-6168 (Lab. 207)          Fax: +55 51 3316-7308
Porto Alegre - RS                                          Brazil

_______________________________________________
Myrinet mailing list
Myrinet <at> osc.edu
http://email.osc.edu/mailman/listinfo/myrinet

Re: MVICH and VI-GM

Leonardo Alves de Paula e Silva wrote:
> Hi,
> 
> I would like to know if somebody have been tried compiling MVICH (MPI 
> over VIA, from NERSC) with VI-GM implementation.
> 
> Myricom people: have you tried? Is it possible?

No, I haven't try. It makes more sense to use MPICH-GM instead of MVICH 
on top of VI-GM though, but it could be a way to validate  VI-GM.

Patrick
--

-- 

Patrick Geoffray
Myricom, Inc.
http://www.myri.com

asymmetric DMA speed on myrinet

Hello. I've seen that the DMA speed of Myrinet NIC is different based on 
the direction of DMA. 
L2E(from NIC to host memory) is faster than E2L(from host memory to NIC), 
especially for large messages.
Is it a Myrinet specific feature or a wide-accepted one?
(I guess giving higher speed to L2E might be good since it can reduce 
the number of packets in the network and resolve the congestion of the network.)
Can I balance the DMA speed with simple operation to special registers
on Myrinet NIC or any other method?
Thanks.

---
with regards
sang

Re: asymmetric DMA speed on myrinet

Hi,

mslee <at> camars.kaist.ac.kr wrote:
> Hello. I've seen that the DMA speed of Myrinet NIC is different based on 
> the direction of DMA. 
> L2E(from NIC to host memory) is faster than E2L(from host memory to NIC), 
> especially for large messages.
> Is it a Myrinet specific feature or a wide-accepted one?

Write DMAs (NIC to host) are always faster than Read DMAs (Host to NIC), 
unless something is broken. A Read DMA is actually composed of a request 
sent from the NIC to the host, then the data is collected from host 
memory and pushed to the NIC. This request/fetch delay is a penalty 
compared to Write DMAs where the data is immediately pushed by the NIC.

It's not a feature, it's just how it works. Not only Myrinet, all PCI 
devices.

> Can I balance the DMA speed with simple operation to special registers
> on Myrinet NIC or any other method?

A good PCI implementation would keep the difference to a minimum, but 
you can't change it, it's already best effort. Actually, you could slow 
things down by various means, but you really don't want to do that 

Patrick
--

-- 

Patrick Geoffray
Myricom, Inc.

MPICH-GM and iptables

Hi:

I've compiled MPICH-GM version 1.2.6..14b with GM version 2.0.21. The OS
is Fedora Core 3 with the 2.6.11-1.35_FC3smp kernel. I've been testing
MPICH-GM with the simple example routine hello++ installed in
/opt/mpi/examples.

When I run the routine using processors on the local node, the program
executes fine.

When I run the routine using processors on different nodes, the program 
hangs indefinitely and exits after 5 or 10 minutes with a "failure to 
connect" error.

If I turn iptables off ("/etc/init.d/iptables stop") and try hello++ with 
processors on different nodes, the program executes fine.

Iptables appears to be the problem. Does anybody have a working iptables 
rule that lets MIPCH-GM communicate?

Thanks!

Dan Strahs

Re: MPICH-GM and iptables

Hello all:

	Never mind: I've formed the iptables rule. 
		
						Dan Strahs

On Mon, 25 Jul 2005, Dan Strahs wrote:

> 
> Hi:
> 
> I've compiled MPICH-GM version 1.2.6..14b with GM version 2.0.21. The OS
> is Fedora Core 3 with the 2.6.11-1.35_FC3smp kernel. I've been testing
> MPICH-GM with the simple example routine hello++ installed in
> /opt/mpi/examples.
> 
> When I run the routine using processors on the local node, the program
> executes fine.
> 
> When I run the routine using processors on different nodes, the program 
> hangs indefinitely and exits after 5 or 10 minutes with a "failure to 
> connect" error.
> 
> If I turn iptables off ("/etc/init.d/iptables stop") and try hello++ with 
> processors on different nodes, the program executes fine.
> 
> Iptables appears to be the problem. Does anybody have a working iptables 
> rule that lets MIPCH-GM communicate?
> 

Re: Re: MPICH-GM and iptables

Hi People,

I've a setup of 8 nodes, each having M3F-PCIXD-2 interface. I want to
pass messages in Real-Time using MPI. Does this Myrinet interface
support Real-Time MPI(MPI/RT)? If not, then how can I pass messages in
Real-Time across nodes?

Thanks
-Ashish

MPI/RT for Myrinet

Hi People,

I've a setup of 8 nodes, each having M3F-PCIXD-2 interface. I want to
pass messages in Real-Time using MPI. Does this Myrinet interface
support Real-Time MPI(MPI/RT)? If not, then how can I pass messages in
Real-Time across nodes?

Thanks
-Ashish

Re: MPI/RT for Myrinet

Ashish Bijlani wrote:
> Hi People,
> 
> I've a setup of 8 nodes, each having M3F-PCIXD-2 interface. I want to
> pass messages in Real-Time using MPI. Does this Myrinet interface
> support Real-Time MPI(MPI/RT)? 

No. MPI/RT never really took off. I don't know anybody using it.

> If not, then how can I pass messages in
> Real-Time across nodes?

You may want to look at products from MPI Softech (www.mpi-softtech.com, 
now part of Verari). Tony was leading the MPI/RT effort a while ago, he 
may have support for it in his MPI implementations.

Patrick
--

-- 

Patrick Geoffray
Myricom, Inc.
http://www.myri.com

Re: MPI/RT for Myrinet

On 7/28/05, Patrick Geoffray <patrick <at> myri.com> wrote:
> Ashish Bijlani wrote:
> > Hi People,
> >
> > I've a setup of 8 nodes, each having M3F-PCIXD-2 interface. I want to
> > pass messages in Real-Time using MPI. Does this Myrinet interface
> > support Real-Time MPI(MPI/RT)?
> 
> No. MPI/RT never really took off. I don't know anybody using it.
>

Can I use MPICH-GM in Real-Time to pass messages to other nodes? If
not, then what are the bottlenecks involved?

> 
> > If not, then how can I pass messages in
> > Real-Time across nodes?
> 
> You may want to look at products from MPI Softech (www.mpi-softtech.com,
> now part of Verari). Tony was leading the MPI/RT effort a while ago, he
> may have support for it in his MPI implementations.
> 
> Patrick
> --
> 
> Patrick Geoffray
> Myricom, Inc.
> http://www.myri.com
>