Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Tim Mattox (timattox_at_[hidden])
Date: 2007-04-30 07:43:02


Hi Adrian,

On 4/29/07, Adrian Knoth <adi_at_[hidden]> wrote:
> On Sun, Apr 29, 2007 at 10:18:01AM -0400, George Bosilca wrote:

[snip]
> BTW: How does multi TCP BTL works? I see num_links, but I wonder if
> kernel channel bonding would achieve the same results...

Unfortunately, kernel channel bonding is very unlikely to achieve the
same results as Open MPI's striping with multi TCP BTL.

Here is a relevant snippet of the documentation for kernel channel bonding:

========
balance-rr: This mode is the only mode that will permit a single
TCP/IP connection to stripe traffic across multiple
interfaces. It is therefore the only mode that will allow a
single TCP/IP stream to utilize more than one interface's
worth of throughput. This comes at a cost, however: the
striping often results in peer systems receiving packets out
of order, causing TCP/IP's congestion control system to kick
in, often by retransmitting segments.
========

The Open MPI way avoids this packet reordering problem, since
any individual TCP stream is carried by only a single NIC at either
end of the transmission. Since Open MPI splits the MPI traffic
across multiple TCP streams, additional throughput is possible
without the low-level packet reordering problem.

If you really want to know more about this, I could go on at length,
but I'll spare the rest of the readers on the mailing list for now.

-- 
Tim Mattox - http://homepage.mac.com/tmattox/
 tmattox_at_[hidden] || timattox_at_[hidden]
    I'm a bright... http://www.the-brights.net/