FreeTDS Development Group

[Peter Deacon <peterd AT iea-software.com>]

On Sun, 22 May 2011, James K. Lowden wrote:

> http://www.freetds.org/mars.html

> A little explanation of what I see as the problems inherent in MARS and
> why it's bad for TDS and FreeTDS, even if it's necessary to maintain an
> ODBC driver compatible with Microsoft's.  I thought it would be better
> to state my case plainly, where it can be evaluated and corrected, than
> to make occasional snide comments as we slouch our way there.  Maybe
> someone will even write a unit test to measure the "MARS tax".

FWIW tend to agree on network side.

My guess depending on network conditions (esp high RTT WAN/Intl links) at 
some point you are much better off with a connection pool of multiple real 
TCP sessions and shared session/transaction context (sp_bindsession) vs 
MARS.

No matter how good the protocol if you start multiplexing within a single 
TCP stream you are still limited by the head-of-line blocking nature of 
TCP.

With low RTTs MARS may significantly reduce concurrent TCP session counts 
at little to no cost and folks writing code for heavily async environments 
might tend to appreciate MARS more than those of us used to a synchronous 
environment.

regards,
Peter