A kernel level message passing service

Patrick Geoffray pgeoffra at cs.utk.edu
Thu Aug 31 12:45:54 PDT 2000


On Thu, 31 Aug 2000, Bryce Bockman wrote:
 
> > Does is make sense that this sort of thing could be applied to a message
> > passing interfaces for any sizable gain in performance?  If so, is there
> > any work already being done on this topic?

> My initial reaction to the thought is that there would be all kinds of
> security issues, but maybe some sort of "safe" API could be developed to
> address these issues.


On Thu, 31 Aug 2000, Tony Skjellum wrote:

> Bryce, see www.viarch.org ... the experience here is that good VI Arch  
> implementations appear to do what you are asking about (low overhead,

Hi Bryce,

Actually, almost all of the driver for high performance interconnects
provides OS-bypass functionnalities. It's true for Myrinet, SCI, Quadrics,
VIA is finally one more flavour.

But you are right : it's a tradeoff between security and performance.
Usually the systeme calls overhead for security is very small and out
of the critical path. The kernel is the only way to protect against other
users in a multi-users environnment.

Patrick Geoffray
---
Aerospatiale MATRA - Sycomore
Universite Lyon I - RESAM
http://lhpca.univ-lyon1.fr





More information about the Beowulf mailing list