[openib-general][PATCH 1 of 3] repost: Client Reregister support for kernel space

[Eitan Zahavi]

> -----Original Message-----
> From: Sean Hefty [mailto:mshefty at ichips.intel.com]
> Sent: Wednesday, May 31, 2006 9:15 PM
> To: Eitan Zahavi
> Cc: Leonid Arsh; Roland Dreier; openib-general at openib.org
> Subject: Re: [openib-general][PATCH 1 of 3] repost: Client Reregister
support for
> kernel space
> 
> Eitan Zahavi wrote:
> > Leonid just sent an example for a race that might happen if the SM
is to
> > be the maintainer of the data.
> 
> The race Leonid mentioned is a client sending a request when the SM is
down.
> That request will fail, so there's no data for the SM to maintain for
that node.
>   That's a retry condition that the client must deal with.
[EZ] The race is happening when the SM received the request and
responded but the other SMs or the file system did not fully stored that
registration and the SM crashed.
> 
> > [EZ] The SM is a single entity that has to respond to all requests
from
> > the entire cluster. (Even redirection requests). When you require
that
> > SM to also provide transaction safe storage or even worse then that
> > consistency with multiple standby SMs you worsen the problem. The
> > clients on the their side only need to maintain their own
registrations.
> 
> I don't believe that there's any requirement that the SM be a single
system.
> But I do believe that the SM should be able to recover from all SM
problems
> without interrupting any existing communication that is occurring the
fabric.
> SM failover or failure/restart should be as transparent to the clients
(i.e the
> non-SM nodes in the fabric) as possible.  (Btw, I also believe that
the SM
> should run on top of a real DBMS and support SQL style queries...)
[EZ] Please do the math of how many transactions per second you are
going to get from a reasonably priced SM if you are going to have a
distributed DBMS for that sake.
> 
> You don't want to push this problem to every application running in
the fabric,
> so why even push it to every node in the fabric?
> 
> - Sean