[PVFS2-developers] question

[Phil Carns]

Yeah, this is a limit in our protocol right now.  The problem is that (for 
performance and network/memory friendlyness reasons) we need to limit the 
maximum size of unexpected requests that the server can accept.  

To enforce this limit we cap the size of various inputs that the encoder can 
handle (the first error you ran into) and also cap the size of bmi unexpected 
messages (the second error that you ran into).

I'm not sure what the best way to handle this is.  I can think of 2 
possiblities offhand:

1) make functions that are capable of generating large requests (sys-io being 
the first culprit, but maybe others) break up operations like that into 
multiple smaller steps somehow.

2) work out some protocol for sending larger requests to a server.  For 
example, the client could send a message that says "hey, I'm about to send 
something big" and the server could respond by posting a recv buffer of the 
appropriate size, and then treat its contents as an unexpected message to 
start a state machine.  I'm not sure what level this would be done, but it 
could perhaps be made to work for the general case...

Anyone know what the right answer is here?  :)

-Phil

On Tuesday 22 June 2004 14:46, Murali Vilayannur wrote:
> Hi All,
> I am trying to incorporate a client-side cache in the pvfs2-client daemon
> that adheres to close-to-open consistency semantics and I am using the
> PVFS_Request_hindexed() function to writeback all dirtied blocks to the
> server at the time of a close.
> For a particular workload (cp -Rf <X11R6 fonts directory> to a mounted
> pvfs2 volume), the client-daemon triggered the following
> message before it crashed,
> lebf_encode_req: encode_PVFS_servreq_io: demands 274 elements, more than
> maximum 100
>
> Setting PVFS_REQ_LIMIT_PINT_REQUEST_NUM to something larger triggers a
> message,
>  sys-io.sm: msg_p->send_status.error_code: message too long.
>
> Any ideas, thoughts, suggestions?
> Thanks
> Murali
> _______________________________________________
> PVFS2-developers mailing list
> PVFS2-developers at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers