[PVFS2-developers] Mechanism for forwarding data

[Bradley W Settlemyer]

From: Brad Settlemyer <bradles at clemson.edu>
To: [List] PVFS2 Developers <pvfs2-developers at beowulf-underground.org>
Subject: Mechanism for forwarding data
Date: Wed, 07 Jul 2004 13:51:29 -0400

Hello,

  I am interested in developing a mechanism to allow a client to send
data to a data server, and that server then simultaneously persists the
data and forwards it to another data server.

  The 2 obvious mechanisms to do this are to modify the flow subsystem
or the trove subsystem.

  The approach to modifying the flow subsystem would seem to be to
develop a new aggregated endpoint type, e.g. the
forwarding_trove_endpoint.  In servicing the data, this endpoint would
write the data thru trove, and also write the data into another flow
from the first server to the second server's trove repository.  This
endpoint would be different than existing endpoints in that it would be
uni-directional.

  The other obvious approach would be to provide a new trove
implementation. i.e. a forwarding trove implementation.  For writes, the
implementation would persist the data using whatever the existing
mechanism is (aio I think) and it would also construct a flow to the 2nd
server to forward the data over.  I'm not sure how well the write end of
trove meshes with the send end of a flow, so there may be some impedance
mismatch of some sort there.

  Is either one of these approaches superior to the other?  Is there
another simple or simpler approach that is superior to these?

Cheers,
Brad