[libvirt] [PATCH 1/2] rpc: Switch to dynamically allocated message buffer

[Daniel P. Berrange]

On Thu, May 10, 2012 at 04:48:10PM +0200, Michal Privoznik wrote:
> On 03.05.2012 10:01, Daniel P. Berrange wrote:
> > On Wed, May 02, 2012 at 01:52:27PM -0600, Eric Blake wrote:
> >> On 04/27/2012 07:22 AM, Michal Privoznik wrote:
> >>> Currently, we are allocating buffer for RPC messages statically.
> >>> This is not such pain when RPC limits are small. However, if we want
> >>> ever to increase those limits, we need to allocate buffer dynamically,
> >>> based on RPC message len (= the first 4 bytes). Therefore we will
> >>> decrease our mem usage in most cases and still be flexible enough in
> >>> corner cases.
> >>> ---
> >>>  src/rpc/virnetclient.c       |   16 ++-
> >>>  src/rpc/virnetmessage.c      |   12 ++-
> >>>  src/rpc/virnetmessage.h      |    6 +-
> >>>  src/rpc/virnetserverclient.c |   20 ++-
> >>>  tests/virnetmessagetest.c    |  393 +++++++++++++++++++++++-------------------
> >>>  5 files changed, 264 insertions(+), 183 deletions(-)
> >>
> >> I haven't looked closely at this, but I do have a generic question:
> >>
> >> Are you malloc'ing and free'ing the buffer for each rpc on every call,
> >> or are you tracking a pool of buffers and only alloc'ing when needed?
> >> Furthermore, can you use the stack or a statically allocated array to
> >> receive short messages, saving alloc only for the long messages?  I'm
> >> worried about performance if we malloc on every single RPC call.
> > 
> > To be honest I'm not convinced malloc() would be a big factor in the
> > speed of the RPC calls. That said you could optimize it using the
> > VIR_ALLOC_VAR  to allocate the virNetMessage struct + the buffer
> > in one go. ALlocate all messages with 256kb initially and then you
> > can relloc larger only if needed, and re-alloc back to 256 kb when
> > done.
> > 
> > 
> > Daniel
> 
> Okay, point taken. Although I'd rather avoid allocating virNetMessage
> struct and buffer at once. It would complicate the code again, because
> the client side code has this struct _virNetClient; which already
> contains virNetMessage. However, if I'd need to re-alloc, I'd need to
> realloc whole message (!) therefore change the member to
> virNetMessagePtr. This implies too many changes.

Lets just go with your current code for now. We can easily revisit if
it proves to be a performance problem later.

Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|