[omniORB] Sanity Check on Large Data Transfers

Sai-Lai Lo s.lo@uk.research.att.com
Fri, 15 Jun 2001 09:44:25 -0000


> I now have my graph-as-structs transfer working. I can transfer a 3Meg
> struct in about 5 seconds over a lightly-loaded 100BaseT network. That
> seems like good performance to me. The client and server machines for
> this test are reasonably fast PIII PCs.
>
> This data reflects 5100 unique graph nodes and about 1 Meg of string
> data for a total of about 3Meg of data over the wire (estimated by
> turning down the message size until I get a COMM failure--I assume
> that's telling me the size of the data being transfered).
>
> My sanity check questions:
>
> 1. Is this good performance? Is there anything I could do in the ORB
> configuration or in the way I've defined the structs to improve it?

I think you will see a significant performance improvement with omniORB4 to
omniORB4. This is because omniORB4 uses GIOP 1.2 by default and with that
protocol only a single pass is needed to marshal the data. omniORB3 supports
GIOP 1.0 and in order to limit the amount of buffering required, data
marshalling is done as a 2 pass process. The first pass calculates the total
message size.

For a complex structure like yours, removing the 1st pass will make a
noticable difference.

Before you ask, I'll announce the pre-release of omniORB4 real soon now.
Still have to write up some documentation before this can be done.


> 2. What would people consider to be a safe maximum message size? For
> this test, obviously I had to set the value to > 3Meg (I chose 10Meg as
> a round number for my testing).


With omniORB, the buffering requirement (default to 8K for each direction)
stays the same irrespective of the message size. This may not be the case
for other ORB implementations, i.e. the message may be fully buffered
internally before it is unmarshalled. If you worry about memory footprint
and have to use another ORB, say a java one, a large message size may not be
desirable.

Regards,

Sai-Lai
---
Sai-Lai Lo                                   S.Lo@uk.research.att.com
AT&T Laboratories Cambridge           WWW:   http://www.uk.research.att.com
24a Trumpington Street                Tel:   +44 1223 343000
Cambridge CB2 1QA                     Fax:   +44 1223 313542
ENGLAND