[Spread-users] Big messages in Spread

[Roel van der Goot]

On Fri, 20 Jul 2001, Jonathan Stanton wrote:

> Here is the real question, which I would be interested in anyone's
> opinions on.
>
> What should the maximum message size supported by a Spread
> client library be?

This is not an easy question to answer. It will most likely depend on the
amount of memory in the machines that you are running on and the amount of
clients that are expected to participate. Given that big message sizes
result in big buffers, the buffer size could potentially grow to
number_of_clients * max_msg_size. In short, I think it should be
configurable.

> Should that size depend on the way the daemon is compiled (i.e.
> MAX_PACKET_SIZE)?

I am fine with a compile-time maximum (see above) although a run-time maximum
is always nicer. However I do see drawbacks: Unlimited message sizes result in
unlimited buffer sizes. A limited buffer size could lead to deadlock! (Hmmm,
is this already a problem in the current implementation?) If you decide to go
with dynamic you will definitely have to inform your users about the
consequences. There is no free lunches!

> Currently the answers are MAX_SCATTER_ELEMENTS * 1440 and sortof ( it
> depends only on MAX_SCATTER_ELEMENTS, but if you shrink MAX_PACKET_SIZE
> then too large messges can be sent which is buggy)

I think that MAX_PACKET_SIZE should be used by both the clients and the daemon
because they are dependant on the network that you are running on (udp/ip or
fcp [fibrechannel protocol]).

> I think there should be better answers :-)

May be if there were simpler questions. :-)

Roel.