[Spread-users] Maximum Message Size
James.Wagner at mci.com
Thu Jun 24 08:59:56 EDT 2004
We have tested with message sizes up to 1000M I believe. The following only
tests 2M, but the ideas are the same.
#define MAX_SCATTER_ELEMENTS 10000
* user.c - just used in test program
#define MAX_MESS_LEN 2000000 /* for testing larger buffers */
o change static char buf to use
MAX_MESS_LEN - something around that size...
The biggest issue in user.c is that it doesn't use the MAX_MESS_LEN constant
everywhere, but changing the above seemed to fix it.
We did this a while back and I don't remember testing larger buffers with
flooder, we probably just used user.c. I also believe we are currently
running with these larger sizes and sending data of larger sizes. I'm no
longer working on this exact code, so I'd have to wait until the person
working it shows up to confirm this. If the above doesn't help and you still
have problems, send another message later today and I'll ask them about it.
> -----Original Message-----
> From: spread-users-admin at lists.spread.org [mailto:spread-users-
> admin at lists.spread.org] On Behalf Of Killeen, Damien
> Sent: Thursday, June 24, 2004 3:31 AM
> To: Spread Mailing List (E-mail)
> Subject: [Spread-users] Maximum Message Size
> Yes, yet another post about the maximum message size in spread ... our
> application requires very large messages, up to 16MB.
> Going back through the message archives I see one solution is to fragement
> the messages before sending them, then reconstructing the message at the
> other end. This solution is not viable for us as we cannot afford the
> latency/throughput overheads of the reliable delivery service required to
> acheive the reliable reconstruction of the message.
> I have also seen advice that implies that by changing MAX_SCATTER_ELEMENTS
> in scatter.h and/or MAX_PACKET_SIZE in data_link.h the resulting
> MAX_MESSAGE_BODY_LEN in sess_types.h will change accordingly. I am
> advised that MAX_PACKET_SIZE is already as large as it can be for the UDP
> packet size, so my only option is changing MAX_SCATTER_ELEMENTS, from its
> current value of 100 to around 120000.
> I have done this, and successfully started the daemon, and used spflooder
> and spuser successfully. However, these programs have hardcoded limits
> relating to the 100000byte message size that I can't seem to circumvent
> (changing MAX_BYTES in flooder results in it seg faulting no matter what
> its input args are). My own code fails to send the large messages.
> In short, my question is, has anyone successfully increased the maximum
> message size by a substantial amount, and if so what were all the steps?
> Am I missing something by only changing MAX_SCATTER_ELEMENTS?
> Damien Killeen
> Spread-users mailing list
> Spread-users at lists.spread.org
More information about the Spread-users