Previous | Next --- Slide 11 of 36
Back to Lecture Thumbnails
amolakn

If I'm understanding correctly, I believe that the Optimistic method that allocates the buffer is probably better for smaller messages (such as maybe just sending one integer) since the allocated buffer is better than the overhead of tag matching in this conservative method, right?

I do also remember the Professor mentioning with the optimistic method that even if shorter messages are ok, you never know if there may be several processors sending short messages so the even though the buffer required for each message is small, the buffer required to fit all of them may still be large so that still may not be ideal.

autumnust

@amolakn Why do you think Optimistic method will be better for short messages? If there are huge amounts of short messages and each allocate buffer on sender side, will it cause potential resource contention ?

acortes

@autumnust I think short messages means a single or a few short messages being sent. That in contrast with a few large messages is what makes the difference. That being said, you are never guaranteed you will only receive a few short messages so optimistic can lead to contention.

tcm

Actually, a concern that we discussed with regard to short messages was the latency of sending them, since the conservative approach involves 3 messages. For example, if you only wanted to send a 4-byte integer, that is a lot of latency for sending a short message. The optimistic approach can have lower latency (just 1 message), but the input buffer problem is a concern (even with short messages).

lusiliang93

For conservative approach, if send-ready request fails at first, sender will resend the ready request.If receive-ready request is successful the next time, the data finally sent to the destination should be different from what the sender wants to send in the beginning since there seems to be no buffer in this approach?