LAM/MPI General User's Mailing List Archives

From: Ross Torkington (ELA01RMT_at_[hidden])
Date: 2004-04-01 13:11:17

• Next message: Ross Torkington: "Re: LAM: fault tolerance: lamboot -x"
• Previous message: Vishal Sahay: "Re: LAM: Dynamic allocation of request array"
• Next in thread: Bogdan Costescu: "Re: LAM: sharp increases in send time for large messages"
• Reply: Bogdan Costescu: "Re: LAM: sharp increases in send time for large messages"
• Maybe reply: Ross Torkington: "Re: LAM: sharp increases in send time for large messages"

Hello! I've been timing how long it takes to send various sized messages and found that the cost per kb begins high for short messages, drops sharply and remains at around 50us for messages between 10 and 100kb but then increases in a staircase fashion beyond that. I understand that LAM uses two send protocols with the default crossover threshold being 64kb but I don't understand why the cost per kb has more than one sharp increase beyond the 100kb mark.

Many thanks for your help!
Ross

• text/html attachment: attachment

• Next message: Ross Torkington: "Re: LAM: fault tolerance: lamboot -x"
• Previous message: Vishal Sahay: "Re: LAM: Dynamic allocation of request array"
• Next in thread: Bogdan Costescu: "Re: LAM: sharp increases in send time for large messages"
• Reply: Bogdan Costescu: "Re: LAM: sharp increases in send time for large messages"
• Maybe reply: Ross Torkington: "Re: LAM: sharp increases in send time for large messages"