larree@altos86.Altos.COM (Larry Snyder) (11/22/89)
I am looking for ways of describing network performance in terms which can be easily compared to other systems and configurations. Are there any widely used programs which will measure the performance of various layers of a TCP/IP protocol stack? How does one measure performance degradation when adding virtual circuits or playing with block sizes? Is it useful to measure the inter-frame delay between a 'ping' and the response? By using a network analyzer to help calculate the average transmission rate between stations, am i getting an accurate measurement of my ethernet driver? What about NFS or RFS? And, most important, are any numbers available for other systems? (any useful sources mailed to me will be "shar"ed and posted) -- Larry Snyder Altos Computer Systems Internet: larree@altos.com 2641 Orchard Parkway UUCP : uunet!altos!larree San Jose CA 95134 (my opionions have nothing to do with microsoft.) 408.432.6200
haas@cs.utah.edu (Walt Haas) (11/22/89)
In article <28@altos86.Altos.COM> larree@Altos.COM (Larry Snyder) writes: >I am looking for ways of describing network performance in terms >which can be easily compared to other systems and configurations. Well I've found it important to not stop at a simple packets/second or bytes/second figure. Another important property is the ability to handle bursts of closely spaced packets. Some devices drop packets closer than X microseconds; this value is frequently more important than the average sustained throughput. Other devices can only buffer up Y packets at minimal spacing before losing packets. Another figure of merit that's very important in interactive use is the ability to return a quick response to the user's keystroke. In some cases minimizing this response time will reduce other performance figures, such as average throughput. Cheers -- Walt