bm@sumikin.co.jp (Bunpei Matoba) (11/07/90)
I'm not familar with both ultrix and DECstation hardware. However, I'm involved in a project making a real time control system using a DECstation 3100 and I have to evaluate software performance prior to designing the whole system. My interests are; 1. How long does it take to send a message from a process to another process in the same machine using a socket inter-process communication mechanism? Does the message transfer mechanism require delay of micro-second order or millisecond order? 2. How long does it take to receive a single packet message from another machine via TCP connection on ethernet? I measured those items before on Multibus-based computers using a logic analyzer monitoring internal hardware signals. For example, in order to measure the delay for a message exchange, I inserted a simple code for changing voltage of a hardware line just before a "send" system call and just after a "receive" system call respectively. By monitoring the H/L state change of the hardware line connected to a external logic analyzer I could know the precise delay time between two system calls. I'd like to apply the simular method to the case of DECstion3100. Is there any hardware lines which satisfy the following conditions? 1. The lines are available on a DECstation 3100 machine of standard hardware configuration. 2. The lines are easily found outside of the machine. (if it is impossible, inside of the machine OK) 3. We can easily change a voltage of the lines in a test program by calling c language level functions. And the change takes place just after calling the function within a period shorter than 100 micro-seconds. I know this question is too "hard" in this newsgroup. I would be very happy, however, if anyone could post a reply. Thank you very much in advance. bm@sumikin.co.jp Bunpei Matoba Computer Engineer Sumitomo Metal Industries Control & Engineering Corporation, JAPAN -- $BE*>lJ8J?(J (MATOBA, Bunpei) bm@sumikin.co.jp$B!J=;M'6bB0!K(J $B=;6b@)8f%(%s%8%K%"%j%s(J$B%03t<02q<R(J $B%3%s%T%e!<%?@)8fIt(J phone. NTT (06)418-4715 SUN-NET +671-2602 fax. NTT (06)418-4764 CompuServe 72300,3551
jg@crl.dec.com (Jim Gettys) (11/07/90)
This is not quite the data you were asking for. In any case, measured using X11perf (i.e. user level code talking to a real X server), a ROUND TRIP (message from x11perf, processing of it in the X server, and a response) is 2 milliseconds on machine using Unix domain IPC, 3 milliseconds using TCP, with Ultrix 4.0 on a DECstation 3100, and across the network with TCP, it is 3.6 milliseconds for the ROUND TRIP. This includes the process scheduling, and everything, for a complete round trip. One could probably do marginally better than this, and whether the measurement methodology is what you'd like is another question as well. - Jim
bm@sumikin.co.jp (Bunpei Matoba) (11/16/90)
Thank you very much for a benchmarking report for inter-process communication performance on a DECstation 3100 by Jim Gettys. > In any case, measured using X11perf (i.e. user level code talking to > a real X server), a ROUND TRIP (message from x11perf, processing of > it in the X server, and a response) is 2 milliseconds > on machine using Unix domain IPC, 3 milliseconds using TCP, with Ultrix 4.0 > on a DECstation 3100, and across the network with TCP, it is 3.6 > milliseconds for the ROUND TRIP. This includes the process scheduling, > and everything, for a complete round trip. I just know that X11perf is a bench marking codes for X-windows and it outputs messages including a line like this; Sync time adjustment is xx.xxxx msecs. I guess that the ROUND TRIP time Jim said is equivalent to the " Sync time adjustment". My understanding of Jim's report is; -----------DECstation 3100 -----( standalone )------------------------- | -------------------- Unix domain IPC(socket) --------------- | | | X-client: X11perf| -------------------------- | X-server: | | | | | 2.0 msec | | | | | | | <------------------------- | | | | -------------------- --------------- | ----------------------------------------------------------------------- --- DECstation 3100 ----- ----DECstation 3100-- | -------------------- | TCP/IP Ethernet | --------------- | | | X-client: X11perf| |------------------------ | | X-server: | | | | | | 3.6 msec | | | | | | | | |<----------------------- | | | | | -------------------- | | --------------- | ------------------------- --------------------- Is it correct? Please let me know, if my understanding is wrong. bm@sumikin.co.jp Bunpei Matoba Sumitomo Metal Industries Control and Engineering Corporation Amagasaki Japan -- $BE*>lJ8J?(J (MATOBA, Bunpei) bm@sumikin.co.jp$B!J=;M'6bB0!K(J $B=;6b@)8f%(%s%8%K%"%j%s%03t<02q(J$B<R(J $B%3%s%T%e!<%?@)8fIt(J phone. NTT (06)418-4715 SUN-NET +671-2602 fax. NTT (06)418-4764 CompuServe 72300,3551