whm@sunquest.UUCP (Bill Mitchell) (07/09/88)
In article <14318@hc.DSPO.GOV>, tomlin@hc.DSPO.GOV (Bob Tomlinson) writes: > > ... > Meanwhile the NeWS client side is trivially portable to Encores > (as well as MIPS, Alliant, etc.) and is actually being used in large > distributed applications that would *kill* a network if the applications > were using X. ... Bob, how well is your statement re X killing a network based on fact? If you folks have done some performance studies, I'm sure that everyone would like to hear about the results. If you're just speculating, what are the suppositions that led to your conclusion? (i.e., what was the server/ client mix, how much client/server interaction was required for X and NeWS respectively, and what was the network bandwidth?) We've been unable to find anyone who's done performance comparisons of X and NeWS. Let me throw out a couple of simple problems for X and NeWS fans to try on their favorite system: a) Create a C array of endpoint coordinates for 100,000 lines and then draw them. x and y values for endpoint coordinates should be selected using a uniform random distribution and scaled to produce lines with a mean length of five inches. Display the lines in a 10x10 inch window. Variations: After all the lines after been drawn, erase them in reverse order. Erase each line after it is drawn. Draw rectangles using the endpoints as cornerpoints. Generate the endpoints on the server, if you can. b) Using 10-point characters put up an 80x60 character window and scroll through 1000 80-character lines. The 61st line will cause the 1st line to disapear and all other lines to move up one line, just like a scrolling CRT. Use /usr/dict/words as the source of text. Note that each line should be 80 non-blank characters, for example: 10th1st2nd3rd4th5th6th7th8th9thaAaronABAAbabaabackabaloneabandonabaseabashabatea Variations: After displaying the last line, scroll back through the lines to the first line. Use any non-repetitive text from any source that you want to. ------------- Bill Mitchell sunquest!whm@arizona.edu uunet!sunquest!whm {allegra,cmcl2,noao}!arizona!sunquest!whm
liam@cs.qmc.ac.uk (William Roberts) (07/11/88)
In article <111@sunquest.UUCP> whm@sunquest.UUCP (Bill Mitchell) writes: > a) Create a C array of endpoint coordinates for 100,000 lines and then > draw them. x and y values for endpoint coordinates should be > selected using a uniform random distribution and scaled to > produce lines with a mean length of five inches. Display the > lines in a 10x10 inch window. > > Variations: > After all the lines after been drawn, erase them in reverse order. > > Erase each line after it is drawn. > > Draw rectangles using the endpoints as cornerpoints. > > Generate the endpoints on the server, if you can. OK, I'll try it if I can find the time - the only snags are: NeWS can't selectively erase lines X can't generate random numbers in the server Suggested experimental parameters: Local vs remote clients backing store/no backing store Thick lines/fast lines Would it matter if I used the standard UNIX random number generator with user-supplied seed value (so you can reproduce the same sets of endpoints) and ignored the normalisation? -- William Roberts ARPA: liam@cs.qmc.ac.uk (gw: cs.ucl.edu) Queen Mary College UUCP: liam@qmc-cs.UUCP LONDON, UK Tel: 01-975 5250
whm@sunquest.UUCP (Bill Mitchell) (07/13/88)
In article <111@sunquest.UUCP>, I proposed a couple of problems for studying
the relative performance of X and NeWS. In a couple of the proposed variations
of the first problem, I spoke of "erasing" lines that had been drawn. That
makes things sound a little more complicated than what I mind, namely, erase
the lines by just drawing lines in the background color; don't worry about
preserving the continuity of the intersecting lines.
Bill Mitchell
sunquest!whm@arizona.edu
uunet!sunquest!whm
{allegra,cmcl2,noao}!arizona!sunquest!whm