Subject: | |
From: | |
Reply To: | |
Date: | Mon, 4 Nov 1996 13:20:43 -0800 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Hi,
I've duplicated the CGI "lost data" problem that some people have
reported recently. Here's some comments/observations.
1) When transferring large amounts of data, sometimes the transfer
is interrupted after a portion is sent
(typically 32KB, but sometimes 16KB)
2) When the interruption occurs, the cgi program is aborted!
I deduced #2 by writing a cgi program that sends the following text
to the web server:
<BODY>0 1 2 3 4 5 6 7 8 9 ... 18 19 20 21 22 23 24 25 26 27 <BR>
28 29 30 31 32 33 34 35 ... 43 44 45 46 47 48 49 50 51 52 <BR>
53 54 55 56 57 58 59 60 ... 68 69 70 71 72 73 74 75 76 77 <BR>
...
19980 19981 19982 19983 ... 19988 19989 19990 19991 19992 <BR>
19993 19994 19995 19996 19997 19998 19999 20000 <BR>
Done, #bytes 114528, #recs 1420</BODY></HTML>
and then does a PRINTOP with a summary of the number of bytes...in
normal operation, I get the entire text sent to my web browser *and* the
summary appears on the system console.
When I get a truncated file on the web browser, I do *NOT* see a message
on the system console...indicating that the cgi program was aborted
prior to finishing!
I tested accessing this cgi program over our local Ethernet (10 Mb/sec),
and found:
- an HP-UX workstation received the entire text about 90%+ of the time.
When it failed, it stopped at about number 6336 ... which represented
(counting the unseen overhead) about 32768 bytes (according to the
access_log).
- a Win95 Pentium received the entire text about 0% of the time,
consistently stopping at number 3264 (16384 bytes).
Up to this point, all of the tests described above were done using the
OpenMarket WebServer. I then tried the NCSA server, and found:
- EVERY transmission to the Win95 and HP-UX workstation worked!
I didn't get a single failure.
Also, I noticed:
- An HP-UX workstation (9000/859 aka K220) running Netscape 3.0 was
about twice as fast as a Pentium 75, running Netscape 3.0 in displaying
the entire text.
--
Stan Sieler [log in to unmask]
http://www.allegro.com/sieler.html
|
|
|