Fred Cheng and I turned Poller into an abstract base class, moved the existing poll()-based implementation into class Poller_poll, and added a new...
Dan Kegel
dank@...
Sep 7, 2000 3:29 am
110
... That's a nice feature, but you don't need a hash table, actually. Since fd's are small positive integers, you can simply use an array, as I do in my...
Dan Kegel
dank@...
Sep 8, 2000 11:47 pm
111
... they don't remain small forever :) And maintaining the array is a bit of an annoyance. Its way cooler to just dereference the pointer you get back.. ... ...
Zach Brown
zab@...
Sep 9, 2000 2:41 pm
112
... I agree, and a while ago was going to ask for F_SETSIG to also take a context pointer for this purpose, but the array stuff isn't soooo awful. Having...
Dan Kegel
dank@...
Sep 10, 2000 6:58 pm
113
... I second this. An extra 4 or 8 bytes of state per descriptor isn't going to kill you. I can't really think of any useful applications that aren't going...
shane@...
Sep 11, 2000 8:34 am
114
Fred implemented Poller_devpoll already, and tested it on his machine! It needs a little cleanup, but we should be able to post it soon. Not hard to use,...
Dan Kegel
dank@...
Sep 12, 2000 1:33 am
115
I was puttering around tonight thinking up a new design for my ftp protocol stack, and decided to try an approach where every part of the stack has a uniformly...
Dan Kegel
dank@...
Sep 12, 2000 5:00 am
116
... Hash: SHA1 ... At this point I had to ask why it was a "stack" and not a stream.... then I read on and I realized great minds think alike. ;-) ... I don't...
Christopher Smith
x@...
Sep 12, 2000 6:15 am
117
Hi, Firstly, appologies to all for this cross post. I have hidden the recepient lists (4) in the bcc field to stop replies posting to all lists. For a long...
Beau Kuiper
kuiperba@...
Sep 13, 2000 3:24 am
118
... Please post a URL rather than the actual document, as egroups strips attachments. Also, yes, you are on crack :-) The crucial steps before writing the RFC...
Dan Kegel
dank@...
Sep 13, 2000 3:36 am
119
... URL is http://www.arach.net.au/~wildfire/ftp-host.txt This specification also considers compatibility with existing clients. Beau Kuiper ...
Beau Kuiper
kuiperba@...
Sep 13, 2000 4:44 am
120
... Comment: user!host is a bit unusual. Wouldn't user@host (as in email) or possibly host:user (as in scp) or host!user (as in uucp -- ok, so I'm dating...
Dan Kegel
dank@...
Sep 13, 2000 5:05 am
121
... I can't use this method because proxies and web browsers will get confused, and won't send the correct username. eg: ftp://anonymous@...@test.com/ is...
Beau Kuiper
kuiperba@...
Sep 13, 2000 5:21 am
122
Three cheers for Michael Elkins and Fred Cheng, who contributed implementations of Poller for kqueue and /dev/poll, respectively! New release of dkftpbench...
Dan Kegel
dank@...
Sep 15, 2000 4:55 am
123
This resolves the issues I mentioned in last post (though I still haven't moved wakeUp to the base class). It's fit to look at now. Next, Fred and I plan to do...
Dan Kegel
dank@...
Sep 16, 2000 1:18 am
124
... God... I feel compelled to do an AIO version... Maybe after my continuously delayed laptop arrives. ;-) --Chris...
Christopher Smith
x@...
Sep 16, 2000 5:32 am
125
... It's been on my mind for a while. We want to allow programs written with Poller to substitute any subclass of Poller transparantly. That might mean we...
Dan Kegel
dank@...
Sep 16, 2000 3:21 pm
126
http://www.kegel.com/c10k.html#zerocopy now links to http://people.freebsd.org/~ken/zero_copy/ which describes some patches for FreeBSD which add support for...
Dan Kegel
dank@...
Sep 18, 2000 2:51 pm
127
#1: http://www.kegel.com/c10k.html#overload links to a new benchmark using phhttpd. It has a good graph showing how bad things can get if you don't actively...
Dan Kegel
dank@...
Sep 24, 2000 5:38 am
128
... I need to look at the source for Poller more directly, but wouldn't it make sense for a Poller instance to detect an overload itself and handle it...
Christopher Smith
x@...
Sep 24, 2000 6:15 am
129
... It might. We'd need to extend Poller to have explicit support for accept operations; right now, no distinction is made between listening sockets and...
Dan Kegel
dank@...
Sep 24, 2000 6:51 am
130
... It seems to me almost like an FSM (I know, all problems can be reduced to FSM's). The only real difference between a read socket and an accept socket is...
Christopher Smith
x@...
Sep 24, 2000 7:03 am
131
... On second thought, this definitely doesn't belong in Poller. Applications which listen on multiple sockets might want to have different behavior for each...
Dan Kegel
dank@...
Sep 24, 2000 2:31 pm
132
can we _please_ move this list to a listserver that doesn't inject bullshit advertising to the messages? I will gladly donate mailman resources @zabbo.net if...
Zach Brown
zab@...
Sep 24, 2000 5:16 pm
133
... Its starting to feel like a layering/complexity problem. Might it be better to try putting it in the app first and see if it makes more sense there? not...
Zach Brown
zab@...
Sep 24, 2000 5:20 pm
134
I'm moving the mailing list to Sourceforge so we can escape this egroups nonsense. Should be ready in a day or two. - Dan...
Dan Kegel
dank@...
Sep 25, 2000 5:47 am
135
I cleaned up the microbenchmark Poller_bench.cc, and now have microbenchmark results for poll() and select(). poll() wins at 500 total + either 5 or 50 active...
Dan Kegel
dank@...
Sep 25, 2000 6:40 am
136
See http://www.kegel.com/dkftpbench/Poller_bench.txt Looks like /dev/poll is a win on Solaris, but not yet on Linux. Comments welcome. I look forward to seeing...
Dan Kegel
dank@...
Sep 27, 2000 3:10 pm
137
... mhz? /usr/platform/sun4u/sbin/prtdiag :) ... can you set profile=2 on the command line and boot and send the output of readprofile after _lots_ of...
Zach Brown
zab@...
Sep 27, 2000 4:02 pm
138
... 167 MHz. ... I worked out how to use that kernel profiler (thanks) and have updated the page with profiling data for poll and select; see ...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
