View previous topic :: View next topic |
Author |
Message |
jeremy_c
Joined: 09 Apr 2005 Posts: 16 Location: Ohio, USA
|
Posted: Tue Apr 19, 2005 4:56 pm Post subject: Stability? |
|
|
I compiled the example servlets then ran apache benchmark on 100 requests, this is what I got:
Program received signal SIGUSR1, User defined signal 1.
[Switching to Thread -1209555264 (LWP 19326)]
0xffffe410 in ?? ()
(gdb) bt
#0 0xffffe410 in ?? ()
#1 0xbffff6c0 in ?? ()
#2 0xb7d7b7c0 in ?? ()
#3 0x00000000 in ?? ()
#4 0xb7f04f46 in __nanosleep_nocancel () from /lib/tls/libc.so.6
#5 0xb7f2eb0a in usleep () from /lib/tls/libc.so.6
#6 0x08055049 in _D5mango4base6System6System5sleepFkZv ()
#7 0x0804e554 in _D8servlets10testServerFC5mango4http6server5model9IProvider9IProviderZv ()
#8 0x0804e6f6 in _D8servlets17testServletEngineFZv ()
#9 0x0804e72c in _Dmain () at example/servlets.d:588
#10 0x080621af in main ()
(gdb)
Any ideas?
Jeremy |
|
Back to top |
|
|
kris
Joined: 27 Mar 2004 Posts: 1494 Location: South Pacific
|
Posted: Wed Apr 20, 2005 5:18 pm Post subject: |
|
|
Mango is typically tested on Win32, so these kinds of things can sometimes get overlooked (unfortunately). The lack of D debugging info and demangling on the linux platform just made the task too hideous to bear so far, for most folks. There's two issues that have been noted before regarding multiple threads and sockets under linux:
- The sockets were not being scavanged quickly enough (by the OS) for one particular stress-test
- A sleeping thread has been seen to die a horrible death whilst waiting for a multicast packet (when the program terminated)
The issue you ran into may be related to the latter? I mean, your thread apparently broke in _nanosleep_nocancel(), for Bob's sake
I thnk this will ultimately require a solid session with valgrind (and someone who knows how to drive that software). Is this something you might be able to help with?
- Kris |
|
Back to top |
|
|
jeremy_c
Joined: 09 Apr 2005 Posts: 16 Location: Ohio, USA
|
Posted: Thu Apr 21, 2005 7:57 am Post subject: |
|
|
kris wrote: | I thnk this will ultimately require a solid session with valgrind (and someone who knows how to drive that software). Is this something you might be able to help with?
- Kris |
I've heard of it, never used it. Hm.
Jeremy |
|
Back to top |
|
|
kris
Joined: 27 Mar 2004 Posts: 1494 Location: South Pacific
|
Posted: Thu Apr 21, 2005 5:17 pm Post subject: |
|
|
jeremy_c wrote: | kris wrote: | I thnk this will ultimately require a solid session with valgrind (and someone who knows how to drive that software). Is this something you might be able to help with?
- Kris |
I've heard of it, never used it. Hm.
Jeremy |
I'd be interested to see the results of apache benchmark against a server running Win32 ... the servlet examples are not ideal, but the basic http server is exceptionally efficient (due to heavy-usage of thread-locals and slicing, it makes zero memory allocations whilst servicing a request).
For example, a 1.6GHz Pentium M (the original one) in a laptop was handling over 3000 server requests per second, and more than half the CPU was being eaten up by the stress-test-client itself. |
|
Back to top |
|
|
JJR
Joined: 22 Feb 2004 Posts: 1104
|
Posted: Thu Apr 21, 2005 6:34 pm Post subject: |
|
|
*Ahem* It was a 1.4 GHz Pentium-M, Mr. Bell... And I think you busted it.
(or maybe you were referring to your own laptop?) |
|
Back to top |
|
|
kris
Joined: 27 Mar 2004 Posts: 1494 Location: South Pacific
|
Posted: Thu Apr 21, 2005 7:54 pm Post subject: |
|
|
JJR wrote: | *Ahem* It was a 1.4 GHz Pentium-M, Mr. Bell... And I think you busted it.
(or maybe you were referring to your own laptop?) |
I do beg your pardon, sir. Please accept my most humble and gracious apologies
(for those who don't know: JJR has been instrumental in getting Mango operating on linux. The aforementioned problems with sockets were identified and traced by JJR also)
Hey John; Doesn't this latest issue sound rather like the 'multicast-socket' problem? Where the app would throw-up on exit if there was an active multicast-listener? I'm wondering whether it might be a config-issue in the linux version of Sockets.d ... maybe it would be worth switching to Chris Miller's latest version, and trying that instead? |
|
Back to top |
|
|
JJR
Joined: 22 Feb 2004 Posts: 1104
|
Posted: Thu Apr 21, 2005 10:32 pm Post subject: |
|
|
kris wrote: | JJR wrote: | *Ahem* It was a 1.4 GHz Pentium-M, Mr. Bell... And I think you busted it.
(or maybe you were referring to your own laptop?) |
I do beg your pardon, sir. Please accept my most humble and gracious apologies
|
kris wrote: | (for those who don't know: JJR has been instrumental in getting Mango operating on linux. The aforementioned problems with sockets were identified and traced by JJR also) |
Oop... wasn't fishing for credits, Kris, but thanks!
kris wrote: | Hey John; Doesn't this latest issue sound rather like the 'multicast-socket' problem? Where the app would throw-up on exit if there was an active multicast-listener? I'm wondering whether it might be a config-issue in the linux version of Sockets.d ... maybe it would be worth switching to Chris Miller's latest version, and trying that instead? |
Yes, it does look very much like the same problem. Swapping in Chris' version couldn't hurt to try... Use of Phobos' thread class may be still part of the problem too. Didn't we discover that there was a bug in phobos' threading package also (on Linux specifically) that possibly related to this?
I'd like to give this another try when I get my Laptop back... might be back in a couple weeks... I'll let you know.
All the best,
John |
|
Back to top |
|
|
kris
Joined: 27 Mar 2004 Posts: 1494 Location: South Pacific
|
Posted: Fri Apr 22, 2005 11:43 am Post subject: |
|
|
JJR wrote: | Oop... wasn't fishing for credits |
Aye, but credit where it's due.
JJR wrote: | I'd like to give this another try when I get my Laptop back... might be back in a couple weeks... I'll let you know. |
Sounds good. |
|
Back to top |
|
|
|