summaryrefslogtreecommitdiff
path: root/open_issues/libpthread.mdwn
diff options
context:
space:
mode:
authorThomas Schwinge <tschwinge@gnu.org>2013-01-08 21:31:31 +0100
committerThomas Schwinge <tschwinge@gnu.org>2013-01-08 21:31:31 +0100
commit51c95fc11727532e3b0d98c8470a6b60907a0680 (patch)
tree6a8dd54654398bb2b05ce3f7cbcdc211d1dfbfb7 /open_issues/libpthread.mdwn
parent160d0597cb94d58f8ab273226b1f3830589a500b (diff)
IRC.
Diffstat (limited to 'open_issues/libpthread.mdwn')
-rw-r--r--open_issues/libpthread.mdwn135
1 files changed, 134 insertions, 1 deletions
diff --git a/open_issues/libpthread.mdwn b/open_issues/libpthread.mdwn
index befc1378..05aab85f 100644
--- a/open_issues/libpthread.mdwn
+++ b/open_issues/libpthread.mdwn
@@ -1,4 +1,4 @@
-[[!meta copyright="Copyright © 2010, 2011, 2012 Free Software Foundation,
+[[!meta copyright="Copyright © 2010, 2011, 2012, 2013 Free Software Foundation,
Inc."]]
[[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable
@@ -1257,6 +1257,19 @@ There is a [[!FF_project 275]][[!tag bounty]] on this task.
<braunr> i'll add traces to know which step causes the error
+### IRC, freenode, #hurd, 2012-12-11
+
+ <youpi> braunr: mktoolnix seems like a reproducer for the libports thread
+ priority issue
+ <youpi> (3 times)
+ <braunr> youpi: thanks
+ <braunr> youpi: where is that tool packaged ?
+ <pinotree> he probably means the mkvtoolnix source
+ <braunr> seems so
+ <braunr> i don't find anything else
+ <youpi> that's it, yes
+
+
## IRC, freenode, #hurd, 2012-12-05
<braunr> tschwinge: i'm currently working on a few easy bugs and i have
@@ -1326,3 +1339,123 @@ There is a [[!FF_project 275]][[!tag bounty]] on this task.
<braunr> i wondered for a long time why the load average was so high on the
hurd under even "light" loads
<braunr> now i know :)
+
+
+## IRC, freenode, #hurd, 2012-12-27
+
+ <youpi> btw, good news: the installer works with libpthread
+ <youpi> (well, at least boots, I haven't tested the installation)
+ <braunr> i can do that if the image is available publically
+ <braunr> youpi: the one thing i suspect won't work right is the hurd
+ console :/
+ <braunr> so we might need to not enable it by default
+ <youpi> braunr: you mean the mode setting?
+ <braunr> youpi: i don't know what's wrong with the hurd console, but it
+ seems to deadlock with pthreads
+ <youpi> ah?
+ <youpi> I don't have such issue
+ <braunr> ah ? i need to retest that then
+
+Same issue as [[term_blocking]] perhaps?
+
+
+## IRC, freenode, #hurd, 2013-01-06
+
+ <youpi> it seems fakeroot has become slow as hell
+ <braunr> fakeroot is the main source of dead name notifications
+ <braunr> well, a very heavy one
+ <braunr> with pthreads hurd servers, their priority is raised, precisely to
+ give them time to handle those dead name notifications
+ <braunr> which slows everything else down, but strongly reduces the rate at
+ which additional threads are created to handle dn notifications
+ <braunr> so this is expected
+ <youpi> ok :/
+ <braunr> which is why i mentioned a rewrite of io_select into a completely
+ synchronous io_poll
+ <braunr> so that the client themselves remove their requests, instead of
+ the servers doing it asynchronously when notified
+ <youpi> by "slows everything else down", you mean, if the servers do take
+ cpu time?
+ <braunr> but considering the amount of messaging it requires, it will be
+ slow on moderate to large fd sets with frequent calls (non blocking or
+ low timeout)
+ <braunr> yes
+ <youpi> well here the problem is not really it gets slowed down
+ <youpi> but that e.g. for gtk+2.0 build, it took 5h cpu time
+ <youpi> (and counting)
+ <braunr> ah, the hurd with pthreads is noticeably slower too
+ <braunr> i'm not sure why, but i suspect the amount of internal function
+ calls could account for some of the overhead
+ <youpi> I mean the fakeroot process
+ <youpi> not the server process
+ <braunr> hum
+ <braunr> that's not normal :)
+ <youpi> that's what I meant
+ <braunr> well, i should try to build gtk+20 some day
+ <braunr> i've been building glibc today and it's going fine for now
+ <youpi> it's the install stage which poses problem
+ <youpi> I've noticed it with the hurd package too
+ <braunr> the hurd is easier to build
+ <braunr> that's a good test case
+ <braunr> there are many times when fakeroot just doesn't use cpu, and it
+ doesn't look like a select timeout issue (it still behaved that way with
+ my fixed branch)
+ <youpi> in general, pfinet is taking really a lot of cpu time
+ <youpi> that's surprising
+ <braunr> why ?
+ <braunr> fakeroot uses it a lot
+ <youpi> I know
+ <youpi> but still
+ <youpi> 40% cpu time is not normal
+ <youpi> I don't see why it would need so much cpu time
+ <braunr> 17:57 < braunr> but considering the amount of messaging it
+ requires, it will be slow on moderate to large fd sets with frequent
+ calls (non blocking or low timeout)
+ <youpi> by "it", what did you mean?
+ <youpi> I thought you meant the synchronous select implementation
+ <braunr> something irrelevant here
+ <braunr> yes
+ <braunr> what matters here is the second part of my sentence, which is what
+ i think happens now
+ <youpi> you mean it's the IPC overhead which is taking so much time?
+ <braunr> i mean, it doesn't matter if io_select synchronously removes
+ requests, or does it by destroying ports and relying on notifications,
+ there are lots of messages in this case anyway
+ <braunr> yes
+ <youpi> why "a lot" ?
+ <youpi> more than one per select call?
+ <braunr> yes
+ <youpi> why ?
+ <braunr> one per fd
+ <braunr> then one to wait
+ <youpi> there are two in faked
+ <braunr> hum :)
+ <braunr> i remember the timeout is low
+ <braunr> but i don't remember its value
+ <youpi> the timeout is NULL in faked
+ <braunr> the client then
+ <youpi> the client doesn't use select
+ <braunr> i must be confused
+ <braunr> i thought it did through the fakeroot library
+ <braunr> but yes, i see the same behaviour, 30 times more cpu for pfinet
+ than faked-tcp
+ <braunr> or let's say between 10 to 30
+ <braunr> and during my tests, these were the moments the kernel would
+ create lots of threads in servers and fail because of lack of memory,
+ either kernel memory, or virtual in the client space (filled with thread
+ stacks)
+ <braunr> it could be due to threads spinning too much
+ <braunr> (inside pfinet)
+ <youpi> attaching a gdb shows it mostly inside __pthread_block
+ <youpi> uh, how awful pfinet's select is
+ <youpi> a big global lock
+ <youpi> whenever something happens all threads get woken up
+ <pinotree> BKL!
+ * pinotree runs
+ <braunr> we have many big hurd locks :p
+ <youpi> it's rather a big translator lock
+ <braunr> more than a global lock it seems, a global condvar too, isn't it ?
+ <youpi> sure
+ <braunr> we have a similar problem with the hurd-specific cancellation
+ code, it's in my todo list with io_select
+ <youpi> ah, no, the condvar is not global