Re: mach_msg fails to copy out-of-line data when length >= 512MB

2011-09-22 Thread Roland McGrath
The whole point of OOL data is that it is relatively cheap to transfer huge quantities. So I don't think there should be any arbitrary limits.

Calling vm_object_deactivate_pages for each vm_object_deallocate kills performance.

2011-09-22 Thread Sergio López
Hi, Every time a persistent object is deallocated (this happens for every vm_map->memcpy->vm_deallocate operation), vm_object_deactivate_pages is called, which iterates over all pages in the object deactivating all active pages. The impact of this operation is really high when dealing with large o

mach_msg fails to copy out-of-line data when length >= 512MB

2011-09-22 Thread Sergio López
Hi, Sending messages with out-of-line data length equal or greater than 512 MB fails silently. The message arrives to the receiver, dataCnt indicates the amount pointed by the sender, but the buffer is not mapped into the receiver's space. This usually results in a unsolved page fault in the recei

Re: Degradation of GNU/Hurd ``system performance''

2011-09-22 Thread Svante Signell
On Thu, 2011-09-22 at 17:40 +0200, Sergio López wrote: > An easy way to reproduce this, is > sending a SIGINT in the middle of an opertation like "dd if=/dev/zero > of=test.bin bs=1M count=100". > > I think threads running vm_fault_copy doesn't deal properly with > thread_abort, but I didn't have

Re: Degradation of GNU/Hurd ``system performance''

2011-09-22 Thread Sergio López
2011/9/22 Svante Signell : > On Thu, 2011-09-22 at 04:45 +0200, olafbuddenha...@gmx.net wrote: >> Hi, >> >> On Mon, Sep 05, 2011 at 06:42:22AM +0200, olafbuddenha...@gmx.net wrote: >> >> > As recently I aquired the (bad) habit of running my system 24/7, >> > almost never voluntarily rebooting, I wa

Re: Degradation of GNU/Hurd ``system performance''

2011-09-22 Thread Svante Signell
On Thu, 2011-09-22 at 15:18 +0200, Svante Signell wrote: > On Thu, 2011-09-22 at 04:45 +0200, olafbuddenha...@gmx.net wrote: > > Hi, .. > several /bin/bash -noprofile /dev/fd/3 in state Sslo These seems to be started by /libexec/runttys, so they should probably be there. State flag 'o' still unex

Re: Degradation of GNU/Hurd ``system performance''

2011-09-22 Thread Guillem Jover
On Thu, 2011-09-22 at 04:45:55 +0200, olafbuddenha...@gmx.net wrote: > On Mon, Sep 05, 2011 at 06:42:22AM +0200, olafbuddenha...@gmx.net wrote: > > As recently I aquired the (bad) habit of running my system 24/7, > > almost never voluntarily rebooting, I was able to make further > > observations. >

Re: Degradation of GNU/Hurd ``system performance''

2011-09-22 Thread Svante Signell
On Thu, 2011-09-22 at 04:45 +0200, olafbuddenha...@gmx.net wrote: > Hi, > > On Mon, Sep 05, 2011 at 06:42:22AM +0200, olafbuddenha...@gmx.net wrote: > > > As recently I aquired the (bad) habit of running my system 24/7, > > almost never voluntarily rebooting, I was able to make further > > observ

Re: Diagram of Hurd I/O, read

2011-09-22 Thread Samuel Thibault
Thomas Schwinge, le Thu 22 Sep 2011 12:51:55 +0200, a écrit : > +> `trivfs_S_io_read`. Depending on the internatl state, either a new memory spurious t^ Apart from that, looks right. Samuel

Diagram of Hurd I/O, read

2011-09-22 Thread Thomas Schwinge
Hi! Yesterday night (while other Europeans have been sleeping) ;-), Sergio, Olaf, Marcus have been discussing about how Hurd I/O, read in particular, is spread over the various components, whishing there to be a diagram of it. And, I have the answer, and it's the same one as so often: there alrea