October 2004 - Darwin-kernel

Re: Very poor PPP interface performance
by Jean-Edouard BABIN 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Microsoft-Entourage/11.1.0.040913 Hello You're right, i also already setup a ppp tunnel over ssh With something like pppd pty 'ssh -x -t -e none user@server /usr/sbin/pppd passive noauth 9600' noauth 10.0.0.1:10.0.0.2 and get the same poor performance, and bad ping also :/ on 26/10/04 20:55, « Jamie Wood » <eimaj79(a)hotmail.com> wrote: > I have noticed that the performance of the PPP interface in Mac OS X 10.3.5 > is not very good at all. I have written a simple program that brings up a > PPP interface, configures it using "ifconfig", and reads incoming packets. > If I use another program to repeatedly send UDP packets to my newly-created > PPP interface in a tight loop, the sendto() function quickly begins to > return a "No buffer space available" error. The program which is reading > packets from the PPP interface only reports that about 1% of the packets > even get through. > > If I run a similar set of tests on Linux, none of the UDP packets are > dropped, and sendto returns no errors. > > In another test, I tried using iperf to measure the throughput of a ssh-pppd > tunnel. The most I can get without compression enabled is 1.2 Mbits/sec over > a 10 Mbit link. The same test on Linux results in numbers which are close to > link speed (one of our tests resulted in ~28 Mbits/sec). > > One interesting thing I and my collegues have observed is that if a Linux > box is used for one end of the ssh-pppd tunnel, we get about 7 Mbits/sec > download speed (Linux->Mac OS X) but only about 800 Kbits/sec upload speed > (Mac OS X->Linux). > > All of this seems to indicate that there is some sort of performance problem > between the TCP stack and the PPP interface code, which is causing some sort > of serious delay when the interface picks up events from the stack. > > Any information about this behavior, and suggestions for possible > workarounds/fixes would be greatly appreciated. > > Thanks, > Jamie Wood -- Jean-Edouard BABIN Mail: Jeb(a)jeb.com.fr Web : www.Jeb.com.fr IRC : Jeb @ Undernet _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Very poor PPP interface performance
by Jamie Wood 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thanks, Jamie Wood _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… I have noticed that the performance of the PPP interface in Mac OS X 10.3.5 is not very good at all. I have written a simple program that brings up a PPP interface, configures it using "ifconfig", and reads incoming packets. If I use another program to repeatedly send UDP packets to my newly-created PPP interface in a tight loop, the sendto() function quickly begins to return a "No buffer space available" error. The program which is reading packets from the PPP interface only reports that about 1% of the packets even get through. If I run a similar set of tests on Linux, none of the UDP packets are dropped, and sendto returns no errors. In another test, I tried using iperf to measure the throughput of a ssh-pppd tunnel. The most I can get without compression enabled is 1.2 Mbits/sec over a 10 Mbit link. The same test on Linux results in numbers which are close to link speed (one of our tests resulted in ~28 Mbits/sec). One interesting thing I and my collegues have observed is that if a Linux box is used for one end of the ssh-pppd tunnel, we get about 7 Mbits/sec download speed (Linux->Mac OS X) but only about 800 Kbits/sec upload speed (Mac OS X->Linux). All of this seems to indicate that there is some sort of performance problem between the TCP stack and the PPP interface code, which is causing some sort of serious delay when the interface picks up events from the stack. Any information about this behavior, and suggestions for possible workarounds/fixes would be greatly appreciated. This email sent to site_archiver(a)lists.apple.com

1 0

Re: [APPL:DarwinKernel] Re: type sizes
by Jay A. Kreibich 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Mutt/1.4.1i On Tue, Oct 26, 2004 at 02:59:19PM +1000, Wade Tregaskis scratched on the wall: > >With the move to 64bit on Tiger, I'm wondering how the types will > >change. For example, will sizeof(size_t) == 8 instead of 4 as it is on > >OSX 10.3 (which seems to violate the ISO C spec)? > > Most things will remain the same size - int's are 4 bytes, shorts are > 2, doubles are 8, floats are 4, etc. The main difference is long, > which is 8 bytes instead of 4. Additionally, long long used to be 8 > (under 32-bit PPC)... I'm not sure what it is now (on my mythical G5 ;) > ), although I'd suspect the same (watch out though for when we move to > 128-bit processors ;) ). It's not an ANSI type anyway, so it's > probably implementation-dependant. Don't forget that most of these numbers are going to be process environment specific, and not system-wide. Since Tiger will support running 32-bit applications and 64-bit applications side-by-side, it needs to support both environments and be able to "context switch" between process and system environments. The big question is how the kernel is going to run. Since Apple will be supporting both 32 and 64 bit processors with the Tiger kernel, there is some assumption the kernel itself will remain largely 32 bit, except for the VM system on 64 bit machines. In short, I expect things to work pretty much like they have for every other vendor that has gone through the 32 -> 64 bit transition. We played most of these games with multiple system libs and different process/kernel environments when SGI made the jump 7 years ago. In that case, things were even worse because their "32 bit application in a 64 bit VM" was a third environment, so you had a whole heck of a lot of shared libs sitting around and three LD_LIBRARY_PATH vars you needed to keep straight. > The key changes from the horses mouth (IBM) This is also up on Apple's site: http://www.apple.com/macosx/tiger/64bit.html Tiger supports the industry standard LP64 programming model supported by other 64-bit Unix systems. This means developers can easily port 64-bit code to Tiger. LP64 support in Tiger provides for 64-bit pointer, long, and long long but preserves 32-bit integer data types. -j -- Jay A. Kreibich | Comm. Technologies, R&D jak(a)uiuc.edu | Campus IT & Edu. Svcs. <http://www.uiuc.edu/~jak> | University of Illinois at U/C _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: MacOS X Scheduler
by Brian Bergstrand 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Use 'sysctl kern.clockrate'. On my machine this returns: kern.clockrate: hz = 100, tick = 10000, profhz = 100, stathz = 100 The one you want is 'tick': 10000usecs You can get this value in your app. with a call to sysctl() using the defines in /usr/include/sys/sysctl.h. HTH. On Oct 26, 2004, at 8:23 AM, Pierre Jomain wrote: > Hi all, > > I need to know what is the quantum (in time or cycles) of the Darwin > scheduler built in Mac OS X.3. > I tried to look at the source ... but it's like finding a needle in a > haystack! > > Thanks in advance, > > Pierre > Brian Bergstrand <http://www.bergstrand.org/brian/>, AIM: triryche206 PGP Key: <http://www.bergstrand.org/brian/misc/public_key.txt> I disapprove of what you say, but will defend to the death your right to say it. - Voltaire As of 09:17:04 AM, iTunes is playing "Say Hello To Heaven" from "Temple of the Dog" by "Temple of the Dog" -----BEGIN PGP SIGNATURE----- Version: PGP 8.1 iQA/AwUBQX5QOnnR2Fu2x7aiEQLLYwCgw3SFuVKktL8N5jTlXaIWEpCUVbEAn3C9 kCmZr9/lLnrqAMAYj+C4KtaE =vYYX -----END PGP SIGNATURE----- _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

mach_msg and mach_msg_type_t missing
by Matt Connolly 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thanks, Matt _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Hello all, I'm a newbie (at kernel programming) and am experimenting with a driver that will send messages to a user client. However, when I use 'mach_msg' in my code, it fails to load at run time because the kernel linker cannot find "_mach_msg". I also notice that the 'struct mach_msg_type_t' is missing (along with a whole bunch of things) from the <mach/message.h> header file. Any help, or pointers to some more detailed documentation for using mach_msg to communicate from the kernel to a user client would be much appreciated. This email sent to site_archiver(a)lists.apple.com

1 0

MacOS X Scheduler
by Pierre Jomain 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi all, Thanks in advance, Pierre _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… I need to know what is the quantum (in time or cycles) of the Darwin scheduler built in Mac OS X.3. I tried to look at the source ... but it's like finding a needle in a haystack! This email sent to site_archiver(a)lists.apple.com

1 0

Re: Our FS make Finder hangs on
by yf-263 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.3) Gecko/20040922 Hi, Quinn, Following is output of all bt: (gdb) thread apply all bt Quinn wrote: At 11:37 +0800 22/10/04, yf-263 wrote: Recently we ported a Linux FS to Darwin, and it can works in CLE. But on openning with Finder, it always hangs with following info: (gdb) thread apply all bt _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Thread 2 (process 1137 thread 0x5b33): #0 0x90293508 in getdirentries_exhaustive(int, unsigned long*) () #1 0x9028a6d8 in DCBlock_BuildContents(int, DCBlock*) () #2 0x9029aa60 in DCBlock_Create(VolumeInfo const*, unsigned long, DCBlock**) () #3 0x90290550 in GetIndexedDirectoryItem(VolumeInfo*, unsigned long, unsigned long, unsigned long, char*, unsigned char*) () #4 0x9028c3cc in POSIXMount::getattrsbulk(void*, unsigned long, unsigned long*, unsigned long, unsigned long, FSAttributeInfo*, unsigned char*) () #5 0x9028d310 in PBGetCatalogInfoBulkSync () #6 0x902a2104 in FSGetCatalogInfoBulk () #7 0x9091d494 in THFSPlusIterator::Next(THFSPlusRef&) () #8 0x9092b6dc in THFSPlusIterator::First(THFSPlusRef&) () #9 0x9091d40c in THFSPlusIterator::Next(THFSPlusRef&) () #10 0x9091c4d0 in THFSPlusSynchronizer::Next(THFSPlusStore&) () #11 0x9091f724 in THFSPlusStore::CreateChildren(TChildrenList&, TNodeTask*) () #12 0x9092b610 in TNode::CreateChildren(TNode::StInitialCreate const&, TNodeTask*) () #13 0x90923964 in TNodeOpenSyncTask::OpenSyncTaskProc(void*) () #14 0x902c6da8 in PrivateMPEntryPoint () #15 0x900246e8 in _pthread_body () Thread 1 (process 1137 thread 0x40b): #0 0x90014528 in semaphore_wait_trap () #1 0x900021f0 in pthread_mutex_lock () #2 0x90281798 in TSLockMutex () #3 0x90284584 in IsForkedDirectoryItem(VolumeInfo*, unsigned long, char const*, unsigned char*) () #4 0x90288efc in POSIXMount::getattrscore(bool, char const*, char const*, POSIXNode*, unsigned long, FSAttributeInfo*, unsigned long, unsigned char*) () #5 0x90298584 in POSIXMount::_getattrs(unsigned long, char const*, unsigned long, unsigned long, FSAttributeInfo*, unsigned long, unsigned char*) () #6 0x902888c8 in FSMount::getattrs(unsigned long, char const*, unsigned long, unsigned long, FSAttributeInfo*, unsigned long, unsigned char*) () #7 0x9028ba90 in GetFSRefAttributes(FSMount*, FSRefPrivate const*, unsigned long, FSAttributeInfo*, unsigned long, char*) () #8 0x9028bb3c in PBGetCatalogInfoSync () #9 0x902955d8 in FSGetCatalogInfo () #10 0x902a62e0 in AL_fillInternalAliasRecord () #11 0x902a7e88 in AL_fillAlias () #12 0x902c0c90 in AL_newAliasCommon () #13 0x902cf45c in FSNewAlias () #14 0x0005200c in ?? () #15 0x00051dd0 in ?? () #16 0x000251b4 in ?? () #17 0x0002a898 in ?? () #18 0x0003f100 in ?? () #19 0x0003eae8 in ?? () #20 0x0003e92c in ?? () #21 0x0002c36c in ?? () #22 0x000439d0 in ?? () #23 0x00073dc8 in ?? () #24 0x0005c71c in ?? () #25 0x0002734c in ?? () #26 0x0009da18 in ?? () #27 0x0005c128 in ?? () #28 0x0002fe3c in ?? () #29 0x0005ba60 in ?? () #30 0x0005b8b4 in ?? () #31 0x0005b874 in ?? () #32 0x0005b75c in ?? () #33 0x0005b100 in ?? () #34 0x0000f954 in ?? () #35 0x00021990 in ?? () #36 0x0000a078 in ?? () #37 0x0000ca54 in ?? () #38 0x927d2330 in DispatchEventToHandlers () #39 0x927d25a4 in SendEventToEventTargetInternal () #40 0x927e4a34 in SendEventToEventTarget () #41 0x927f31b4 in HandleMouseEventForWindow(OpaqueWindowPtr*, OpaqueEventRef*, unsigned short) () #42 0x927e8ad0 in HandleMouseEvent(OpaqueEventRef*) () #43 0x927e2fd4 in ToolboxEventDispatcherHandler(OpaqueEventHandlerCallRef*, OpaqueEventRef*, void*) () #44 0x927d23ec in DispatchEventToHandlers () #45 0x927d25a4 in SendEventToEventTargetInternal () #46 0x927e4a34 in SendEventToEventTarget () #47 0x927e892c in ToolboxEventDispatcher () #48 0x92805c5c in HLTBEventDispatcher () #49 0x927fbaf8 in RunApplicationEventLoop () #50 0x00035798 in ?? () #51 0x00034090 in ?? () #52 0x0003313c in ?? () #53 0x00032fb0 in ?? () (gdb) Can anyone give us some hints about it here ? This backtrace isn't very useful because it only shows one thread, and that thread is blocked waiting for a mutex that arbitrates access to Core Services File Manager's AppleDouble code. Can you post the results of: S+E This email sent to site_archiver(a)lists.apple.com

1 0

Re: Our FS make Finder hangs on
by yf-263 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.3) Gecko/20040922 Oh, sorry, That didn't resolve all problems, now the mountpoint can only be enterred; but can't list anything. does here means I need to implement the ??_getdirattr() ? Reply is very welcomed and appreciated :) /yftty Quinn wrote: At 18:42 +0800 22/10/04, yf-263 wrote: I feel the alias "" fetched via _getattrlist(), or I'm wrong here. And during that time, when I want to gdb -p Finder_pid. The box hang ! What's the tricky then ? If the GUI is unresponsive, SSH into the machine and run GDB from there. If you can't SSH, you should attach with the kernel debugger. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Can you tell me what I omitted ? Making aliases work properly is a tricky business. It's definitely something that I'd defer investigating until you have made more progress on having the volume work in the Finder. S+E This email sent to site_archiver(a)lists.apple.com

1 0

Re: Our FS make Finder hangs on
by yf-263 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.3) Gecko/20040922 Hi, Quinn, With your following article http://developer.apple.com/qa/qa2004/qa1351.html I got Finder work well with our FS now :) Thank you very much. /yftty Quinn wrote: At 18:42 +0800 22/10/04, yf-263 wrote: I feel the alias "" fetched via _getattrlist(), or I'm wrong here. And during that time, when I want to gdb -p Finder_pid. The box hang ! What's the tricky then ? If the GUI is unresponsive, SSH into the machine and run GDB from there. If you can't SSH, you should attach with the kernel debugger. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Can you tell me what I omitted ? Making aliases work properly is a tricky business. It's definitely something that I'd defer investigating until you have made more progress on having the volume work in the Finder. S+E This email sent to site_archiver(a)lists.apple.com

1 0

Re: type sizes
by Wade Tregaskis 26 Oct '04

26 Oct '04

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… With the move to 64bit on Tiger, I'm wondering how the types will change. For example, will sizeof(size_t) == 8 instead of 4 as it is on OSX 10.3 (which seems to violate the ISO C spec)? Most things will remain the same size - int's are 4 bytes, shorts are 2, doubles are 8, floats are 4, etc. The main difference is long, which is 8 bytes instead of 4. Additionally, long long used to be 8 (under 32-bit PPC)... I'm not sure what it is now (on my mythical G5 ;) ), although I'd suspect the same (watch out though for when we move to 128-bit processors ;) ). It's not an ANSI type anyway, so it's probably implementation-dependant. The key changes from the horses mouth (IBM) are just pointer sizes (obviously 8 bytes instead of 4) and long's, which apparently are supposed to always match the size of pointers - I'd never heard this before, but it seems reasonable enough (and now that I look, this is what all the BSD's expect). So anyone converting pointers to ints for whatever reason is going to have headaches. Meaning, quite a lot of people. For completely safety they should be using intptr_t anyway (or uintptr_t). So, knowing that, you should be able to trace down to size_t or whatever other system type you like. It appears on Darwin size_t is a _BSD_SIZE_T_, which is roughly equitable to unsigned long (although it could also be __SIZE_TYPE__, which isn't defined anywhere in the system headers... so I guess not). So, size_t will be 8 bytes on a 64-bit system. Since this is defined in <ppc/ansi.h>, I'm guessing it's ANSI C valid. If not, file a bug report, but I wouldn't expect the behaviour to change any time soon. Wade Tregaskis (AIM, Yahoo & Skype: wadetregaskis, ICQ: 40056898, MSN & email: wjtregaskis(a)students.latrobe.edu.au, Jabber: wadetregaskis(a)jabber.org -- Sed quis custodiet ipsos custodes? This email sent to site_archiver(a)lists.apple.com

1 0