| 06:28.53 | *** join/#asterisk-dev philipp64|laptop (~chatzilla@66-232-81-197.ctcweb.net) |
| 08:19.01 | *** join/#asterisk-dev schmidts (~schmidts@Lo20-rt01.lm33.sil.at) |
| 08:19.04 | schmidts | good morning |
| 09:21.06 | schmidts | i want to ask you what you think about an idea i have. i want to build an own thread which allocates and stores memory and just hand over the pointers to this memory if its need. i allready tried this in a very small range to allocate 1000 buckets for an ao2 container and then just give back these allready allocated memory pointers instead of allocate a new bucket every time an object is linked into an ao2_container |
| 09:21.31 | schmidts | by using this very simple version of my idea i was able to increase the cps rate for sip, from 1750 up to 2750 cps |
| 09:22.06 | schmidts | and i thought that there might be a lot of other stuff like sip_pvt structures or even rtp frames which could use this allready allocated memory instead of alloc it directly |
| 09:24.49 | schmidts | i think when i have an own thread which handles the memory allocations and just givie back pointers when needed, it should be easy to "learn" how much a system would need and allready have the right amount of memory allocated to serve the needs |
| 09:33.00 | wdoekes2 | aren't you then replicating the efforts of say the hoard allocator? |
| 09:33.38 | wdoekes2 | have you tried what that does to your cps rate? |
| 09:36.44 | schmidts | wdoekes2 i will take a look of it, sounds interesting ;) |
| 09:48.22 | schmidts | wdoekes2 maybe it dont understand it cause i have never head of hoard before but on http://www.hoard.org/ i see this: |
| 09:48.29 | schmidts | Open source projects using Hoard include the Asterisk Open Source Telephony Project, |
| 09:48.34 | schmidts | ? |
| 09:49.41 | schmidts | sorry forget it |
| 09:50.11 | schmidts | its a compiler flag ;) |
| 10:00.11 | schmidts | using hoard is a little bit more complicated than i thought :( why isnt there a debian package for it? |
| 10:09.20 | *** join/#asterisk-dev snuff-home (~snuffy@203-219-29-182.tpgi.com.au) |
| 10:09.20 | *** mode/#asterisk-dev [+o snuff-home] by ChanServ |
| 10:11.36 | schmidts | wdoekes2 ok using hoard increase it the same amount but i am now looking if both my test and hoard will increase it even further |
| 10:14.03 | schmidts | wdoekes2 i can reach 2750 cps with normal asterisk source + hoard using and 3100 cps with my changes + hoard |
| 10:15.31 | *** join/#asterisk-dev razu (~razu@2001:ad0:1:1:202:b3ff:fe36:921c) |
| 10:15.43 | schmidts | but right now i still have a very weak solution which allocates 1000 buckets at once when the list is empty, so for every 1000 link into a container it takes a lot of time, thats why i want to use a thread for it |
| 10:19.53 | *** join/#asterisk-dev sgimeno (~sgimeno@163.117.206.10) |
| 10:29.23 | wdoekes2 | why does allocating a 1000 buckets take so much time? do you have to initialize them at alloc time? |
| 10:35.15 | schmidts | what i have done now is just a function which holds a list of free bucket entries and when i need one i just call this function. if the list is empty i allocate 1000 new bucket entries |
| 10:40.09 | wdoekes2 | yes.. but isn't allocating a matter of calling a single malloc on the buffer and perhaps calloc/realloc on a list that keeps track of what is used/unused? or is even that taking up that much time? |
| 10:40.51 | wdoekes2 | oh, you're using a linked list? |
| 10:49.21 | schmidts | the problem with bucket entries is that i cannot reuse it easily, i havent tried it this much, but i had some problems cause normaly the bucket entries / astobj is unlinked from the container and then just freed after this. |
| 10:49.44 | schmidts | and yes i am using a linked list |
| 10:50.15 | schmidts | cause bucket entries are allready designed to be used in a list so i dont have to change them |
| 10:53.21 | schmidts | http://pastebin.com/unneRwYw |
| 10:53.32 | schmidts | this is what i am trying |
| 10:54.22 | wdoekes2 | p = ast_malloc(sizeof(*p)); <-- that's the expensive bit |
| 10:54.27 | wdoekes2 | (I think) |
| 10:54.37 | wdoekes2 | why aren't you doing that outside the for? |
| 10:54.48 | schmidts | do you think there is a difference between a calloc and a malloc? |
| 10:54.59 | wdoekes2 | no.. sizeof(*p) * BUCKETS_TO_REFILL |
| 10:55.17 | wdoekes2 | malloc that at once and profit |
| 10:55.27 | schmidts | good idea ;) |
| 11:03.13 | snuff-home | calloc = zero's mem.. vs malloc unknown entity |
| 11:03.42 | snuff-home | so for large structure calloc can be expensive |
| 11:03.46 | snuff-home | goes back to bf3 |
| 11:05.08 | schmidts | snuff-home have fun :P |
| 11:33.02 | *** join/#asterisk-dev kpfleming (~kpfleming@asterisk/digium-director-of-software-tech/kpfleming) |
| 11:33.02 | *** mode/#asterisk-dev [+o kpfleming] by ChanServ |
| 11:43.03 | *** join/#asterisk-dev ccesario (~ccesario@linux.unialco.com.br) |
| 11:51.29 | *** join/#asterisk-dev F|shie (~chatzilla@182.177.51.96) |
| 11:54.28 | *** join/#asterisk-dev pnlarsson (~niklas@fw1.gml.g.icnet.infracom.se) |
| 12:43.36 | *** join/#asterisk-dev w9sh (~chatzilla@64.238.96.125) |
| 12:45.04 | schmidts | wdoekes2 i am stuck with my thoughts, maybe you can help me ;) when i use malloc like this: p = malloc(sizeof(struct bucket_entry) * 100); then i should be able to get the next struct with p++ right? |
| 12:50.00 | kpfleming | how was 'p' declared? |
| 12:51.23 | schmidts | kpfleming http://pastebin.com/1P24rBav here is the whole function |
| 12:53.45 | kpfleming | well... the simple answer to your question is 'yes', although that code has a number of issues |
| 12:53.55 | kpfleming | first, there is no reason to set p to NULL before calling malloc() |
| 12:54.12 | schmidts | yes i allready have removed this, this was left when i was doing the alloc in the loop before |
| 12:54.28 | kpfleming | second, you should use something like 'p = ast_malloc(100 * sizeof(*p))', to ensure that you don't have to match the type of p (let the compiler do it for you) |
| 12:54.56 | kpfleming | third, yes, you can increment p to point to the next element in the allocated array... but p is not going to magically become NULL at the end of the array. |
| 12:55.00 | schmidts | i had this before and just changed it to make sure |
| 12:55.08 | schmidts | aaah ok ;) |
| 12:55.20 | kpfleming | p is a pointer, it will point to wherever you tell it to point. if you want to iterate over the array, you have to know how many elements there are. |
| 12:55.31 | schmidts | ok got it ;) |
| 12:55.33 | schmidts | thank you kevin |
| 12:55.37 | kpfleming | sure! |
| 12:56.35 | schmidts | ok didnt helped |
| 12:57.23 | schmidts | what i can see is that when internal_ao2_callback tries to do an ast_free on one of this bucket_entries, asterisk dies |
| 12:58.06 | schmidts | this allways happens on the same place, when asterisk first unlink one object from a container and calls ast_free on the actual bucket_entrie object |
| 12:58.39 | *** join/#asterisk-dev anonymouz666 (~anonymouz@189.25.35.63) |
| 13:05.14 | *** join/#asterisk-dev hehol (~hehol@buero-gw.dortmund.loca.net) |
| 13:08.47 | *** join/#asterisk-dev elred_ (~elred_@unaffiliated/elred-/x-5010831) |
| 13:11.39 | kpfleming | so the ao2 internals can't handle what you are doing... they expect every bucket_entry to be its own allocation |
| 13:12.36 | kpfleming | you'll have to add some sort of flag to the bucket_entry itself to indicate that it was not allocated separately, and can't be freed. of course, when it is no longer in use, it will have to be put back on some sort of 'free' list so it can be reused. |
| 13:18.27 | *** join/#asterisk-dev mjordan (~mjordan@nat/digium/x-fuycqddvsqtoeena) |
| 13:18.27 | *** mode/#asterisk-dev [+o mjordan] by ChanServ |
| 13:39.58 | *** join/#asterisk-dev mjordan (~mjordan@nat/digium/x-ztjpxfnoqutrtgzc) |
| 13:39.58 | *** mode/#asterisk-dev [+o mjordan] by ChanServ |
| 13:41.58 | *** join/#asterisk-dev leifmadsen (~Leif@asterisk/documenteur-extraordinaire/blitzrage) |
| 13:41.58 | *** mode/#asterisk-dev [+o leifmadsen] by ChanServ |
| 16:35.50 | *** join/#asterisk-dev infobot (~infobot@rikers.org) |
| 16:35.50 | *** topic/#asterisk-dev is Asterisk Development Discussion -=- http://www.asterisk.org/developers -=- Tier 2 and 3.14159265 support is in #asterisk -=- Tier 4 support is in #2,000 |
| 16:37.12 | *** join/#asterisk-dev anthm (~anthm@freeswitch/developer/anthm) |
| 16:39.30 | *** join/#asterisk-dev frawd (~francois@221.red-80-28-139.adsl.static.ccgg.telefonica.net) |
| 16:45.13 | *** join/#asterisk-dev F|shie (~chatzilla@182.177.32.55) |
| 17:00.20 | seanbright | putnopvut: thanks |
| 17:00.22 | seanbright | updating |
| 17:05.18 | *** join/#asterisk-dev otherwiseguy (~otherwise@69.76.221.79) |
| 18:13.49 | *** join/#asterisk-dev frawd (~francois@221.red-80-28-139.adsl.static.ccgg.telefonica.net) |
| 18:57.13 | *** join/#asterisk-dev moy (~moy@216.172.42.74) |
| 19:28.31 | *** join/#asterisk-dev jeffg (~jeffg@pdpc/supporter/professional/jeffg) |
| 19:47.58 | kpfleming | seanbright: unless you are planning to do a bunch of work, you'll have to disable support for IAX2 transfers when IPv6 is in use |
| 19:51.48 | Qwell | kpfleming: what makes transfers tricky? |
| 19:52.15 | kpfleming | the IP addresses are communicated in the signaling, and the existing IEs can only carry IPv4 addresses |
| 19:53.03 | Qwell | the apparent address IE talks about IPv6 in the RFC |
| 19:53.16 | kpfleming | huh |
| 19:53.25 | Qwell | I mentioned the IE earlier, and I was corrected |
| 19:53.38 | Qwell | it's got a v6 diagram and everything |
| 19:53.49 | Qwell | (page 69) |
| 19:53.50 | kpfleming | well there you go |
| 19:55.33 | putnopvut | seanbright: thanks for getting that fixed up |
| 19:58.22 | kpfleming | jrose_atDigium: r357051 doesn't reference a review or any issue numbers in the commit message |
| 20:01.09 | jrose_atDigium | kpfleming: That's because it's a simple patch which isn't tied to a review and isn't really in response to any issue. |
| 20:01.15 | jrose_atDigium | I had Richard check the patch. |
| 20:21.16 | seanbright | putnopvut: thanks for the feedback, sorry it didn't go in RB first |
| 20:21.51 | putnopvut | you will be spared this time |
| 20:22.08 | seanbright | phew |
| 20:22.35 | seanbright | blindly commits iax ipv6 changes to trunk |
| 20:24.15 | file | blindly commits seanbright to rewriting Asterisk in perl |
| 20:25.59 | seanbright | annnnnd go |
| 20:43.24 | seanbright | so netsock2.h says: "It is important to always initialize ast_sockaddr before use" |
| 20:43.49 | seanbright | but i feel like: struct ast_sockaddr foo = {{0}}; |
| 20:43.53 | seanbright | is ugly as hell |
| 20:44.39 | seanbright | i guess i could make AST_SOCKADDR_INIT |
| 20:47.01 | file | delete seanbright; |
| 20:47.45 | seanbright | thanks for the feedback |
| 20:47.48 | seanbright | douche |
| 20:47.58 | file | <3 |
| 21:02.57 | *** join/#asterisk-dev Georger (~Georger@79.103.238.193.dsl.dyn.forthnet.gr) |
| 21:03.03 | Georger | hi |
| 21:03.10 | Georger | i already posted in asterisk but no reply |
| 21:03.49 | Georger | http://justpaste.it/rkn |
| 21:04.43 | seanbright | [#asterisk-dev] Welcome to #asterisk-dev! This channel is for discussion about the Asterisk source code. If you are looking for support, please use #asterisk. Thanks! |
| 21:27.12 | pabelanger | Odd, testsuite tests seem to fail the dependency check for the TESTSUITE and DONT_OPTIMIZE compiler flag. |
| 21:27.23 | pabelanger | time to fix that |
| 21:28.24 | putnopvut | seanbright: excellent branch name |
| 21:29.24 | *** part/#asterisk-dev mjordan (~mjordan@nat/digium/x-nvwfwxusojxiwsqg) |
| 21:29.39 | seanbright | putnopvut: thank you, sir. |
| 21:29.42 | jrose_atDigium | JunK-Y: Hey, you around? |
| 21:41.18 | pabelanger | Q: is https://issues.asterisk.org/jira/browse/ASTERISK-19289 related to gcc and weak ref support (or what ever it is referred to as)? |
| 21:42.53 | kpfleming | jrose_atDigium: ahh, ok, i confused it with the patch you had on reviewboard. never mind :-) |
| 21:43.09 | jrose_atDigium | kpfleming: No problem |
| 21:44.09 | pabelanger | basically chan_iax2 is still compile, but res_crypto.so is not because libssl-dev is missing. |
| 21:44.29 | pabelanger | then chan_iax2 fails to load, because undefined symbol: ast_aes_set_encrypt_key is missing |
| 21:45.46 | JunK-Y | jrose_atDigium: yep |
| 21:50.51 | kpfleming | pabelanger: it could be related yes, but i'd really prefer to see the failure occur without 'live-ast' being used |
| 21:52.00 | pabelanger | kpfleming: Ya, this is a clean compile under Ubuntu 12.04, no additional scripts are used. |
| 21:52.08 | pabelanger | WAIT one |
| 21:52.17 | pabelanger | actually, I am wrong |
| 21:52.21 | pabelanger | you are right |
| 21:52.26 | pabelanger | let me reproduce it |
| 21:52.53 | kpfleming | thanks |
| 22:19.31 | kpfleming | pabelanger: http://blogs.atlassian.com/2012/02/artifact-sharing-build-automation-bamboo/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+AtlassianDevToolsBlog+%28Atlassian+Products+-+Dev+Tools%29 |
| 22:34.38 | leifmadsen | omg I see kpfleming |
| 22:34.47 | kpfleming | scary |
| 22:34.53 | kpfleming | seanbright too |
| 22:34.56 | leifmadsen | from my perspective... perhaps |
| 22:35.40 | leifmadsen | assumes when he sees kpfleming it must be near the end of the day |
| 22:35.59 | kpfleming | heh... i just finally dug my webcam out of my backpack and put it on the stand |
| 22:36.03 | kpfleming | but yeah... that too |
| 22:36.07 | leifmadsen | heh :) |
| 22:36.18 | leifmadsen | you take ti with you on a trip recently? |
| 22:36.23 | kpfleming | belgium |
| 22:37.28 | leifmadsen | ah gotcha |
| 22:37.36 | leifmadsen | how was that conference? Fosdem right? |
| 22:38.57 | kpfleming | yep. as usual, it was overwhelming and enjoyable... but this year, it was also disturbingly cold. |
| 22:41.51 | leifmadsen | seanbright: nice :) |
| 22:42.03 | seanbright | i ordered 10 of them in various sizes |
| 22:42.11 | seanbright | so i can gain weight as necessary |
| 22:42.27 | leifmadsen | seanbright: :) |
| 22:43.43 | seanbright | ok, i'm out. |
| 22:45.48 | *** join/#asterisk-dev jeffg (~jeffg@pdpc/supporter/professional/jeffg) |
| 23:10.16 | pabelanger | kpfleming: ya, I've been using it more recently. Seems to work good so far |
| 23:10.29 | kpfleming | pabelanger: cool |
| 23:33.29 | *** join/#asterisk-dev sruffell (~Adium@asterisk/the-kernel-guy/sruffell) |
| 23:33.29 | *** mode/#asterisk-dev [+o sruffell] by ChanServ |
| 23:37.48 | *** join/#asterisk-dev sruffell (~Adium@asterisk/the-kernel-guy/sruffell) |
| 23:37.49 | *** mode/#asterisk-dev [+o sruffell] by ChanServ |