Chat log from the meeting on 2016-09-06

From OpenSimulator

Jump to: navigation, search

[11:08] Andrew.Hellershanks How is everyone today?
[11:08] Andrew.Hellershanks Any new issues since last week?
[11:08] Gavin.Hird Doing good here, and you?
[11:09] James atLLOUD: Did a bunch of yardwork of the weekend - soreness is notable.
[11:09] Andrew.Hellershanks Doing ok. Just fighting a minor issue in a program I've been working on over the last week.
[11:09] Gavin.Hird I've been testing the code as of yesterday morning and it does good so far
[11:10] Andrew.Hellershanks Good.
[11:11] Andrew.Hellershanks There were several changes to the scripting engine in the past week.
[11:13] Gavin.Hird all scripts I have seem OK, but I have not done any systematic testing
[11:15] Robert Adams: kind of quiet... how are things going these days?
[11:15] James atLLOUD: @scripting engine - would this be in the git?
[11:16] Robert Adams: I've been away from my computer and will be some more but I wanted to check on things
[11:16] Andrew.Hellershanks Some of the other changes in this past week were some changes to the (mostly) ubODE physics engine and removal of some megaregions code from the physics engines.
[11:16] Andrew.Hellershanks James, yes
[11:16] Robert Adams: Ubit has been madly fixing things (cheering noises)
[11:17] Ubit Umarov: that was dead code already
[11:17] Andrew.Hellershanks He has indeed.
[11:17] Ubit Umarov: :)
[11:17] Andrew.Hellershanks He has been a one man coding army the last few weeks.
[11:17] Gavin.Hird he's been very busy - Impressive again
[11:17] James atLLOUD: OK thanks. Megaregions also noted in the OSgrid news page.
[11:17] James atLLOUD: (cheering too)
[11:18] Ubit Umarov: ( wait to see the bugs :p )
[11:19] Andrew.Hellershanks ok. I missed the exact commit(s) that removed MegaRegion support other than seeing the recent ones that affected the physics engine.
[11:19] Robert Adams: megaregions had their problems... but it's not like varregions don't have any rough corners... but varregions are a much better solution
[11:20] Andrew.Hellershanks nods to Robert.
[11:20] Ubit Umarov: main issue on vars is the single set of terrain textures
[11:20] Andrew.Hellershanks There are still a number of issues with vars. With mega support disappearing more attention can be given to vars.
[11:20] Robert Adams: I don't think megaregions have been explicitly removed... they are like ninja joints -- mostly gone and broken but pieces still hang around
[11:21] Andrew.Hellershanks I had an issue with a var just today. Loaded an oar file and the number of objects reported for the region didn't make sense.
[11:21] Ubit Umarov: ohh ninja.. im to remove them for ages :)
[11:22] Robert Adams: prim in region counts could easily have problems. there are places where computations are based on the region area
[11:22] Ubit Umarov: but Robert i did a extensive remove of mega regions code
[11:22] Robert Adams: that, and the code is not in one place but is scattered around
[11:22] Robert Adams: that would be a Good Thing, Ubit :)
[11:23] Robert Adams: glad to hear it
[11:24] Andrew.Hellershanks I just logged in to the other region and the stats look ok at the moment. The problem might be viewer specific or it could be that OS isn't sending an update to the viewer after the OAR has finished loading.
[11:25] Robert Adams: and/or it's not counting all the objects loaded by the 'load oar'... I think that uses a different object add path for performance reasons
[11:25] Andrew.Hellershanks Could be.
[11:29] Robert Adams: Wright Plaza doesn't have any neighbors any more... was it moved or did they just go away over time?
[11:29] Andrew.Hellershanks IIRC, it was moved some time ago.
[11:29] Andrew.Hellershanks I think it was either due to some issues or when something was being tested.
[11:30] Andrew.Hellershanks If nebadon was here he might remember what was the reason for it being moved.
[11:33] George Equus: Robert, think was Dan making sure there would never be any neighbouring regions around here, due to "region chatter", creating lag
[11:34] Andrew.Hellershanks oh, right. Region chatter.
[11:34] George Equus: This region is pretty heavy...
[11:34] Andrew.Hellershanks yes, there was an increase in chatter between regions some time ago.
[11:34] George Equus: Region blabla iform.... and so on
[11:35] George Equus: see it in my own server, a lot of "informing" going on a soon as you move
[11:35] James atLLOUD: moving or crossing region?
[11:36] George Equus: crossing
[11:36] James atLLOUD: ok ty
[11:36] George Equus: sorry..
[11:36] James atLLOUD: np - I appreciate the help understanding
[11:37] Robert Adams: there should be some analysis of all the inter-region protocols -- they have kinda grown over time
[11:37] Andrew.Hellershanks Ah yes. The change in region chatter occurred during some of the earlier work on improving region crossings.
[11:38] stiofain mactomais: having neighbour regions up when we run events absolutely kills club when over 12 avs
[11:38] Simulator Version v0.5 ruft: OpenSim Dev ed06bfb: 2016-09-05 13:41:06 +0100 (Unix/Mono)
[11:38] Gavin.Hird that's not good
[11:39] Gavin.Hird this is on dev or 0.8.2?
[11:40] stiofain mactomais: dev
[11:40] Gavin.Hird ah
[11:41] Gavin.Hird are people running viewers with radars scanning across the sim borders?
[11:42] James atLLOUD: I would run radar if exploring a grid.
[11:42] Gavin.Hird I think many have them on regardless
[11:45] stiofain mactomais: no idea gav it started when osg started running 9 we were still on 8,2 and we went from handling 30+ avs to having probs with 7 or 8 update to 9 on our regions let us get up to 12 / 14 then rezz time jumped from few min to 30 mins for some
[11:46] Andrew Hellershanks: Can you still only handle about half the number of avatars as you could under 0.8.2?
[11:47] Gavin.Hird I have made a casual observation that the scene build / refresh time has increased over the last month or so
[11:48] Robert Adams: we should set up a load party on some test region and get some logs and numbers
[11:48] stiofain mactomais: yea andrew we dumped 10k prims and replace surrounding regions with voids for shows but still get probs
[11:48] Robert Adams: would have to add some stats for inter-region messages (not sure if that exists) so we can see where any load is coming from
[11:49] Andrew Hellershanks: Sounds like time to do run more load tests but using the 0.9 code. What that problem the OSCC regions would have to stay on 0.8.2
[11:49] Ubit Umarov: inter regions protocols are as bad as before... they need a a udp part
[11:49] James atLLOUD: had the same question @oscc.
[11:50] James atLLOUD: HG safari went to some well attended social things last spring - thought that region was v.9
[11:50] Ubit Umarov: but didn't change that much, except a bit on crossings information
[11:51] Ubit Umarov: and there isn't that much other that crossings
[11:51] Ubit Umarov: only a too slow child positions update
[11:51] Ubit Umarov: that should had been made using UDP from start
[11:52] stiofain mactomais: well we usually have 20+ on sundays at 2pm anyone is welcome to observe
[11:52] Ubit Umarov: the friends, IMs.. etc yeack
[11:52] George Equus: Andrew, OSCC staying on older is mostly playing it safe by Admins I think, ran really well last year, too short time now to switch and find bugs  :)
[11:53] Andrew Hellershanks: yea. That will be something for next year.
[11:53] stiofain mactomais: dev would need similar load tests to before last yr
[11:54] stiofain mactomais: tbh move to 9 has been total nightmare for us
[11:54] Andrew Hellershanks: Those load tests really helped identify a number of issues in the code.
[11:54] Ubit Umarov: why a nightmare?
[11:55] George Equus: Terrible lag when 12+ avatars
[11:55] George Equus: 11 is OK 12 is disaster
[11:55] Ubit Umarov: what kind of problems are you having with the migration to 0.9 ?
[11:56] Robert Adams: last year we spent several months before OSCC doing load testing and smashing performance bugs
[11:56] Robert Adams: there would be load tests just after this meeting, as I remember
[11:56] stiofain mactomais: as quick summary we went from being able handle 30+ avs on 15k prim region with hich prim neighbours to iit falling over and load time going from few mins to 30 mins after 12/14 avs
[11:56] Andrew Hellershanks: Robert: Yes, the load tests were run after these meetings.
[11:57] Robert Adams: when is OSCC?
[11:57] Andrew Hellershanks: Robert, December.
[11:57] Gavin.Hird speaking of 11 vs 12 avatars. I remember testing one of the dance floors with NCPs a few weeks ago, and noticed a significant decrease in performance when adding NPC #12
[11:57] Ubit Umarov: what kind of fail ?
[11:57] Gavin.Hird I thought it was the viewer
[11:57] stiofain mactomais: ah gav that is when we hit wall
[11:58] Ubit Umarov: what happens at avatar #12?
[11:58] stiofain mactomais: did a load of load tests and always 12 avs broke it
[11:58] Gavin.Hird animations skipping
[11:58] stiofain mactomais: rezz time jumps to 10x normal
[11:58] Ubit Umarov: well i have not seen any mantis about those issues
[11:59] Gavin.Hird NPC #12 failing to rezz
[11:59] stiofain mactomais: rubber banding inside region
[11:59] Andrew Hellershanks: Those issues would need to be fixed for an 0.9 release.
[11:59] Ubit Umarov: assuming they are really region issues
[12:00] Ubit Umarov: did you check the avatar impostors and the new jelly crap ?
[12:00] Gavin.Hird no
[12:00] Gavin.Hird will not have the jelly crap in the OS version of Kokua
[12:00] Andrew Hellershanks: jelly?
[12:00] Gavin.Hird those who impemented it are now scrambling to turn it off
[12:00] Ubit Umarov: well there are 10 avas here
[12:01] stiofain mactomais: Dan is actually gonna start hosting the club so we can get more accurate info
[12:01] Robert Adams: let me add some more
[12:01] Gavin.Hird LL has made a metric called avatar complexity
[12:01] Gavin.Hird all avatars over a certain complexity is shown as jelly dolls
[12:01] Andrew Hellershanks: hehe
[12:01] Andrew Hellershanks: Sounds fun
[12:02] Ubit Umarov: its another variant on impostors Andrew
[12:02] Ubit Umarov: well same goal...
[12:02] Gavin.Hird all sculpted / mesh hairdos turn them all to jelly haha
[12:02] stiofain mactomais: i did bring all this up at the time it started all advice was switch to 9 just
[12:03] James atLLOUD:
[12:03] Gavin.Hird could it be the dancefloor code stiofain?
[12:03] Ubit Umarov: 11 avatars here now
[12:04] Gavin.Hird we are all seated so we don't do much
[12:04] Ubit Umarov: the magic 12 now
[12:05] stiofain mactomais: gav we just use aines danceball and tried one load test where all avs sat but same result
[12:05] Gavin.Hird ok
[12:06] Gavin.Hird what is your ping time between the simulator and the asset server - any idea?
[12:06] Andrew Hellershanks: During the load tests we usually asked people to sit. It might be different if we were all standing and moving about.
[12:07] Ubit Umarov: tests arent that easy to do
[12:07] Andrew Hellershanks: no
[12:07] Ubit Umarov: and analyse
[12:07] Ubit Umarov: 0.9 load time per agent increased
[12:07] Ubit Umarov: but possible not for all at same time
[12:08] Robert Adams: the sitting need was mostly about physics
[12:08] Andrew Hellershanks: There were changes to code to reduce inventory load time. The per agent load time still went up?
[12:08] Ubit Umarov: bc 0.8 whould load all fast for now...
[12:08] stiofain mactomais: not sure gav but is heavy duty server ntel Xeon E5-1650 v3, 6-core
6 Cores @ 3.5 GHz / 3.8 GHz Turbo
16 32 GB RAM (DDR4)
*250 GB SSD Primary Drive, RAID 1
5 TB Outbound Bandwidth
[12:08] Robert Adams: there are now 13 agents here
[12:08] Ubit Umarov: in some points 0.9 does increase load time per agent to avoid having agents with no data
[12:08] Robert Adams: anyone notice problems?
[12:09] Ubit Umarov: textures for example have throttling now
[12:09] Robert Adams: stats say there is lots of spare time on the region
[12:09] Andrew Hellershanks: I didn't notice any issue while walking around here.
[12:09] Ubit Umarov: but keep in mind that opensim regions have more work that sl ones now
[12:10] Ubit Umarov: textures and meshs service for example
[12:10] Gavin.Hird can the texture throttling be turned off with a variable?
[12:10] Ubit Umarov: that whould be absurd gavin.Hird
[12:10] Gavin.Hird no it would not
[12:10] Ubit Umarov: you say so..
[12:10] Gavin.Hird Justincc had some throttling in the server, and it performed significantly better when turned off
[12:11] Robert Adams: I must say, my NVidia 1080 is running 6 viewers with shadows with no problems :)
[12:11] Ubit Umarov: those where udp
[12:12] Ubit Umarov: and was a global one.. that was bad
[12:12] Ubit Umarov: besides udp throttles are now diferent also
[12:12] Andrew Hellershanks: My nVidia 9500GT used to work well but not so much any more. They are putting a higher load on my machine. I think that is due to changes in the viewers.
[12:12] Gavin.Hird there are changes in the viewers too Andrew
[12:13] Robert Adams: if it isn't an absolute 12 AV problem... I wonder what problem stiofain is seeing
[12:13] Gavin.Hird definitely so
[12:13] Gavin.Hird to me it looked like texture trashing of some sort when adding NPC #12
[12:13] Ubit Umarov: 0.9 did try to make "badness" democratic :p
[12:13] Gavin.Hird which is why I suspected the viewer
[12:14] Ubit Umarov: as i said in 0,8 some users could get all http bandwith others none
[12:14] Andrew Hellershanks: stiofain, I would suggest filing a mantis with the issue(s) you have so they won't get forgotten on the way to releasing 0.9
[12:14] Gavin.Hird so if textures are throttled it would be interesting to run it at full blast
[12:14] stiofain mactomais: am surprised happened with npcs too i didnt think to test that
[12:14] Ubit Umarov: you cannot send textures at full blast
[12:14] Ubit Umarov: viewers get into trouble also
[12:15] Gavin.Hird viewers will stall about at around 4Mbit/s
[12:15] Ubit Umarov: you can flood a client tcp connection also
[12:15] stiofain mactomais: ok andrew as i said we have further testing planned with dan
[12:17] Andrew Hellershanks: ok.
[12:17] Andrew Hellershanks: Once you have some more information you can file one or more mantis reports based on your findings..
[12:18] Ubit Umarov: yeap, without reports we don't see the issues
[12:18] Robert Adams: it could be throttling... this region has excellent CPU and networking... possible anything less is causing throttling or retries to kick in
[12:18] Ubit Umarov: i see no retries or packet loss here
[12:19] Ubit Umarov: just constant stream of 160Kb/s
[12:19] Robert Adams: like I said... this place has excellent networking
[12:19] Ubit Umarov: almost the same i seen when i arrived here
[12:20] Gavin.Hird if you have a lot of light sources, animations and avatars moving about in the scene fast (as in dancing) you get an enormous amount of updates going to each viewer
[12:20] Ubit Umarov: yeap
[12:20] Gavin.Hird so a club scene is posibly the most challenging
[12:21] Andrew Hellershanks: Perhaps we should move over to the club after one of these meetings to see the issue
[12:21] Ubit Umarov: my viewer is in trouble
[12:21] Ubit Umarov: not the region
[12:21] Ubit Umarov: 32000 prims here
[12:21] Ubit Umarov: all freebies...
[12:21] Ubit Umarov: don't expect performance miracles
[12:21] Gavin.Hird In what way is the viewer in trouble ubit?
[12:22] stiofain mactomais: sunday at 2 would be best for test then you get usual avs wearing mesh etc
[12:22] Ubit Umarov: had 2 running was with 2fps
[12:22] Gavin.Hird oh
[12:22] Ubit Umarov: back to 14 now
[12:22] Ubit Umarov: my alt was here behind me
[12:23] Ubit Umarov: this is a easy thing to debug
[12:23] Gavin.Hird avatars wearing mesh is always an issue
[12:23] Gavin.Hird there is a limit on the number of concurrent mesh streams
[12:23] Gavin.Hird in th viewer
[12:23] Ubit Umarov: as i said viewers also choke with too much data
[12:24] James atLLOUD: I think my fps dropped to about 8 as new avatars arrived, but worked back up to 12
[12:24] Gavin.Hird so LL made jelly dolls to curb it
[12:25] stiofain mactomais: hop:// is club
[12:26] Andrew Hellershanks: My frame rate kind of sucks. I'm only getting about 12.
[12:26] Andrew Hellershanks: When I had an nVidia 8600 I got much better frame rates.
[12:26] James atLLOUD: I'm on a computer at work - laptop Nvidia
[12:26] Ubit Umarov: 32362 freebie prims around you andrew
[12:26] Ubit Umarov: :)
[12:27] Andrew Hellershanks: Ubit, yes but I also have a low setting for draw distance.
[12:27] Andrew Hellershanks: I have it set to 48m
[12:27] Ubit Umarov: hm ive it to 480m similar :)
[12:28] Andrew Hellershanks: Your video card must be a lot better than mine.
[12:28] James atLLOUD: true, 48m does bump the fps
[12:28] Ubit Umarov: yeap 64m 27fps
[12:28] Ubit Umarov: but still 160kb/s of updates
[12:28] Ubit Umarov: ( of course )
[12:29] Andrew Hellershanks: I wanted to set it to 64m but its so hard to set the limit using that slider.
[12:29] Ubit Umarov: you can type the number :)
[12:29] Andrew Hellershanks: where?
[12:29] Andrew Hellershanks: There is only a slider
[12:29] Ubit Umarov: in fs, there is a entry just in from of the slider
[12:30] Ubit Umarov: front
[12:30] James atLLOUD: I'm using the phototools panel in Firestorm - nice gadgets there.
[12:30] Andrew Hellershanks: Ubit, they have hidden that. I see nothing in front of the slider to indicate there is any input widget.
[12:30] Ubit Umarov: but 48 or 64 same thing :)
[12:30] Ubit Umarov: im in 4.7.9
[12:31] Andrew Hellershanks: ok. its new. I'm in 4.7.7
[12:31] James atLLOUD: new pie menu in 4.9
[12:31] Ubit Umarov: well exact number doesn't matter that much...  :)
[12:32] Andrew Hellershanks: Slider must have been added after 4.7.7
[12:32] Andrew Hellershanks: I'm downloading 4.7.9 now
[12:32] stiofain mactomais: and can't tell ppl coming to a show " you need to reduce draw " " take off mesh" dont move about"
[12:33] Metropolis Testparcel: Your testparcel located in OpenSim2 (225.234000,60.777000,21.338000) will expire in one hour.Reactivation is possible.
[12:34] Andrew Hellershanks: I have 4.7.9 installed. I'll test it later today.
[12:34] James atLLOUD: aw, that sounds lovely - I need to meet co-workers. just not the same
[12:34] stiofain mactomais: ok andrew was an interesting meet
[12:34] Ubit Umarov: don't rush
[12:34] Ubit Umarov: 4.7.7 is better for opensim
[12:34] Andrew Hellershanks: The 4.7.9 is still shown on the website as being for SL and OS
[12:35] Ubit Umarov: .9 is a crap
[12:35] Andrew Hellershanks: If that is the case I better keep both for now.
[12:35] Ubit Umarov: tons of bad messages going on
[12:35] James atLLOUD: thank you Andrew for facilitating
[12:35] Ubit Umarov: i considered several times to "ban it"
[12:35] Andrew Hellershanks: My experience is I have lots of issues with many of the current viewers so it can't get too much worse with a 0.0.2 update
[12:36] Andrew Hellershanks: yw, James.
[12:36] George Equus: Thanks Andrew
[12:36] stiofain mactomais: well i will plan a load test for this sundays show at 2pm
[12:36] Ubit Umarov: i can say that we don't suport fs 4.7.9 actually
[12:37] Andrew Hellershanks: The one thing I'll do is just to check if FS 4.8.9 has the entry widget for setting draw distance
[12:37] Ubit Umarov: and if next version of FS doesn't fix issues we may ban fs
[12:37] stiofain mactomais: bring on onlook
[12:37] James atLLOUD: hmm, ban seems very strong - maybe not recommend?
[12:38] Ubit Umarov: well all opensim bans mean that :)
[12:38] Andrew Hellershanks: Looks like FS messed up the preferences dialog box. The sliders are not even drawing properly.
[12:38] Simulator Version v0.5 ruft: OpenSim Dev ed06bfb: 2016-09-05 13:41:06 +0100 (Unix/Mono)
[12:38] James atLLOUD: OK :)
[12:38] Ubit Umarov: but current fs is a performance problem
[12:38] Ubit Umarov: it does insane repeats of the same requests
[12:39] Ubit Umarov: you don't see it without looking to packets
[12:39] Andrew Hellershanks: ok. I think I'll be sticking with 4.7.7 then.
[12:39] Andrew Hellershanks: It doesn't always work properly for voice chat but is ok for the most part other than voice.
[12:39] Ubit Umarov: 4.7.7 don't use map zoom much
[12:39] James atLLOUD: I'll see about 4.7.9 in Great Canadian Grid this Sunday. Will have a standby ready.
[12:40] Ubit Umarov: it sends a insane repeat of map blocks requests
[12:40] James atLLOUD: I do love me that phototools panel tho  :(
[12:40] Ubit Umarov: i added code to block those for 5 minutes :)
[12:41] Ubit Umarov: so your map will fail for 5 minutes :)
[12:41] Ubit Umarov: but on rest its not as bad as .9 ( that actually fixed that )
[12:41] Ubit Umarov: a example
[12:41] James atLLOUD: I must go to - peeps needs answers in RL
[12:41] Ubit Umarov: if you move the mouse on terrain
[12:42] Ubit Umarov: you get a flood or parcelinformation requests
[12:42] Ubit Umarov: some 10 per second
[12:42] Ubit Umarov: on a large region thats seveal 100KB per request
[12:43] Sheera.Khan that's quite a bit of load ...
[12:43] Ubit Umarov: oh thye all left :)
[12:43] Ubit Umarov: yeap
[12:43] Ubit Umarov: we have a problem now
[12:43] Ubit Umarov: this is LL bugs actually
[12:44] Ubit Umarov: this one i told FS, FS told alchemy Alch filed a LL jira :)
[12:44] Ubit Umarov: several other things i told FS where.. ignored :p
[12:45] Ubit Umarov: since they are LL issues possible other viewers will have them
[12:45] Ubit Umarov: as they copy ll code
[12:45] Ubit Umarov: actually this issues are also bad at SL
[12:45] Ubit Umarov: just they have better machines
[12:46] Ubit Umarov: see the bandwidth use here?
[12:46] Ubit Umarov: 140Kb/S only us
[12:47] Sheera.Khan is that kBit/s or kByte/s ?
[12:47] Ubit Umarov: bit i think
[12:47] Ubit Umarov: viewers say Kbits
[12:48] Sheera.Khan ok...
[12:48] Ubit Umarov: well its about 10bits per byte
[12:48] Ubit Umarov: ( packets overload)
[12:48] Sheera.Khan that's not a big deal then ...
[12:49] Ubit Umarov: on a empty region well :)
[12:49] Ubit Umarov: and 10 avas its 1000Kb/s
[12:49] Sheera.Khan we are supposed to have several megabits/s as users and Gigabits/s at the server
[12:49] Ubit Umarov: forget gigabits
[12:50] Ubit Umarov: we don't have Gbits
[12:50] Ubit Umarov: some ISPs have a few GB for ALL their clients
[12:51] Ubit Umarov: we expect a few Mbits per user avaiable..
[12:51] Ubit Umarov: not much
[12:52] Ubit Umarov: on a data center
[12:52] Ubit Umarov: worse on home regions
[12:52] Sheera.Khan sure, we won't have Gigabit/s guaranteed ...
[12:52] Ubit Umarov: i have 1Mbit/s :)
[12:52] Sheera.Khan but at my home I have 10 MBit/s upload
[12:53] Sheera.Khan and that's usable :-)
[12:53] Ubit Umarov: i've 1
[12:53] Ubit Umarov: no idea how much osg has on the datacenter they use
[12:54] Ubit Umarov: but thats limit per client on datacenters also
[12:54] Ubit Umarov: possible all osg doesn't have more than 100Mb for all machines
[12:54] Ubit Umarov: don't know
[12:55] Sheera.Khan me neither, we would have to ask Nebadon
[12:55] Ubit Umarov: but there is a lot hidden about these things
[12:55] Ubit Umarov: well our code is still bad in any case :p
[12:56] Sheera.Khan but you're working on it and I really appreciate that :-)
[12:56] Ubit Umarov: :)

Personal tools
About This Wiki