Chat log from the meeting on 2018-10-09

From OpenSimulator

Jump to: navigation, search

[10:32] Sheera Khan: I only see unknowns today :-(
[10:34] Bill Blight: there are two files in your viewer cache you can delete that help with that if they get stuck ... , after closing the viewer they are like Avatar name cache .db and .xml
[10:38] LaNani Sundara: yesterday morning (my time) we had a test session on one of my regions, using my code. I had turned off all throttles. Dan was there, so was Ubit and Paela. Plus a lot of people using mesh avatars. It went great, So last night Ubit wrote code that is very similar to mine, but of course his own version of the same idea. I have read his code. He took out the horrible throttle code from the GetMeshModule and GetTextureModule, like i have a year ago
[10:38] LaNani Sundara: :) glad he saw the light finally
[10:39] Bill Blight: :)
[10:40] LaNani Sundara: they have it on Event Plaza and LBSA now
[10:40] Bill Blight: I still think it's a bad idea, but oh well, no throttles means someone with crazy high bandwidth can suck all the bandwidth from the server given the right circumstances
[10:40] LaNani Sundara: he put in a cap
[10:40] LaNani Sundara: there are still throttles
[10:41] Bill Blight: well bad idea is not what I meant , possible cause of trouble is a better description
[10:41] LaNani Sundara: hmm i doubt it
[10:41] LaNani Sundara: hihi
[10:41] LaNani Sundara: i have been running this kind of code for over a year now
[10:42] Bill Blight: the new code is running here
[10:42] LaNani Sundara: seems ok
[10:42] Bill Blight: but it was ok before, so I don't see a difference, but time will tell, this is not a busy region
[10:43] LaNani Sundara: hmm
[10:43] JayR Cela: OS 0.9.1.1 / seems to be ok so far for me
[10:43] LaNani Sundara: its the mesh avies who suffered in the old code
[10:43] Bill Blight: yep, and honestly , I had no issue with that :P
[10:43] LaNani Sundara: that is the issue tho
[10:43] Bill Blight: mainly because they caused others to have viewer lag
[10:44] Bill Blight: why do they get "special" treatment
[10:44] LaNani Sundara: that you have no issue with that is an issue
[10:44] LaNani Sundara: and not true
[10:44] LaNani Sundara: they do not cause the viewer lag
[10:44] LaNani Sundara: the bad code did
[10:44] Bill Blight: we can agree to disagree on that
[10:44] Bill Blight: high poly counts sent to the viewer take longer to render
[10:44] Bill Blight: contain more data
[10:45] Bill Blight: so to people not on high end systems
[10:45] Bill Blight: they cause viewer side lag
[10:45] Bill Blight: but derendering or jelly dolls is always an option there I guess
[10:45] LaNani Sundara: am i cause you lag?
[10:45] Bill Blight: I never lag
[10:45] LaNani Sundara: well then
[10:45] Bill Blight: my system is very high end
[10:45] Bill Blight: and gigabit internet
[10:46] LaNani Sundara: people want to look nice :) hence the mesh avies
[10:46] LaNani Sundara: and they want to rez well
[10:46] Bill Blight: my issue with it is the attitude, "to hell with everyone else as long as I look good and rez fast"
[10:47] Bill Blight: but I'm grumpy today
[10:47] LaNani Sundara: its false sentiment Bill
[10:47] LaNani Sundara: because the mesh avies do not need to cause lag
[10:47] Bill Blight: I don't think it is false as much as you think it is
[10:47] Bill Blight: but again we don't have to agree
[10:47] LaNani Sundara: a lot of the lag was due to the code
[10:48] LaNani Sundara: the throttling actually caused lag
[10:48] Bill Blight: disagreements does not mean we can't have a good debate
[10:48] LaNani Sundara: o i am fine :) dont worry
[10:48] Bill Blight: I had events on my grid regularly with 30 avatars and NOBODY lagged including the mesh avatars, and that was with the throttles, so I don't find that a consistent argument
[10:48] LaNani Sundara: having things load slow on purpose makes it seem like there is lag caused by the things being loaded
[10:49] LaNani Sundara: well you just proved my argument, mesh does not cause lag
[10:49] LaNani Sundara: :D
[10:49] Bill Blight: when I say lag I mean server lag
[10:49] Bill Blight: but they did cause some viewer lag
[10:50] LaNani Sundara: minor
[10:50] Bill Blight: but my people tend not to complain as much as , they come for the people , and don't care if they are a fancy mesh avatar right off the bat or a prim box
[10:50] LaNani Sundara: people wont complain when they do not lag as you say they do not on your server
[10:50] LaNani Sundara: nor do they on my sims
[10:51] Bill Blight: yes mesh avatars took at most 60 seconds to rez but they did rez fully
[10:51] LaNani Sundara: good thats how it should be
[10:51] Bill Blight: but that was with the throttles
[10:51] LaNani Sundara: not the 15 minutes as it was at LBSA
[10:52] LaNani Sundara: well you must have a better server then ;)
[10:52] LaNani Sundara: and Ubit did not take out the UDP bitbucket queue throttles
[10:53] LaNani Sundara: he just took out the horrid code in GetMeshModule and GetTextureModule
[10:53] Bill Blight: well, that is also the point, can't run windows 10 on a 286 , hardware has to grow with the software, and people need to realize it is time to move from 2005 LOL
[10:53] LaNani Sundara: which is tcp
[10:53] LaNani Sundara: so true
[10:54] Bill Blight: I mean some people I have talked to are running their servers on the same systems they ran them on when opensim was born
[10:54] JayR Cela: why are we even running the UDP protocol still ?
[10:55] LaNani Sundara: *nods
[10:55] LaNani Sundara: do a lot of things
[10:55] Bill Blight: also one of the biggest things, is windows networking and even newer linux networking is very anemic on the UDP side, and even tcp windows and TTL's there are some tweaks that can fix that,
[10:55] LaNani Sundara: with UDP still
[10:55] Bill Blight: don't disagree with that
[10:55] Bill Blight: two words ............ The Viewers
[10:56] LaNani Sundara: Ubit will explain ;)
[10:56] Bill Blight sees Ubit get run over by the bus
[10:56] LaNani Sundara: let me teleport him in
[10:57] JayR Cela: i know ubuntu 16.06 and 18.04 / I am having problems with Samba / my 18.04 machines dont see my 16.04 box'x / but vice versa 16.04 can see the 18.04 ones :_(
[10:57] JayR Cela: ubuntu 16.04
[10:58] Bill Blight: use NFS scrap samba
[10:58] Bill Blight: unless you have a specific need to share with a windows device
[10:58] LaNani Sundara: newer samba protocol maybe
[10:58] Bill Blight: but you can actually mount NFS shares on windows these days
[10:58] JayR Cela: oh ? NFS / I not tried that
[10:59] Bill Blight: yes I think 18.04 requires lanmanV2 where 16 could do both
[10:59] Bill Blight: but not sure
[11:06] Andrew.Hellershanks @hg.osgrid.org: My cat was being a nuisance. I think she has given up on trying to get me to play with her (for now).
[11:06] Sheera Khan: it's always the cats ^^
[11:07] Andrew.Hellershanks @hg.osgrid.org: Sheera, I had to stop what I was doing earlier as she wouldn't stop meowing at me.
[11:08] Sheera Khan: that's what cats do ^^
[11:09] Sheera Khan: Meowing till Ubit pulls the throttles ^^
[11:09] Andrew.Hellershanks @hg.osgrid.org: Sheera, yea. Often the moment I sit down to start doing something.
[11:09] Ubit Umarov: pull ?
[11:09] Sheera Khan: pull them out of the code ...
[11:09] Arielle.Popstar @grid.hgluv.com:8302: give them adjustable sliders?
[11:10] Gavin.Hird @grid.xmir.org:8002: git pull
[11:10] Sheera Khan: I hadn't found the time to look into the sources yet ...
[11:10] Andrew.Hellershanks @hg.osgrid.org: Ubit, care to start us off by talking about the recent throttle related changes.
[11:10] Sheera Khan: so I don't know if they are removed, adjusted or made adjustable ...
[11:10] LaNani Sundara: :)
[11:12] Andrew.Hellershanks @hg.osgrid.org: The latter
[11:12] Ubit Umarov: errr? hmm?
[11:13] Ubit Umarov: well i did modified them a bit
[11:13] LaNani Sundara: sorry we were in an Im discussion
[11:13] Ubit Umarov: now there is a single control for textures and meshes per client
[11:14] LaNani Sundara: *nods
[11:14] Ubit Umarov: previous code did it for meshes and textures independently
[11:14] LaNani Sundara: well.. and in a not so nice way ;p
[11:14] LaNani Sundara: *hushes :)
[11:14] Ubit Umarov: ie now if you only ask for textures, you get all the allowed bandwith
[11:14] Ubit Umarov: or meshes...
[11:15] Bill Blight: you still need to push an update for the OpenSimDefaults.ini so people have the option and an explanation
[11:15] Bill Blight: and maybe the OpenSim.ini.example
[11:15] Ubit Umarov: by default the limit is still derived from the user settings for lludp bandwidth
[11:16] LaNani Sundara: how did you arrive at those hard coded default now Ubit, 100K and 50K
[11:16] Andrew.Hellershanks @hg.osgrid.org: If it is a setting that most people should leave alone there is no need to put it in the .ini.example file.
[11:16] Ubit Umarov: now it gets more or less the same as the total bandwidth for udp
[11:16] LaNani Sundara: *nods
[11:16] Bill Blight: that is in mbits so I think it is 10MB and 5MB ...
[11:16] Ubit Umarov: well a bit more, bc that is not exact math
[11:17] LaNani Sundara: i know
[11:17] Ubit Umarov: ( no need to be :) )
[11:17] Ubit Umarov: and yes i did add a option to overide that by a fixed value
[11:17] Bill Blight: I know you knew Lanani but someone reading the notes might think it is a 100K limit
[11:17] LaNani Sundara: and the burst /5 bit?
[11:18] Ubit Umarov: i may tune that /5
[11:18] Andrew.Hellershanks @hg.osgrid.org: I think some clarification is in order. If there is only one setting for textures and meshes why is Bill stating two different rates?
[11:18] LaNani Sundara: why the /5? i just wonder how you deternine those values
[11:18] Ubit Umarov: that allows for small scarse requests to pass without th
[11:19] Bill Blight: if there had been the example in the opensim .ini it would explain that andrew. 5000 is the lower limit you can't set it lower than that
[11:19] LaNani Sundara: k
[11:19] Bill Blight: 10000 is the upper limit "default"
[11:19] Bill Blight: and I may have missed a zero
[11:19] Ubit Umarov: upper limit?
[11:20] Andrew.Hellershanks @hg.osgrid.org: Bill, ok. I thought you were indicating two different settings, one for textures, and one for mesh while Ubit had said there is one setting covering both.
[11:20] Ubit Umarov: hmm i actually did not placed a hardcoded high limit
[11:20] LaNani Sundara: there is now a Cap_AssetThrottle
[11:20] LaNani Sundara: option
[11:20] Ubit Umarov: i did place a lower limit of 20000 :)
[11:20] Bill Blight: + m_bandwidth = 100000;
+ m_lastBandwithTime = Util.GetTimeStamp() + 0.1;
[11:21] LaNani Sundara: if(m_bandwidth < 50000)
+ m_bandwidth = 50000;
[11:21] Ubit Umarov: 100 000 is the start default until viewer sends throttles
[11:21] LaNani Sundara: ok
[11:21] Ubit Umarov: ohh ok not 20k but 50k as lower :)
[11:21] Bill Blight: yeah that is what I meant , just said it wrong
[11:21] LaNani Sundara: yes
[11:21] Ubit Umarov: ( those are bytes per sec )
[11:22] LaNani Sundara: yes
[11:22] LaNani Sundara: i got that
[11:22] Ubit Umarov: in section [ClientStack.LindenCaps]
[11:22] Ubit Umarov: one can add the option you said like
[11:22] LaNani Sundara: cos you kepe track of asset data size
[11:22] LaNani Sundara: keep*
[11:22] Ubit Umarov: ; Cap_AssetThrottle = 500000 ; aprox 5Mbps
[11:23] Sheera Khan: 50000Bytes/s as in 400kBit/s?
[11:23] Ubit Umarov: this does override the automatic guessing from up
[11:23] Ubit Umarov: no more like 5Mbps :)
[11:23] Sheera Khan: did I miss a zero?
[11:23] Bill Blight: 500000 bytes to MB is .5
[11:24] Sheera Khan: ah, I saw 50k in the chat before
[11:24] Ubit Umarov: no you passed from bytes to bits
[11:24] LaNani Sundara: are you still using that limit or the number of textures of the canibalizing option, or is that purely udp?
[11:24] Ubit Umarov: think i did killed the canab code, and yes that is udp only
[11:25] Sheera Khan: yes I did that intentionally as network speeds are mostly specified in MBits/s
[11:25] Ubit Umarov: so you need to x8 no ?
[11:25] LaNani Sundara: also i recall gettextureModule had this silly rule of only max 4 textures in the queue while getmesh did not
[11:26] Sheera Khan: that's why I went from 50kByte/s to 0.4MBit/s
[11:26] Ubit Umarov: so 500 000 is 4 000 000 bits right ?
[11:26] Bill Blight: ok makes sense
[11:26] Bill Blight flüstert: for the MB to Mb challenged like me ..
[11:26] Bill Blight: https://convertlive.com/u/convert/bytes/to/megabits#500000
[11:26] LaNani Sundara: if (m_queue.Count >= 4)
{
// Never allow more than 4 fetches to wait
reqinfo.send503 = true;
[11:26] LaNani Sundara: this
[11:26] LaNani Sundara: why?
[11:26] Ubit Umarov: i used 10x because of network overloads..
[11:27] LaNani Sundara: its not in the GetMesh version
[11:27] LaNani Sundara: why the 4 max in the queue
[11:27] Ubit Umarov: so i said 500 000 aprox 5MB ( in decimal )
[11:27] LaNani Sundara: i never got that
[11:27] Ubit Umarov: opoes 5Mb
[11:27] Ubit Umarov: :)
[11:28] LaNani Sundara: *waves hand :)
[11:28] Ubit Umarov: just real bandwith use will be higher like 9 to 10x
[11:28] LaNani Sundara: my question please Ubit :) why the only 4 in the queue limit in the GetTextureModule
[11:28] Ubit Umarov: bc as i said network adds more things
[11:29] Ubit Umarov: because reasons lanani
[11:29] LaNani Sundara: LOL?
[11:29] Ubit Umarov: :)
[11:29] LaNani Sundara: what reasons, and why not in the Mesh module then
[11:29] Ubit Umarov: i don't like the viewers option to have 32 requests in parallel
[11:29] Ubit Umarov: that is just silly
[11:29] Ubit Umarov: not even 8
[11:29] LaNani Sundara: you know that that 4 texture limit sends the viewer a response to tell it to try again later?
[11:30] Ubit Umarov: yeap
[11:30] LaNani Sundara: which just makes it pile on again
[11:30] LaNani Sundara: asking for things again
[11:30] Ubit Umarov: yeap
[11:30] LaNani Sundara: how is that good?
[11:30] Ubit Umarov: but not killing all my FDs
[11:30] Arielle.Popstar @grid.hgluv.com:8302: FD's?
[11:30] LaNani Sundara: and why if its a good thing, do you not have the same mechanism in the GetMeshModule
[11:30] Sheera.Khan @hypergrid.org:8002: File descriptors
[11:30] Ubit Umarov: file discriptores
[11:30] Arielle.Popstar @grid.hgluv.com:8302: kk
[11:31] LaNani Sundara: or the GetInventoryDesc one
[11:31] Bill Blight: is it giving room for other requests in there or is that limit client specific
[11:31] Bill Blight: if it is client specific I can understand
[11:31] Ubit Umarov: there is a limit on the max number of opened files ( or tcp) connections
[11:31] LaNani Sundara: look arbitrary to me if its only for textures
[11:31] Bill Blight: stops one client from filling up the queue ahead of others
[11:31] LaNani Sundara: Ubit, its not there for the mesh
[11:32] Sheera.Khan @hypergrid.org:8002: would it be a big thing to make that limit depending on the number of clients?
[11:32] LaNani Sundara: you are basicly telling every client to come back later
[11:32] Sheera.Khan @hypergrid.org:8002: I can see 100 clients draining the pool of FDs available
[11:33] Sheera.Khan @hypergrid.org:8002: but if there are only few avatars we could increase that number per viewer
[11:34] LaNani Sundara: it is a queue guys.. its a limit on the number of requests in the queue, not on the number of files being sent
[11:34] LaNani Sundara: its an extra throttle
[11:34] LaNani Sundara: on accepting requests
[11:34] Gavin.Hird @grid.xmir.org:8002: what files are kept open server side for a texture request when it is hauled from the database?
[11:34] LaNani Sundara: the request itself has no file in it
[11:35] LaNani Sundara: you are already limiting the number of files by the bandwidth throttle basicly
[11:35] Gavin.Hird @grid.xmir.org:8002: I mean from a database blob
[11:36] LaNani Sundara: it will only get the file when it handles the request and creates the response with the data in it
[11:36] Gavin.Hird @grid.xmir.org:8002: it will only get a file when it reaches the client
[11:37] Gavin.Hird @grid.xmir.org:8002: the blob is streamed from the db server to the client without being represented as a disk file
[11:37] Arielle.Popstar @grid.hgluv.com:8302: is there a capacity limit to the amount of requests the server can handle before being flooded?
[11:37] LaNani Sundara: in my view, all this 4 textures requests in the queue limit does is extra traffic cos it keeps telling the viewer, ask me later
[11:38] LaNani Sundara: and 4 ... why 4
[11:38] LaNani Sundara: and like i said why not do it for the mesh as well then
[11:38] Arielle.Popstar @grid.hgluv.com:8302 nudges Ubit to wake him up
[11:39] Gavin.Hird @grid.xmir.org:8002: I think there is a limit of 8 or 16 meshes that can be requested at any time from a viewer
[11:39] LaNani Sundara: i am talking about.. line 246 i think in GetTextureModule
[11:39] Gavin.Hird @grid.xmir.org:8002: some viewer have a debug setting to increase that
[11:40] LaNani Sundara: its probably ancient code nobody looked at in a while ;)
[11:41] LaNani Sundara: i would make it the same as for the Mesh, no weird 4 requests at a time limit on Textures
[11:41] Gavin.Hird @grid.xmir.org:8002: I can't see any reason to cap it at 4 request server side
[11:41] LaNani Sundara: ikr
[11:41] Gavin.Hird @grid.xmir.org:8002: one have to assume the server is a more resourcefull system than the client
[11:41] LaNani Sundara: i have taken that out over a year ago and never had a problem
[11:41] Gavin.Hird @grid.xmir.org:8002: most operating systems these days allows 64k open files per process
[11:42] Sheera.Khan @hypergrid.org:8002: tell that to my raspberry^^
[11:42] Gavin.Hird @grid.xmir.org:8002: although you might have to set it for the session
[11:43] Gavin.Hird @grid.xmir.org:8002: haha - I was thinking of getting a raspberry just for the fun of getting opensim running on it
[11:43] Gavin.Hird @grid.xmir.org:8002: you could build a docker swarm of raspberries
[11:45] Andrew Hellershanks: Talk to nebadon about getting OS running on a Pi.
[11:45] Bill Blight wonders what is going to happen to "Home Users" with allowing a bazillion requests at once via home routers
[11:45] Arielle.Popstar @grid.hgluv.com:8302: good question Bill
[11:45] Bill Blight doesn't care, just doesn't want to listen to the whining
[11:46] Arielle.Popstar @grid.hgluv.com:8302: but one wouldnt run a big grid on a home server
[11:46] LaNani Sundara: Bill the requests are still being sent...
[11:46] LaNani Sundara: the code even responds
[11:46] LaNani Sundara: it says ask again in 30 sec
[11:46] Bill Blight: but does not send data back
[11:46] Bill Blight: other than a 503 I think?
[11:46] LaNani Sundara: no but the requests will be sent again by the viewer
[11:47] Bill Blight: again , does not equal same time
[11:47] LaNani Sundara: more requests then without that limit
[11:47] Ubit Umarov: sorry had a phone call
[11:47] Bill Blight: more data too at once
[11:47] LaNani Sundara: not at once
[11:47] Ubit Umarov: lanani I think i did answer that
[11:47] LaNani Sundara: its still a queue
[11:48] Ubit Umarov: we can't allow the 32 parallel requests per user
[11:48] LaNani Sundara: why?
[11:48] Andrew Hellershanks: It is almost 10 minutes before the top of the hour. I need to go soon after the hour. Does anyone else have a topic beside throttles and bandwidth? If not we can stay on this topic.
[11:48] Bill Blight: you have a million requests, a million with data, or 7500000 with data 250000 with 503's that is a big diff in the data at once
[11:48] Ubit Umarov: bc we can't
[11:48] LaNani Sundara: they are not parallel
[11:48] Ubit Umarov: yes they are
[11:48] LaNani Sundara: there is a queue?
[11:48] Arielle.Popstar @grid.hgluv.com:8302: i do but this is more important at the moment
[11:48] Ubit Umarov: its 32 FDs
[11:48] Gavin.Hird @grid.xmir.org:8002: the viewer will issue parallel requests
[11:49] LaNani Sundara: this is about the number of requests in the queue
[11:49] LaNani Sundara: look... i do not have that limit in my version... and have not had it in more that a year on 30 or so inhabited regions
[11:49] Ubit Umarov: now they can do parallel and pipeline
[11:49] Gavin.Hird @grid.xmir.org:8002: there is a debug setting for most viewers: mesh2MaxConcurrentRequests
[11:50] LaNani Sundara: on regions where ppl come to party
[11:50] Gavin.Hird @grid.xmir.org:8002: it is set default to 8
[11:50] Bill Blight: this is on textures only Gavin
[11:50] Ubit Umarov: yes gavin and can go up to 32 if i remember
[11:50] LaNani Sundara: yes also why its weird, its only on textures
[11:50] Gavin.Hird @grid.xmir.org:8002: yes 32
[11:51] Ubit Umarov: think the impact of that on number of allocated FDs
[11:51] LaNani Sundara: i say take it off and see what happens.... we had big fundraiser events on regions without that limit
[11:51] Ubit Umarov: also consider the close time of a fd
[11:51] Gavin.Hird @grid.xmir.org:8002: why does it allocate FDs?
[11:51] LaNani Sundara: yeah why, its just data like mesh
[11:51] Ubit Umarov: each tcp connection eats a fd of course
[11:51] Gavin.Hird @grid.xmir.org:8002: where?
[11:51] Bill Blight: maybe make it a config option, because the network guy in me thinks this could kill home users and small grids
[11:52] Ubit Umarov: where where??
[11:52] LaNani Sundara: so... again why then only put that limit on the textures, not on the mesh or the inventory descriptions
[11:52] Gavin.Hird @grid.xmir.org:8002: where are the FDs allocated?
[11:52] Gavin.Hird @grid.xmir.org:8002: in Robust?
[11:52] LaNani Sundara: its only in the GetTextureModule
[11:52] Gavin.Hird @grid.xmir.org:8002: cause the db server sure does not do it
[11:52] Ubit Umarov: (windows hides or uses other name for fds)
[11:52] LaNani Sundara: whats so special about FDs from Textures
[11:52] Ubit Umarov: region
[11:53] Ubit Umarov: and all connections do that
[11:53] Gavin.Hird @grid.xmir.org:8002: region on the simulator?
[11:53] Ubit Umarov: just they try to keep some to reuse
[11:53] Ubit Umarov: everywhere
[11:53] Ubit Umarov: but on this we are talking regions
[11:54] Ubit Umarov: just type netstat on similar on win or linux and see
[11:54] Sheera.Khan @hypergrid.org:8002: could one increase the number of FDs per process in the OS? Like setting an env-var?
[11:54] Gavin.Hird @grid.xmir.org:8002: and it create a FD on the region why?
[11:54] Gavin.Hird @grid.xmir.org:8002: because it needs to be writtend to Flotsam?
[11:54] LaNani Sundara: 30 regions, big ones, small ones, plus other people using my code base... all without that 4 request limit....
[11:54] Ubit Umarov: gavin... a tcp connection is like a file
[11:54] Gavin.Hird @grid.xmir.org:8002: memory file
[11:54] Ubit Umarov: it gets a Fd on linux
[11:54] Ubit Umarov: and there is a limit on those
[11:55] LaNani Sundara: *slowly bangs head on the bar top
[11:55] Gavin.Hird @grid.xmir.org:8002: 64k per process on my system (system max)
[11:55] Ubit Umarov: used to be like 1024 per user
[11:55] Ubit Umarov: and depends on OS and OS version
[11:55] Ubit Umarov: just it is a limited resource also
[11:56] LaNani Sundara: ok so .. at least make that configurable... and optional?
[11:56] Gavin.Hird @grid.xmir.org:8002: the only time I have had OpenSIm run out of FD was for the old ODE when it calculated physics
[11:56] Sheera.Khan @hypergrid.org:8002: ulimit -a --> open files (-n) 1024
[11:56] Gavin.Hird @grid.xmir.org:8002: during startup
[11:56] Ubit Umarov: thx sheera.Khan
[11:56] Ubit Umarov: on win is more confusing
[11:56] Gavin.Hird @grid.xmir.org:8002: ulimit -a is the current setting not the max
[11:57] Ubit Umarov: depends on version and edition
[11:57] Ubit Umarov: Home with lower :)
[11:57] LaNani Sundara: look when a region starts and gets all those assets on a big full region, it makes a lot more connections ... than some users
[11:57] Sheera.Khan @hypergrid.org:8002: @Gavin: I know, It was intended as an ack of Ubits statement
[11:57] Ubit Umarov: no lanani
[11:57] Ubit Umarov: things are a bit more complex
[11:58] LaNani Sundara: well they always are
[11:58] LaNani Sundara: cache and local
[11:58] Ubit Umarov: to analyse i mean
[11:58] LaNani Sundara: but still
[11:58] Ubit Umarov: and now we do have keepalives
[11:58] LaNani Sundara: explain why that limit is NOT in the mesh module or the inventory desc one
[11:58] Ubit Umarov: so getting assets may one use a few FDs
[11:58] LaNani Sundara: but only in the Texture module
[11:58] Ubit Umarov: not even closing them between requests
[11:59] Ubit Umarov: that is also what the mysql connector does
[11:59] LaNani Sundara: i know
[11:59] Ubit Umarov: but answering a tcp connection can't do that
[11:59] LaNani Sundara: *nods
[12:00] LaNani Sundara: so.. even that answer, come back later uses an fd
[12:00] Ubit Umarov: well hopefully keepalives also help on that
[12:00] Ubit Umarov: but 32 connections just for textures or meshes is just 2 much
[12:00] Arielle.Popstar @grid.hgluv.com:8302: how about 16 then
[12:00] Ubit Umarov: how about 4 ?
[12:01] Ubit Umarov: lol
[12:01] Arielle.Popstar @grid.hgluv.com:8302: try it and see if any issue
[12:01] Arielle.Popstar @grid.hgluv.com:8302: 4 is 1/8
[12:01] Ubit Umarov: well that no issue on that
[12:01] LaNani Sundara: ok, i think we are getting nowhere, just think about it... i do not have those limits at all, not for a year
[12:01] Arielle.Popstar @grid.hgluv.com:8302: its not about how slow can we make it run
[12:01] Fu Barr doesn't understand why this isn't configurable with HERE BE DRAGONS warnings in the relevant .ini
[12:01] Gary Monsterous: Guys I have been quiet, go see how well LaNanis regions run, simple :) As a mesh player, i have seen how others regions are much slower with lower quality.
[12:02] Gary Monsterous: now i shush again :D
[12:02] LaNani Sundara: lol k ty
[12:02] Ubit Umarov: config is not the answer for all
[12:02] Gavin.Hird @grid.xmir.org:8002: On the simulators I set the session to allow 2048 open files
[12:02] Gavin.Hird @grid.xmir.org:8002: never run out of file descriptors
[12:02] Ubit Umarov: in fact we already have 2 much config options
[12:03] Bill Blight: What I would suggest LaNani, is put up a proper git , with proper git patches, would make it a lot easier for people to test your changes, and maybe see for themselves
[12:03] Arielle.Popstar @grid.hgluv.com:8302: but hard coding minimums is an answer?
[12:03] Ubit Umarov: yes arielle.Popstar
[12:03] Arielle.Popstar @grid.hgluv.com:8302: ari or ariele is fine
[12:03] Ubit Umarov: some things should be hardcoded
[12:03] LaNani Sundara: Bill my code is freely available and in use by plenty
[12:03] Bill Blight: but not "patchable"
[12:04] Bill Blight: all or nothing
[12:04] Ubit Umarov: i do remember old ode config options
[12:04] Arielle.Popstar @grid.hgluv.com:8302: well idea is to test the upper limit first to see where it is stable
[12:04] Ubit Umarov: well never mind :)
12:05] Bill Blight: test upper limit on what
[12:05] LaNani Sundara: Bill don't try put me in my place or something with an argument about how i make my code available
[12:05] Arielle.Popstar @grid.hgluv.com:8302: what will breakdown if set to 32? Server? Router, viewer?
[12:05] Bill Blight: the slowest pc that someone runs a region from home on?
[12:05] Bill Blight: the fastest one you can find?
[12:05] Arielle.Popstar @grid.hgluv.com:8302: on requests
[12:06] Arielle.Popstar @grid.hgluv.com:8302: thats whats being discussed
[12:06] LaNani Sundara: it works fine without that artificial request limit
[12:06] Ubit Umarov: no comments...
[12:06] Gavin.Hird @grid.xmir.org:8002: Intel announced a 28 core Xeon yesterday
[12:06] Bill Blight: Have you tested on a home hosted, adsl 128k connection with a cheap ass router?
[12:06] Bill Blight: just saying
[12:06] Fu Barr: yes, that's something many home users will be running...
[12:06] Ubit Umarov: if the damm OS does set limits, why will i waste my time discussion this ?
[12:06] LaNani Sundara: why would i
[12:07] Ubit Umarov: really guys..
[12:07] Arielle.Popstar @grid.hgluv.com:8302: no need as 4 requests is managable even on it
[12:07] Gavin.Hird @grid.xmir.org:8002: What is the limit on your linux Ubit?
[12:07] Arielle.Popstar @grid.hgluv.com:8302: i just like to know what will breakdown Ubit
[12:07] Ubit Umarov: sheera.Khan shown the default
[12:07] LaNani Sundara: these days most ISP give home users a dynamic IP and block ports, the number of home run sims will only decline
[12:07] Ubit Umarov: win home editions may have even lower limits
[12:08] Bill Blight: I just checked 2 debian boxes and 3 dif flavors of ubuntu and they were all 1024 by default
[12:08] LaNani Sundara: how far down do you want to stay compatible with old hardware and for how long?
[12:08] Ubit Umarov: and a damm FD can stay in use for minutes after close..
[12:08] Ubit Umarov: well... no point discussing this here
[12:09] Gavin.Hird @grid.xmir.org:8002: yes, but ulimit -n can set it higher
[12:09] Fu Barr: lanani - old hardware _is_ important
[12:09] Arielle.Popstar @grid.hgluv.com:8302: it would if you validate your reasonings
[12:09] LaNani Sundara: you want us to limit ourselves to slow adsl connections and old pcs
[12:09] Fu Barr: and home users _are_ important
[12:09] Bill Blight: well running a 24 core xeon and the limit is still 1024 on the os side, hardware has nothing to do with it , if the users (and I mean home users) don't know how to change that
[12:09] Fu Barr: you cannot leave them behind
[12:09] LaNani Sundara: you tell their ISP then to allow them to have an IP and no blocked ports
[12:10] Fu Barr: my ISP has no issues
[12:10] Bill Blight: Home users are a thorn in my side , but I understand that as a project Opensimulator was made to be used by them
[12:10] Fu Barr: free ports etc.
[12:10] Fu Barr: stable IP for weeks and weeks
[12:10] Fu Barr: months even
[12:10] LaNani Sundara: and even a home user can have good hardware and a decent connection
[12:10] LaNani Sundara: i am throwing nobody aside
[12:10] Arielle.Popstar @grid.hgluv.com:8302: if you worried about home users so much then why is no longer support for sqlite on grid mode?
[12:10] Ubit Umarov: the use of N paralell connects used to work to increase the available bandwith
[12:11] Ubit Umarov: bc a flaw on routers control
[12:11] Bill Blight: well that is simple, you should not run a GRID from home , but connected regions would suffer as well as standalones
[12:11] Ubit Umarov: long gone, and the FCC regulation changes did help
[12:11] LaNani Sundara: make it optional
[12:11] Ubit Umarov: more conns will not give more bandwith
[12:11] LaNani Sundara: thats all i say
[12:11] Gavin.Hird @grid.xmir.org:8002: make the queue lenght configurable
[12:11] LaNani Sundara: yes
[12:12] Ubit Umarov: and will not give more on opensim regions
[12:12] Gavin.Hird @grid.xmir.org:8002: even a single core should be able to handle a queuelenght of 4 without any problem
[12:12] LaNani Sundara: hear hear Gavin
[12:12] Ubit Umarov: who spoke about queue lenghts ?
[12:12] LaNani Sundara: i did
[12:12] LaNani Sundara: cos thats what it is
[12:12] Ubit Umarov: ahh not me :p
[12:13] Sheera.Khan @hypergrid.org:8002: it already is configurable - just edit it in the source code^^
[12:13] LaNani Sundara: if (m_queue.Count >= 4)
{
// Never allow more than 4 fetches to wait
reqinfo.send503 = true;
}
[12:13] Sheera.Khan @hypergrid.org:8002: so it could well be an option in an .ini file as well
[12:13] Ubit Umarov: no it will not be
[12:13] Andrew Hellershanks: Not that useful a comment as it doesn't say why.
[12:14] LaNani Sundara: ok well thank you Ubit for vetoing that :) all hail the decider
[12:14] Ubit Umarov: somethings must be decided at design stage by devs...
[12:14] Arielle.Popstar @grid.hgluv.com:8302: just have faith
[12:14] Arielle.Popstar @grid.hgluv.com:8302: one dev
[12:14] Fu Barr: ubit already said that he thinks 1. it's not a thing to change, 2. there are too many options as it is in his view.
[12:14] Ubit Umarov: not more configs to just scare users and invite them to break all
[12:14] LaNani Sundara: scare users
[12:14] Arielle.Popstar @grid.hgluv.com:8302: eyerolls
[12:14] LaNani Sundara: hide it in OpenSimDefaults.ini
[12:15] Arielle.Popstar @grid.hgluv.com:8302: you say that after all the file changes last release?
[12:15] Gavin.Hird @grid.xmir.org:8002: Frankly most casual users will not even look at OpenSimDefaults.ini
[12:15] Gavin.Hird @grid.xmir.org:8002: set it to 4 there and let those with more powerful HW increase it
[12:15] LaNani Sundara: thats why Gavin
[12:15] Sheera.Khan @hypergrid.org:8002: @ Ubit: I have to agree that is one of the weaker arguments ^^
[12:15] Sheera.Khan @hypergrid.org:8002: *giggles*
[12:16] LaNani Sundara: its a "because i say so" argument
[12:16] Ubit Umarov: whatever :)
[12:16] LaNani Sundara: you are not my dad, Dad!
[12:16] LaNani Sundara: :p
[12:16] Arielle.Popstar @grid.hgluv.com:8302: i am dev, hear me roar :)
[12:17] Ubit Umarov: you taking this issue this way tells a lot more about you than me :p
[12:17] LaNani Sundara: ok guess i will stick to my own version and people will keep getting it from my site and use it, even if its not in a git
[12:17] Arielle.Popstar @grid.hgluv.com:8302: no, it is the normal way for Ubit
[12:17] Ubit Umarov: if you consider this a "daddy" issue.. well your problem :p
[12:17] Arielle.Popstar @grid.hgluv.com:8302: have faith that what he says is the right way
[12:18] LaNani Sundara: *hugs Ubit great improvements in this current version. well done. Thanks
[12:18] LaNani Sundara: i do see progress :)
[12:18] Arielle.Popstar @grid.hgluv.com:8302: yes Ubit thank you for this
[12:18] LaNani Sundara: .•*♫ ♬ APPlAUSE! ♬♫ *•.
[12:18] Arielle.Popstar @grid.hgluv.com:8302: much better
[12:18] Arielle.Popstar @grid.hgluv.com:8302: claps
[12:18] Ubit Umarov: and on this i read some contradition on you :p
[12:18] Arielle.Popstar @grid.hgluv.com:8302: who?
[12:19] Ubit Umarov: well whatever
[12:19] Arielle.Popstar @grid.hgluv.com:8302: you allowed so why not the rest of us
[12:19] LaNani Sundara: was meant as said, i am glad you changed this throttle
[12:19] Fu Barr really needs to dive into to the code to be more clueful about all of this.
[12:19] Ubit Umarov: code is not static we do try to improve it
[12:19] LaNani Sundara: yes
[12:20] LaNani Sundara: and debate
[12:20] LaNani Sundara: :)
[12:20] Arielle.Popstar @grid.hgluv.com:8302: lets not be as conservative as past admins
[12:20] Ubit Umarov: and there are no "daddies " on this
[12:20] Arielle.Popstar @grid.hgluv.com:8302: you are daddy
[12:20] Arielle.Popstar @grid.hgluv.com:8302: what you say, goes
[12:20] Ubit Umarov: yes what core says goes..
[12:21] Arielle.Popstar @grid.hgluv.com:8302: what you say as you are the only active dev
[12:21] Arielle.Popstar @grid.hgluv.com:8302: core dev
[12:22] Sheera.Khan @hypergrid.org:8002: Great discussion today, I really enjoyed it :-)
[12:22] Arielle.Popstar @grid.hgluv.com:8302: but hugs anyway for making things a lot faster this week
[12:22] Sheera.Khan @hypergrid.org:8002: Even though we didn't reach a universally accepted solution there's a lot of food for thoughts and experiments :-)
[12:22] Fu Barr: tc ubit - thanks for all the hard work!
[12:23] Ubit Umarov: my pleasure
[12:23] Sheera.Khan @hypergrid.org:8002: Yes, I thank you too for your work - all of you ;-)

Personal tools
General
About This Wiki