Back to Home / #mythtv / 2002 / 10 / Prev Day | Next Day
#mythtv IRC Logs for 2002-10-02

12:01--> Youare now talking on #mythtv
12:41-!-Chutt [] has quit ["bbiab"]
14:18-!-Chutt [] has joined #mythtv
14:57-!-kleetus [] has quit []
15:02-!-Chutt [] has quit []
15:02-!-davehunn [] has quit []
15:03-!-Chutt [] has joined #mythtv
15:03-!-davehunn [] has joined #mythtv
15:29-!-davehunn [] has quit []
15:29-!-Chutt [] has quit []
15:32-!-davehunn [] has joined #mythtv
15:32-!-Chutt [] has joined #mythtv
15:47-!-Chutt [] has quit []
15:47-!-davehunn [] has quit []
15:54-!-Chutt [] has joined #mythtv
15:54-!-davehunn [] has joined #mythtv
15:54-!-davehunn [] has quit []
15:54-!-Chutt [] has quit []
15:55-!-Chutt [] has joined #mythtv
15:55-!-davehunn [] has joined #mythtv
18:49-!-You are now known as mdz-work-away
19:32-!-Chutt [] has quit ["dead"]
19:37-!-snuckles [] has joined #mythtv
19:37<snuckles>Hi all.
19:39<snuckles>I have a question, how many of you are using a remote contol with MythTV?
20:47<-- snuckles( has left #mythtv
21:14-!-snuckles [] has joined #mythtv
21:32-!-Chutt [] has joined #mythtv
21:56-!-vektor [] has joined #mythtv
21:56-!-snuckles [] has quit [Read error: 104 (Connection reset by peer)]
22:06<vektor>Hey, question.
22:06<vektor>Anyone know: can I open the video device multiple times?
22:06<vektor>Like, if the other time is to just change the channel?
22:06<vektor>Can I do that?
22:16<Chutt>nope, you can't
22:17<vektor>Ok, thanks.
22:22-!-mdz [] has joined #mythtv
22:25<mdz>I am having problems getting smooth playback at any more than 320x240, which is intolerably blurry
22:26<mdz>what should I trade off?
22:28<vektor>What do you mean by smooth playback?
22:28<vektor>And how powerful is your machine?
22:28<mdz>Athlon 1.4GHz
22:29<mdz>and smooth meaning few enough drops that I don't get skips in audio or obvious choppiness
22:29<mdz>there are a lot of knobs here, and I've been fiddling with most of them, and I'm not sure where best to spend my available processing power
22:31<mdz>it looks like I can just barely get reasonable quality with rtjpeg using the default quality settings (which look decent) only by disabling deinterlaced playback and audio compression
22:32<mdz>it would be nice if I could use the mpeg-2 audio encoder from ffmpeg with mythtv; it's so fast as to be almost free
22:33<Chutt>lame at quality level 9 or so should be faster
22:34<Chutt>i wasn't too impressed by the libavcodec mp3 encoder
22:49<mdz>the mp3 encoder in libavcodec just uses lame
22:49<mdz>but it has an mp2 encoder which is very fast
22:51<mdz>quality=9 skips a little at the beginning, but seems to smooth out
22:56<mdz>it's odd, I can't figure out where my bottleneck is
22:56<mdz>my CPU is not maxed out
22:58<vektor>Make sure you turn on DMA.
22:59<vektor>Or you will suffer brutally.
22:59<vektor>Chutt: I ripped some code off you, hope you don't mind.
22:59<Chutt>course not
23:01* mdztries to avoid admitting that he's capturing over NFS
23:02<Chutt>not enough space locally or something to that effect?
23:02<mdz>the machine is diskless
23:02<mdz>it works out nicely because it is very quiet
23:03<mdz>only 2 moving parts, CPU fan and PS fan
23:03<mdz>I've benchmarked it, and the throughput is more than enough for video. I think I may need to use a network card which multitasks better than this one, though
23:03<Chutt>i do a lot of smaller reads, too
23:03<Chutt>which kills throughput on most normal ethernet cards
23:04<mdz>how small?
23:04<mdz>like, less than an ethernet frame?
23:04<mdz>uh oh
23:04<mdz>they are localized, though, right? cache should win
23:04<mdz>unless you're jumping all over the file
23:05<Chutt>should be reading sequentially
23:05<Chutt>you're playing back something previously recorded, right?
23:06<mdz>that, and timeshifting
23:06<mdz>I've been timeshifting for my performance tests
23:06<mdz>to find the maximum where I can watch without worrying about interfering with scheduled recordings
23:06<vektor>the machine is diskless?
23:06<vektor>are you _sure_ the network can handle the video rate?
23:06<Chutt>i don't think it's going to work :p
23:06<vektor>what about latency?
23:07<vektor>are you using nfs??
23:07<mdz>vektor: yes
23:07<vektor>how did you test read speeds?
23:07<vektor>Chutt: Do you write async to disk?
23:07<mdz>vektor: by playing back high-bitrate mpeg-2 over the network
23:07<vektor>two streams at once?
23:07<Chutt>no, i sync writes once a second
23:07<vektor>what about writing?
23:07<vektor>oh man
23:08<mdz>vektor: that's what I'm working on now
23:08<vektor>those write syncs will kill him
23:08<Chutt>when the kernel dumps its buffers
23:08<Chutt>you drop frames
23:08<vektor>mdz: but you can happily play back two high bitrate streams at once?
23:08<mdz>vektor: haven't tried that
23:08<vektor>since if you're timeshifting you're both reading and writing, right?
23:08<vektor>so that's two streams
23:08<vektor>and that will cause alot of crap to happen
23:08<vektor>but the write syncs will really kill you.
23:09<Chutt>really can't do without it
23:09<mdz>yes, that is the trick
23:09<vektor>jeez, get patching :)
23:09<mdz>lower bitrates should naturally give me better results
23:10<vektor>Chutt: I stole some code from you btw...
23:10<Chutt>then you'll go 30 seconds or so, the kernel will say 'hey, look, i've got full disk buffers, i better clear some' and bad things happen =)
23:10<vektor>mdz: but you still have to deal with the inherent nfs latency.
23:10<mdz>320x240 is no problem, never skips at all in my tests
23:10<Chutt>vektor, yeah, i saw what you said
23:10<vektor>Chutt: oh did i say that already?
23:10<vektor>damn i'm stupid
23:10<Chutt>what're you stealing?
23:10<vektor>nothing big
23:11<vektor>just diz' freq code and how you use it
23:11<vektor>have you ever met diz?
23:11<vektor>he's k-razy.
23:11<vektor>hangs on #livid btw
23:11<vektor>hard fucking core too.
23:11<vektor>dude wrote his own composite decoder ;-)
23:12<vektor>(like, composite video stream, not alpha composite) ;-)
23:13<vektor>ok question. around here we sort of have this convention. when we mean alpha composite, we pronounce it 'COMPOS-IT', and when we mean composite video, we say 'compozite'.
23:13<vektor>actually it's hard to write how you pronounce something.
23:13<vektor>i'm just curious if you pronounce those two differently.
23:13<vektor>or if it's just some whack odd canadian thing or something.
23:13<mdz>vektor: emphasis on the first syllable vs. the second?
23:13<Chutt>i pronounce em both the same
23:13<vektor>mdz: i guess so
23:14<vektor>komposite vs comPOS-it.
23:14<vektor>yeah, i guess it's 1st vs 2nd.
23:17* vektorgone
23:17<mdz>looks like utilization on the network is around 8-9mbit/sec in each direction
23:18<mdz>I should probably invest in a switch
23:18<Chutt>aren't the newer seagate drives pretty silent?
23:19<Chutt>maybe like a small drive, used for cache, basically?
23:19<mdz>yeah, that would be an idea
23:19<mdz>I have some small drives lying around, and could experiment to see which is quietest
23:19<Chutt>could stick the ringbuffer on there for timeshifting
23:19<mdz>I like the idea of pooling the majority of the storage, though
23:19<mdz>rather than having to divide it up between machines
23:19<Chutt>and implement a cache mechanism
23:19<Chutt>where it copies stuff over during/while playing
23:19<Chutt>so it reads from the local drive
23:20<mdz>but with 512M RAM, I should be able to buffer quite a bit of compressed video
23:20<Chutt>but the network copy goes all fast because it's using big reads/writes
23:20<mdz>big reads/writes can only help, for disks too
23:21<Chutt>i've been thinking about going that route, actually
23:21<Chutt>smaller machine in the living room for playback
23:21<Chutt>loud beast for recording elsewhere
23:21<mdz>in fact, the best possible setup would be to do the encoding on the big machine
23:21<mdz>I have so much processor to spare on the server it isn't funny
23:22<mdz>while the set top box is struggling
23:22<mdz>I was thinking about a very fast lossless encoder to shoot it over the network, then do a slow high-quality lossy encode on the server
23:22<mdz>but that would just barely fit one stream over the network
23:22<Chutt>my main problem is that the athlon xp i have is too hot for the nice black case i bought for it, especially when i stick it in the stereo rack
23:23<mdz>a rough calculation said that with a 2:1 ratio or so, it'd be pushing 80mbit easy
23:23<Chutt>you'd have to have a dedicated line for that, really
23:23<mdz>Chutt: I have a 1.4GHz t-bird for the living room, and an XP 2000+ in the office
23:24<mdz>which, like any other desktop, sits and does nothing for 99% of the time
23:24<Chutt>maybe i'll start playing more with that
23:24<mdz>while 10 feet away the other box is gasping for air
23:24<Chutt>wonder how hard it'd be to separate the encode/decode processes for timeshifing purposes
23:25<mdz>the only real issue I've run into with mythtv in this NFS configuration is that it can take a VERY long time to update the preview bitmaps
23:25<Chutt>there's quite a bit of syncing going on between them
23:25<mdz>it seems like it practically reads the entire file
23:25<Chutt>on the playback/delete page?
23:25<Chutt>60 seconds worth of it
23:25<Chutt>i've been meaning to make that optional
23:25<Chutt>since it takes awhile even on a local disk
23:26<mdz>how does it decide which frame to take?
23:26<Chutt>and it's not really needed with the little preview window
23:26<mdz>couldn't it just seek in some random amount and grab one?
23:26<Chutt>it goes 60 seconds in
23:26<mdz>it shouldn't need to read much just to seek, though, no?
23:26<Chutt>there's no seek table
23:26<mdz>I don't know much about nuppelvideo
23:26<Chutt>so finding a frame boundary would be kinda difficult
23:27<Chutt>i suppose, though
23:27<Chutt>it'd be doable
23:27<mdz>it looks like I've settled on a pretty usable configuration...480x480, mpeg4, 2200/scale=1, q=1-15, mp3quality=9, deinterlace=1
23:28<mdz>seems pretty smooth, with about 20% idle cpu
23:28<Chutt>not bad
23:29<Chutt>should be fairly watchable
23:37<Chutt>are you using cvs?
23:38<Chutt>if you change the first instance of 'linearblend' in libs/libNuppelVideo/tv.cpp to 'postprocess'
23:38<Chutt>it'll use mplayer's dering/deblock filters as well
23:38<Chutt>little additional cpu, but
23:38<mdz>I'm not using cvs at the moment, but I've been meaning to pull it
23:46<mdz>does it have any new dependencies?
23:47<mdz>I believe I sent you mail because I couldn't find the location of the CVS repository on the website...then I caught it in a message to the list or something
23:48<Chutt>i don't think i've added anything
23:48<Chutt>think the only dependency change is flac 1.0.4 for mythmusic
23:50<mdz>damn, now I can't find the CVS repository there some reason why it's a well-kept secret? :-)
23:51<Chutt>i don't want random people grabbing that instead of the main distribution, then emailing me if it doesn't work
23:51<Chutt>just want to deter those who want to run a cvs checkout just because it's newer
23:51<mdz>well, I promise not to whine if it doesn't work. I might send bug reports and patches, though.
23:52<Chutt>that's acceptible =)
23:52<Chutt>password is mythtv
23:52<Chutt>there's a bit of an upgrade process
23:52<Chutt>but i think i've got it all documented
23:53<mdz>which module is the mythtv stuff in? MC?
23:54<Chutt>i need to change that sometime
23:54<mdz>the other nice thing about the nfsroot is that I can just chroot into it on the server to do builds and installs, much faster than doing it on the set top and doesn't interfere with capture
23:56<mdz>and dpkg over NFS is hellishly slow
23:56<Chutt>i can imagine