Re: Oscillations in disk write compaction, poor interactive performance
> The problem is that at the low point in the cycle, the machine is > unusable. It is utterly unresponsive until the writes complete, which can > take a very long time (in the case of the ppc machine, several minutes!) > Anything that does disk I/O will block for a long time - having 'ls' take > two minutes is not a good thing. Can you chack how much cpu time do dbflush and kswapd get ? > 2.2 does not exhibit this behaviour. 2.2 is much worse IMO. Bye. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Oscillations in disk write compaction, poor interactive performance
The problem is that at the low point in the cycle, the machine is unusable. It is utterly unresponsive until the writes complete, which can take a very long time (in the case of the ppc machine, several minutes!) Anything that does disk I/O will block for a long time - having 'ls' take two minutes is not a good thing. Can you chack how much cpu time do dbflush and kswapd get ? 2.2 does not exhibit this behaviour. 2.2 is much worse IMO. Bye. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Oscillations in disk write compaction, poor interactive performance
On Mon, 16 Apr 2001, Pavel Machek wrote: > Hi! > > > It also seems that in the 2.4 kernels, we can get into a sort of > > oscillation mode, where we can have long periods of disk activity > > where nothing can get done - the low points, where only 2-3 writes > > per second can occur, so completely screw up the interactive > > performance that you simply have to take your hands off the > > keyboard and go get coffee until the disk writes complete. I know > > we get better performance overall this way, but it can be > > frustrating when this occurs in the middle of video capture. > > I see oscilation even in 2.2.X case > > Can you try running while true; do sync; sleep 1; done? It should help. > > If it helps, try playing with bdflush/kupdate or how is it called/ parameters. > > -- > Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, > details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html. > The problem isn't that it oscillates, at least the oscillations shouldn't cause any problems - though we probably shouldn't see large scale oscillations like this anyway. The problem is that at the low point in the cycle, the machine is unusable. It is utterly unresponsive until the writes complete, which can take a very long time (in the case of the ppc machine, several minutes!) Anything that does disk I/O will block for a long time - having 'ls' take two minutes is not a good thing. 2.2 does not exhibit this behaviour. On the plus side, it appears that several other people are reporting this problem in 2.4, so I don't think I'm totally out to lunch. -dennis T - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Oscillations in disk write compaction, poor interactive performance
On Mon, 16 Apr 2001, Pavel Machek wrote: Hi! It also seems that in the 2.4 kernels, we can get into a sort of oscillation mode, where we can have long periods of disk activity where nothing can get done - the low points, where only 2-3 writes per second can occur, so completely screw up the interactive performance that you simply have to take your hands off the keyboard and go get coffee until the disk writes complete. I know we get better performance overall this way, but it can be frustrating when this occurs in the middle of video capture. I see oscilation even in 2.2.X case Can you try running while true; do sync; sleep 1; done? It should help. If it helps, try playing with bdflush/kupdate or how is it called/ parameters. -- Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html. The problem isn't that it oscillates, at least the oscillations shouldn't cause any problems - though we probably shouldn't see large scale oscillations like this anyway. The problem is that at the low point in the cycle, the machine is unusable. It is utterly unresponsive until the writes complete, which can take a very long time (in the case of the ppc machine, several minutes!) Anything that does disk I/O will block for a long time - having 'ls' take two minutes is not a good thing. 2.2 does not exhibit this behaviour. On the plus side, it appears that several other people are reporting this problem in 2.4, so I don't think I'm totally out to lunch. -dennis T - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Oscillations in disk write compaction, poor interactive performance
Hi! > It also seems that in the 2.4 kernels, we can get into a sort of > oscillation mode, where we can have long periods of disk activity > where nothing can get done - the low points, where only 2-3 writes > per second can occur, so completely screw up the interactive > performance that you simply have to take your hands off the > keyboard and go get coffee until the disk writes complete. I know > we get better performance overall this way, but it can be > frustrating when this occurs in the middle of video capture. I see oscilation even in 2.2.X case Can you try running while true; do sync; sleep 1; done? It should help. If it helps, try playing with bdflush/kupdate or how is it called/ parameters. -- Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: Oscillations in disk write compaction, poor interactive performance
Hi! It also seems that in the 2.4 kernels, we can get into a sort of oscillation mode, where we can have long periods of disk activity where nothing can get done - the low points, where only 2-3 writes per second can occur, so completely screw up the interactive performance that you simply have to take your hands off the keyboard and go get coffee until the disk writes complete. I know we get better performance overall this way, but it can be frustrating when this occurs in the middle of video capture. I see oscilation even in 2.2.X case Can you try running while true; do sync; sleep 1; done? It should help. If it helps, try playing with bdflush/kupdate or how is it called/ parameters. -- Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Oscillations in disk write compaction, poor interactive performance
Subject: Oscillations in disk write compaction The following data sets are the output of a small program that reads a random 4k block from a large data file, makes a trivial alteration to the block, and writes the block back into the file (in place). In all three cases the file is larger than the physical memory of the machine. In the first two cases, the file is the exact same. It appears that in 2.4, we are much more aggressive about ordering and combining writes to disk - which is probably a good thing, as it helps increase disk throughput. But it is also a bad thing, as it seems that this ties up the disk and rest of the system for long periods of time. It also seems that in the 2.4 kernels, we can get into a sort of oscillation mode, where we can have long periods of disk activity where nothing can get done - the low points, where only 2-3 writes per second can occur, so completely screw up the interactive performance that you simply have to take your hands off the keyboard and go get coffee until the disk writes complete. I know we get better performance overall this way, but it can be frustrating when this occurs in the middle of video capture. More notes below. Anyone got any ideas? Or have I done something horribly stupid here? -dennis T 2.2.14 - 160 meg intel PII. Relatively slow ide drive, 6 MB/sec --- File size: 209715200 Blocks: 51200 57.10 writes/second (10 second average, 4096 byte blocks). 57.10 writes/second (10 second average, 4096 byte blocks). 56.70 writes/second (10 second average, 4096 byte blocks). 57.30 writes/second (10 second average, 4096 byte blocks). 60.40 writes/second (10 second average, 4096 byte blocks). 64.30 writes/second (10 second average, 4096 byte blocks). 65.70 writes/second (10 second average, 4096 byte blocks). 60.90 writes/second (10 second average, 4096 byte blocks). 44.60 writes/second (10 second average, 4096 byte blocks). 45.30 writes/second (10 second average, 4096 byte blocks). 56.40 writes/second (10 second average, 4096 byte blocks). 67.60 writes/second (10 second average, 4096 byte blocks). 69.40 writes/second (10 second average, 4096 byte blocks). 66.80 writes/second (10 second average, 4096 byte blocks). 70.80 writes/second (10 second average, 4096 byte blocks). 18.80 writes/second (10 second average, 4096 byte blocks). 74.90 writes/second (10 second average, 4096 byte blocks). 76.00 writes/second (10 second average, 4096 byte blocks). 75.70 writes/second (10 second average, 4096 byte blocks). 59.60 writes/second (10 second average, 4096 byte blocks). 42.70 writes/second (10 second average, 4096 byte blocks). 73.00 writes/second (10 second average, 4096 byte blocks). 50.60 writes/second (10 second average, 4096 byte blocks). 102.80 writes/second (10 second average, 4096 byte blocks). 64.10 writes/second (10 second average, 4096 byte blocks). 91.30 writes/second (10 second average, 4096 byte blocks). 29.70 writes/second (10 second average, 4096 byte blocks). 28.80 writes/second (10 second average, 4096 byte blocks). 95.60 writes/second (10 second average, 4096 byte blocks). 58.50 writes/second (10 second average, 4096 byte blocks). 131.90 writes/second (10 second average, 4096 byte blocks). 6.80 writes/second (10 second average, 4096 byte blocks). 118.00 writes/second (10 second average, 4096 byte blocks). 3.10 writes/second (10 second average, 4096 byte blocks). 73.70 writes/second (10 second average, 4096 byte blocks). 30.00 writes/second (10 second average, 4096 byte blocks). 87.80 writes/second (10 second average, 4096 byte blocks). 97.80 writes/second (10 second average, 4096 byte blocks). 54.60 writes/second (10 second average, 4096 byte blocks). 64.60 writes/second (10 second average, 4096 byte blocks). 6.40 writes/second (10 second average, 4096 byte blocks). 126.20 writes/second (10 second average, 4096 byte blocks). 23.50 writes/second (10 second average, 4096 byte blocks). 88.00 writes/second (10 second average, 4096 byte blocks). 85.00 writes/second (10 second average, 4096 byte blocks). 90.70 writes/second (10 second average, 4096 byte blocks). 12.90 writes/second (10 second average, 4096 byte blocks). 40.90 writes/second (10 second average, 4096 byte blocks). 2.4.3 - 160 meg intel PII. Same machine as above, dual boot --- File size: 209715200 Blocks: 51200 57.40 writes/second (10 second average, 4096 byte blocks). 69.20 writes/second (10 second average, 4096 byte blocks). 84.90 writes/second (10 second average, 4096 byte blocks). 58.70 writes/second (10 second average, 4096 byte blocks). 52.60 writes/second (10 second average, 4096 byte blocks). 36.60 writes/second (10 second average, 4096 byte blocks). 35.10 writes/second (10 second average, 4096 byte blocks). 65.80 writes/second (10 second average, 4096 byte blocks). 74.70 writes/second (10 second average, 4096 byte blocks). 88.90 writes/second
Oscillations in disk write compaction, poor interactive performance
Subject: Oscillations in disk write compaction The following data sets are the output of a small program that reads a random 4k block from a large data file, makes a trivial alteration to the block, and writes the block back into the file (in place). In all three cases the file is larger than the physical memory of the machine. In the first two cases, the file is the exact same. It appears that in 2.4, we are much more aggressive about ordering and combining writes to disk - which is probably a good thing, as it helps increase disk throughput. But it is also a bad thing, as it seems that this ties up the disk and rest of the system for long periods of time. It also seems that in the 2.4 kernels, we can get into a sort of oscillation mode, where we can have long periods of disk activity where nothing can get done - the low points, where only 2-3 writes per second can occur, so completely screw up the interactive performance that you simply have to take your hands off the keyboard and go get coffee until the disk writes complete. I know we get better performance overall this way, but it can be frustrating when this occurs in the middle of video capture. More notes below. Anyone got any ideas? Or have I done something horribly stupid here? -dennis T 2.2.14 - 160 meg intel PII. Relatively slow ide drive, 6 MB/sec --- File size: 209715200 Blocks: 51200 57.10 writes/second (10 second average, 4096 byte blocks). 57.10 writes/second (10 second average, 4096 byte blocks). 56.70 writes/second (10 second average, 4096 byte blocks). 57.30 writes/second (10 second average, 4096 byte blocks). 60.40 writes/second (10 second average, 4096 byte blocks). 64.30 writes/second (10 second average, 4096 byte blocks). 65.70 writes/second (10 second average, 4096 byte blocks). 60.90 writes/second (10 second average, 4096 byte blocks). 44.60 writes/second (10 second average, 4096 byte blocks). 45.30 writes/second (10 second average, 4096 byte blocks). 56.40 writes/second (10 second average, 4096 byte blocks). 67.60 writes/second (10 second average, 4096 byte blocks). 69.40 writes/second (10 second average, 4096 byte blocks). 66.80 writes/second (10 second average, 4096 byte blocks). 70.80 writes/second (10 second average, 4096 byte blocks). 18.80 writes/second (10 second average, 4096 byte blocks). 74.90 writes/second (10 second average, 4096 byte blocks). 76.00 writes/second (10 second average, 4096 byte blocks). 75.70 writes/second (10 second average, 4096 byte blocks). 59.60 writes/second (10 second average, 4096 byte blocks). 42.70 writes/second (10 second average, 4096 byte blocks). 73.00 writes/second (10 second average, 4096 byte blocks). 50.60 writes/second (10 second average, 4096 byte blocks). 102.80 writes/second (10 second average, 4096 byte blocks). 64.10 writes/second (10 second average, 4096 byte blocks). 91.30 writes/second (10 second average, 4096 byte blocks). 29.70 writes/second (10 second average, 4096 byte blocks). 28.80 writes/second (10 second average, 4096 byte blocks). 95.60 writes/second (10 second average, 4096 byte blocks). 58.50 writes/second (10 second average, 4096 byte blocks). 131.90 writes/second (10 second average, 4096 byte blocks). 6.80 writes/second (10 second average, 4096 byte blocks). 118.00 writes/second (10 second average, 4096 byte blocks). 3.10 writes/second (10 second average, 4096 byte blocks). 73.70 writes/second (10 second average, 4096 byte blocks). 30.00 writes/second (10 second average, 4096 byte blocks). 87.80 writes/second (10 second average, 4096 byte blocks). 97.80 writes/second (10 second average, 4096 byte blocks). 54.60 writes/second (10 second average, 4096 byte blocks). 64.60 writes/second (10 second average, 4096 byte blocks). 6.40 writes/second (10 second average, 4096 byte blocks). 126.20 writes/second (10 second average, 4096 byte blocks). 23.50 writes/second (10 second average, 4096 byte blocks). 88.00 writes/second (10 second average, 4096 byte blocks). 85.00 writes/second (10 second average, 4096 byte blocks). 90.70 writes/second (10 second average, 4096 byte blocks). 12.90 writes/second (10 second average, 4096 byte blocks). 40.90 writes/second (10 second average, 4096 byte blocks). 2.4.3 - 160 meg intel PII. Same machine as above, dual boot --- File size: 209715200 Blocks: 51200 57.40 writes/second (10 second average, 4096 byte blocks). 69.20 writes/second (10 second average, 4096 byte blocks). 84.90 writes/second (10 second average, 4096 byte blocks). 58.70 writes/second (10 second average, 4096 byte blocks). 52.60 writes/second (10 second average, 4096 byte blocks). 36.60 writes/second (10 second average, 4096 byte blocks). 35.10 writes/second (10 second average, 4096 byte blocks). 65.80 writes/second (10 second average, 4096 byte blocks). 74.70 writes/second (10 second average, 4096 byte blocks). 88.90 writes/second