On Sun, 1 Jul 2007, Carlos E. R. wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
The Sunday 2007-07-01 at 01:16 +0300, Tero Pesonen wrote:
...
How does it show? Well, first of all, there's constantly (every few seconds) from some 8 to few dozens blocks being written, according to iostat. Almost no reading at all. Then, every so often, maybe after every 2 or 3 minutes (I'm not sure yet if this varies), it writes a bigger burst of data, one that I can easily hear the hard drive doing, and it can least a few seconds, but less than 10 seconds I'd say.
Have a look at variables like ACPI_THROTTLED_KUPDATED_INTERVAL in /etc/sysconfig/powermanagement. I think there are more of the sort.
In my earlier reply to this list I considered this issue as mostly solved for what I think is "normal." I also fiddled with the powersave settings through Yast and asked it to set the "powersave" mode as the default. That probably had some effect. Also, I turned off each and every process I do not need. Both or either one of these made some real difference.
I didn't find any way to monitor which files are constantly being written to.
A pity.
So I tried to find via kfind which files have most recently been written to (access date changed to current.) There were tons of those, but at least it seems that in /sys and especially in /proc huge amount of files are updated so that they're always marked as having been updated at the same time as the current time shown by system clock. Not even a minute behind, never.
However, those two folders hold "virtual" files, so they can't be affecting.
OK.
All my hda partitoons were formatted for SuSE 9.3 when it was installed, and are ReiserFS. According to Yast partitioner, they seem to have been set to "ordered data mode" from the three possible (journal, ordered, writeback.) This is what was then set by default, as I don't remember to having changed those.
Add options "noatime,nodiratime" to your fstab file. Every time a file is read, the time when it is read is written to the filesystem, causing activity. I'm also thinking of /proc and /sys dirs: they files are virtual, but the directories themselves are physical, and therefore, their 'atimes" are written to the HD.
I had already added the option noatime, which actually might have helped a bit. I will still add nodirtime, although my system seems to run already quite well. At least the blocks read/written ratio is more even than it used to be and the tps number is much lower overall and what is more, it remains low during inactivity and there are no regular large "bursts" of data any more.
Why is this constant I/O a problem? Well, I have a feeling my old SuSE 9.3 did not do this. What is more, I'm worried if my poor IDE hard drive can take all this strain the system is putting on it. If my hard disk is writing 24h/day, how long will it last before it dies under such server-level use? The heads need to constantly move and write... and these standard IDE HDD's are definitely no server level units that are designed to perform under constant stress.
Well, the disk will certainly not go to sleep, but I have noticed this behaviour ever since I started using linux. Windows did not behave this way (3.11 at the time): I noticed my disk went to sleep because I had to wait for it to wake up a few seconds after 20 minutes of not using the computer. This doesn't happen with the disk holding "/" in linux, but it might happen with data disks.
I've also read on the web about "laptop modes" etc. that avoid excess and unnecessary disk I/O to allow spinn-downs etc. Could something like this be employed on my desktop PC?
Yes, probably.
I think the "Powersave" named power saving mode already does something like this. But not sure. Any way, things seem better now.
Would there be anything in /etc/sysconfig to edit--I already even turned off all weird sounding man pages related jobs put there in "cron".
I don't think cron will have much effect. But yes, there are settings in the file I mentioned, and maybe in /etc/sysconfig/powersave/disk. Probably better to use yast to adjust those - not quite, not all settings are accesible. You can even add a scheme, and it will be created in that dir, where you can further fine tune it.
I may still take a look at those, but if I'm happy how things are now I think I won't go fiddling with those certain disk settings that if done wrong, could have undesirable side effects, even data loss, as has been warned on the web.
And above all, do others' systems show similar disk I/O behaviour?
Yes. I haven't manage to discover "who" is the culprit, but yes.
Oh, almost forgot: I don't offer any network services etc. Not even SSH like I used to.
Don't forget log entries...
Yes, I also checked them, but none was flooded with data nor were there recurring warnings or errors or anything of particular interest. I would consider this mostly solved for as far as I dare to go without making anything too risky. I've used Linux since SuSE 8.2, so I should already know not to expect things to work as they do on Windows. Thanks, Tero Pesonen -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org