Hi

I have an Athlon 2200+ with 256 DDR400 RAM and 256M swap and running Mandrake10 for the past 6 months or so. I have a 24/7 dialup connection and use it. This means I am up and running for weeks at a time, online with dc++ software running(called Valknut). I use it for other things of course, web surfing, word processing, small games like solitaire etc, JUK for my music is always running.

I find that over time, say 3 days, the system starts to become radically slow and chunky. Swapping between desktops takes up to 30 secs, sometimes longer, if I have music playing, it skips and jumps while this is happening. Just swapping between different apps on the same desktop can also cause this. Unless I have just used that app, then it is faster, because it seems to be in memory rather than the swap file.

I have KDE System Guard running with memory usage monitored and it shows most of the time, very little memory free and the swap file full. Even if I shut down all the user program that are running, so only system stuff is up, but still connected to the net with kppp, the memory that gets freed up is very minimal, I only recover about 50M, then as I restart the programs again, this grabs it all again and I am back to square one

I have read a lot of threads saying linux uses all memory to the best of its ability, and there is no need of a Memory flusher, but how do I stop it becoming slow and chunky? If I shut it down and restart it is fast so I know I have enough memory, even if I have all the programs open thatI usually use, so what is it with the time period that causes me strife? When I was using microshit98se I used a proggy called MemTurbo, which flushed the Ram when it reached a preset level and grabbed it all back from what was not using it, this is the type of program I was hoping to find in linux.

Any help with this problem would be appreciated

Cheers
Asif

btw, I know someone will say get more memory, but I cannot afford it at this time and I do not think it is the problem anyway, but, maybe a xmas present

Aright.
Sometimes i found my linux box getting slower over time with MDK10 (but it took me 3 months). And then i decided to upgrade to 10.1. I have to say it never happened with MDK 9.0. I think it was because of my root partition was an ext3, now i'm trying reiserfs. So here i start my suggestions:
-Use ReiserFS in your system partitions, say root, /usr and /tmp. To store /home use the file system of your choice.
-When installing, check how much space are you giving to swap and /tmp partitions. If using a web/ftp server also check the same /var. My box is a Celeron 2.3Ghz, 128Kb Cache L2, 256 DDR 400Mhz and 40Gb in a 7200RPM IDE drive. I assigned 256Mb to swap, 2Gb to root, 10Gb to /usr and 15Gb to /home. Because i don't use my box as a dedicated server, left /tmp and /var in de root partition. Beside, putting all together in one root partition is a bad idea, even more when need to reinstall system after a severe problem or if you want to upgrade again.
-Upgrade to 10.1, cause i found 10 buggy. 9 was far more stable... i've just installed 10.1 and looks much better. Beside, this version comes with mores recent version of the built in software.

Ok. I hope this tips help. They 3 are kind of extreme but that's all i can think about.
see ya!
triaydev

yaa , even i feel that mandrake 9.0 was more stable than 10.0
i have no explanation in its favour, but the latter gets stuck in hours and does not even shutdown on command. the only difference is that it is a bit more attractive in look and feel and ofcourse the 2.6 kernel. I personally believe that the 2.6 kernel has a lot of bugs. the same is happening with suse counterparts for me. i will soon change to the old mandrake 9.0 . We should find out some tweaking tools for linux, to avoid unnecessary memory usage , for graphical tweaking and for removing unnecessary files coming from the net.

do suggest on this, and help us make our linux a better experience with mandrake 10.0 too.

It gets "slow" because it's not flushing the cache out, which in turn causes other applications to further swap things out.

Try this...

The next time it's slow, close all open applications and open a root console.

At the prompt type in

swapoff -a ; swapon -a

And wait for it to complete.

It can take a few minutes to do so depending upon heavily things have fragmented in Vmemory.

Don't use it while it's doing this!

As soon as it returns, you'll find that your system is nice and snappy again.

Hi again, thanks for the replies

triaydev - I am not at this stage contemplating a re install of the system, but I will keep your suggestions in mind when I do. I do not have version 10.1 yet so cannot do an upgrade. If it is the only way to fix it, then I may have to go that way.

opjose - I tried your suggestion, but I do not think anything happened. I did not notice any activity out of the ordinary, so how do I know if it worked? I looked at my processes and useages and found I had made a mistake in the swap file size, it is actually 500M. It said 300M used, 200M free, before and after. So, I tried that command in two parts

swapoff -a then I waited a bit then
swapon -a

But still nothing had changed?? Did I miss something? Also I think it is the RAM cache that needs flushing not the hdd swap file. Because when it is chunky I can see the hdd light on fulltime and the cpu is running at 100% transferring everything from the swap to the RAM. Or am I mistaken here about what should happen?

Cheers
Asif

After issuing the swapoff, swapon (Mr. Myogi... heh...) check the free ram.

This forces a memory defragmentation.

Usually if your system was running slow prior to this, then you issue the commands, things speed up again.

However note that with KDE even 256megs of ram is kind of light, as KDE forces a lot of things to be held in RAM.

It's a big memory hog.

You may find that running one of the lightweight managers will improve the situation greatly.

9.x started reacting as you described and I saw it continuing fairly badly thru 9.2.

10.0 improves upon it, but 10.1 does a far better job and doesn't seem to be as prone to doing this.

Also simply exiting Xwindows is often enough to bring the unit back to normal speed.

Finally is DMA and 32 bit mode both turned on for you hard drives?

Did you edit the /etc/sysconfig/harddisks?

If you don't the default is to resort to rather slow HDD usuage which causes similiar problems.

Check it and see if you haven't done so.

I shall try that

Quote:

However note that with KDE even 256megs of ram is kind of light, as KDE forces a lot of things to be held in RAM. It's a big memory hog. You may find that running one of the lightweight managers will improve the situation greatly.

Shame, I like KDE, might have to get another stick, xmas!

Quote:

Also simply exiting Xwindows is often enough to bring the unit back to normal speed.

Yeah, true, but I am a cheapskate and hate having to reconnect to my ISP again, call charges get too hefty after a while and getting out of X will close the modem connection.

Quote:

Finally is DMA and 32 bit mode both turned on for you hard drives?

Yes I am sure it is for both, is this good or bad, you don't say?

Quote:

Did you edit the /etc/sysconfig/harddisks?

I think I did, but only to have the hdds run continuously, rather than shut down after a time. Nothing to do with what you are on about though.

If you don't the default is to resort to rather slow HDD usuage which causes similiar problems.

Check it and see if you haven't done so. [/QUOTE]

What should I have in here?

Cheers
Asif

"Yeah, true, but I am a cheapskate and hate having to reconnect to my ISP again, call charges get too hefty after a while and getting out of X will close the modem connection."

You must be using a KDE dependant dialer program.

You may want to investigate utilizing any of the dial on demand programs which is not gui dependant, but rather automatically brings up the link as required.

These kick in before X windows and therefor are not affected by much except purposely bringing down the network service.

"--------------------------------------------------------------------------------
Finally is DMA and 32 bit mode both turned on for you hard drives?
--------------------------------------------------------------------------------

Yes I am sure it is for both, is this good or bad, you don't say?"

You want both on if possible.

All you need do is uncomment the respective lines as the defaults are pretty safe.

"I think I did, but only to have the hdds run continuously, rather than shut down after a time. Nothing to do with what you are on about though."

I don't know about that.

If the HDD powers down then the flush service doen't have a chance to operate and things will grind to a halt until the drive powers up and the unit has time to release allocated ram and effect any VM swaps.

Is the drive spinning back up as required all by itself?

Hi again

I think I saw somewhere a while back where I can have my system connect on bootup, is that what you mean? I guess I could try that as I very rarely boot up without going straight back online I shall have another look around.

A slight misunderstanding here, my hdds do NOT shut down, I have them spinning constantly so they will last longer All that spinning up/down wears them out a lot faster than having them spin constantly.

So what I have is two harddisk files, harddiskhda and harddiskhdb and both are exactly the same as below: Now, I think I have to remove the hashes in front of a few lines here, (# USE_DMA=1) remove hash and which of the other ones please?

# These options are used to tune the hard drives -
# read the hdparm man page for more information

# Set this to 1 to enable DMA. This might cause some
# data corruption on certain chipset / hard drive
# combinations. This is used with the "-d" option

# USE_DMA=1

# Multiple sector I/O. a feature of most modern IDE hard drives,
# permitting the transfer of multiple sectors per I/O interrupt,
# rather than the usual one sector per interrupt. When this feature
# is enabled, it typically reduces operating system overhead for disk
# I/O by 30-50%. On many systems, it also provides increased data
# throughput of anywhere from 5% to 50%. Some drives, however (most
# notably the WD Caviar series), seem to run slower with multiple mode
# enabled. Under rare circumstances, such failures can result in
# massive filesystem corruption. USE WITH CAUTION AND BACKUP.
# This is the sector count for multiple sector I/O - the "-m" option
#
# MULTIPLE_IO=16

# (E)IDE 32-bit I/O support (to interface card)
#
# EIDE_32BIT=3

# Enable drive read-lookahead
#
# LOOKAHEAD=1

# Add extra parameters here if wanted
# On reasonably new hardware, you may want to try -X66, -X67 or -X68
# Other flags you might want to experiment with are -u1, -a and -m
# See the hdparm manpage (man hdparm) for details and more options.
#
EXTRA_PARAMS=-S0

nb. The last line is the one that keeps the hdd spinning.

Cheers
Asif

Re: Bootup

No, there are dial on demand programs that intercept attempts to connect to the internet and cause the PPP link to come up. This is different than bringing up the link on boot, which you do not want as it will slow startup greatly and always cause the modem to dial.

Re: HDD configs

# USE_DMA=1
# MULTIPLE_IO=16
# EIDE_32BIT=3
# LOOKAHEAD=1

ALL of the above should be uncommented!!!!

USE_DMA=1
MULTIPLE_IO=16
EIDE_32BIT=3
LOOKAHEAD=1

Then reboot.

You'll enjoy a huge increase in responsiveness in Linux!

Hi
On the dial up issue, I will have a search around

On the hdds, thanks, I was unsure

Cheers
Asif