Re: [coLinux-devel] a. Memory leak ?, b. number of /dev/cobd?

[Dan A. <da...@co...>]

On Wed, Apr 14, 2004 at 11:52:21AM +0200, Axel Jessner wrote:
> Hello everyone,
> 
> I am finding coLinux allready an extremely useful and productive way to help me use the best of both the Linux and
> Win-world. I am running it on a WIN2K Laptop (1 GB RAM and 2 GHz clock) where both are stable. 
> 
> Problem a.
> I made an overnight test using my own software to create an index of several 10k binary raw data files residing in roughly two
> hundred directories on an external 160 GB USB masstorage  drive. This meant finding, opening reading each file and writing some
> catalogue information to a text file.
> All in all this were  about 90 GB of data. The programme started to execute  quickly and efficiently but in the next morning it
> had ground to slow crawl. I found that everthing was swapped out and and swap-use continued to increase.

Did you rule out the possiblity that your userspace program has a leak,
and not the kernel? Unless you see lines such as '__alloc_pages: 0-order 
allocation failed', chances are there was no kernel memory leak. 
The symptoms you describe fit more to the case where a userspace 
program eats up too much memory.

> I then found that even without the programme running, with a constant number of idle processes, free memory is slowly eaten
> up at about 100kB per minute. I attached two successive "top-results" to this mail. Apparently the reservation for Buffer
> increases regularely and this is taken away from free mem. Comparing that to a another (RedHat two-cpus) system I find
> that bufferspace also increases there, but not to the detriment of free mem. Could there be a problem with memory garbage
> collection in colinux?  Or can that be remedied in another way?

On normal conditions pages of the buffer cache is freed at some point. 
I've ran a few tests on coLinux and it clearly shows that it gets freed.

> Problem b.
> One of the great and ingenious advantages of coLinux and its interface to WIN is the fact that one can mount nearly all devices
> that WIN can mount, including i.e. masstorage on the WIN-USB ports either as flash memory sticks or big external
> disks. Nice is the facility to mount even more than WIN, i.e. the reiser partions on which the alternat boot Linux resides.
> This is normally not accessible by WIN. Please include it in the next kernel release as compiled in, if possible.
> Only the UDF-Packet format for CD-RWs posing as external RW-drives doesn't work, but this isn't supported by ordinary
> Linux either. In colinux I encounter a limitation in the number of /dev/cobd slots that can be allocated. On bootup the
> console message say eight are supported, but there are only four /dev/cobd0-3. I could not find the location whre this limit
> is set, nor could I increase the number of cobd in any other way. Does that depend on the kernel or on the debian
> base image?  

You can create the device node numbers for the other 4 devices using
mknod (1). 

About the 8 limit, it is semi-hard-coded at the moment, will be taken
care of in the next release.

-- 
Dan Aloni
da...@co...