OS simulator/Emulator

[Stonecold]

Hello All,
I need a small suggestion from you all...

I am planning to do some research work in the file systems area. To
put it roughly I am planning to introduce some machine learning
techniques while prefetching the files in a file system. In this way
the machine constantly keeps learning and will try to intelligently
guess the next file that might be used (basically to reduce cache hit/
miss). The idea is still in a very nascent form and I still have to
put in a lot of work to get something substantial.

The thing I am looking for right now is an OS simulator where I can do
my experiments on, so that I can come with some kinda report with and
without my prefetching algorithms. Also will a OS emulator (like
bochs) be useful for my purpose?

Please do send across your thoughts...

Regards,
Stcld

[Lokesh]

Hi there!

Well, its very good thought for discussion. Actually coupe of methods
are already in place to reduce cache hits
on the file system usage. In my view, trying out some operations on
file system with respect to ext2/3 will be better idea.

Where to start?
Knowing the internals of the file system would be a better option and
module programming concept and a better kernel
debugging tool like KDB. (you can find a post here for configuting the
KDB http://groups.google.co.in/group/linuxgurus/browse_thread/thread/a60d04bf1510823c/3c8485abab058830?lnk=gst&q=KDB#3c8485abab058830
)


How to start?
You can do these experiments on any of Linux distributions such as,
Suse, RedHat and Fedora also.
Make sure that kernel should be small enough as you need to recompile
sometimes and to plugin some
file system drivers once make some changes to it.

You can use tools such as, VMWare GSX server, Bochs, Cygwin etc.,
Let us review how this Bochs will be useful for your work. I will
review this Emulator tool and you can also do that on Windows / Linux
box.

Keep learning!
Lokesh

[Sameer Samarth P]

Thank you very much Lokesh for your inputs.

 

However there is a slight change in our idea now...The change is that now we are trying to build a Unified filesystem which will make the intelligent decision (machine learning part) to put the bits (blocks) of files across various other filesystems (maybe Gmail FS, ssh, Picassa FS, etc...) depending on various factors such as access time, file size etc..and due to the time constraints we are plnning to have just one level cache.

 

As you can you see most of the work will be done inthe user spcae and nothing in the kernel space (yeah! we kinda chickened out :-D)..we are planning to use FUSE for the unified FS and maybe some benchmarking tools such as Iozone to get the results.

 

Do you have any comments/suggestions in this regard? It wud be great to hear back from you.

 

~Stcld.