Second

[ambarisha b]

Hi,
Second one on saturday at 8PM?
We'll still stick with those papers.If you guys find anything else do
post it.And people try reading those papers.Unless people already
study the papers, its gonna be a one way talk and one way talks tend
to get pretty boring when the topic has fair bit of technical
detail.Discussion of the work that had been done to write those papers
would be great.So, do give those a shot.

Regards
Ambarish

[Aastha Mehta]

Can you please attach the papers here for the benefit of off-campus people? Or atleast post pointers to their sites.

Thanks,

Aastha.

--
Aastha Mehta
Intern, NetApp, Bangalore
4th year undergraduate, BITS Pilani
E-mail: aast...@gmail.com

[ambarisha b]

Actually "those" meant the papers I posted for the First meet.We
didn't come to those during the talk.Anyway I am attaching them again.

Cheers
Ambarisha

[ambarisha b]

So............Everybody OK with saturday 8PM?

Also I will try to finish reading those papers.It would be great if
somebody else could .Would help the discussion. Tapan ?? Pranav ??

Cheers
Ambarisha

[Pranav Agarwal]

I had started Disk Allocation Policies.. didn't finish it yet... will try my best to get through it..

Pranav

[Tapan Avasthi]

I will make an attempt to put up my understanding of the research paper - File Layout and System Performance.
Yes, 8 PM is fine for me.

-- Tapan

PS: I have created a thread named - "Queries-Meet1"
Please don't keep any doubt with yourself. Put up your queries on group after every meet or otherwise.

[ambarisha b]

Great, I too am half way through File Layout and File System
Performance.I will try to finish the other as well.

Regards
Ambarish

[kunreddy ramakrishna reddy]

Good going .. so excited about tomorrows interaction even though i did not prepared any thing :) :)  

But next saturday i will discuss one FileSystem case study(BFS of BeOS) in depth.

--
Ramakrishna Reddy K, 

M.E. Computer Science,

Project Assistant,CAHU,BITS-Pilani,Rajasthan.

[Tapan Avasthi]

Can our meeting be postponed to sunday on a suitable time? I am sorry for my prior ambivalent confirmation of the time.
I need some time for preparation of the paper, else can we meet tomorrow as well?
Has room booking been done?

-- Tapan

[ambarisha b]

Fine.Tommorrow at 8-930 in 1232.Just confirmed with Ramakrishna and
Pranav.I believe its important that we wait till we finish reading
than strict timings.Otherwise we miss the whole point.Hope everybody
is OK with it.(Informal is what we are.)

@Tapan:Carry on with the paper then.I finished File layout and File
system performance.

Cheers
Ambarisha

[ambarisha b]

Pretty interesting meet today, people.Lets post the queries and follow
ups on the same thread , what say ?? It would be great if somebody
could confirm the buddy system issue.What do you guys say we study the
code along the way.We could have a couple of guys studying and
discussing the code during the meet while others continue the study of
research papers.I'll try to get the code for ffs and see if I can
figure anything out.If anybody wants to join in, would be great.

//Ambarish

[kunreddy ramakrishna reddy]

Yaa today's meet was really good.I am ready for both(code as well as papers).And next session also we will try to discuss FFS only.After understanding FFS thoroughly we will move to "Log Structured File System".

Important:( Second Yearites)--- Try to download Operating systems(Galvin) text book from D.C. and try to read File system basics so that it will be easy to understand the next sessions.

Question or Exercise:(Before coming to next session) (Off-campus people also invited to solve this and correct me if the question is wrong)

What is typical Hard Disk Bandwidth utilization of current Local File Systems.Any Techniques to improve bandwidth utilization (Refer research papers).
We will discuss first 5 minutes of next session on the above exercise.

[Aastha Mehta]

Don't know if this is asking too much, is it possible for someone to jot down minutes of each meeting as well? 

Seems like you had a great meet up (FFS = Flash/Flat/FAT/Fast file system???, buddy system of memory allocation,etc.) and have come with (food for thought)/s. One of them has been mentioned here already, thanks Ramakrishna, will think about it. If we could get the context in which the problem came up, we would be able to jump in properly too.

Thanks,

Aastha.

[Tapan Avasthi]

Yesterday, I was wrong about the advent of spintronics field of magnetic storage devices.
Read on GMR Technology to take a slight overview.
GMR Technology:http://goo.gl/AwPOO
Early Storage:http://goo.gl/ki41O

-- Tapan

[kunreddy ramakrishna reddy]

For complete information and source code of Different File systems the following link is useful

http://tldp.org/HOWTO/Filesystems-HOWTO.html#toc9.2

[Aastha Mehta]

well, you could combine information from this as well.

[ambarisha b]

People, little query.
Suppose I make a change in the os code.Is it necessary to compile the
entire thing again and deploy it or is there someway we could
dynamically load just that module into the kernel, say something like
modprobe.Any info on this would be great.

Regards

[Vineet Pandey]

I do not know about modprobe (or whether the following would work):

In ns2, there is a central make to build the entire program, and then there are makefiles in individual folder (wherever necessary) which you can use to change that particular part without having to recompile the entire code. you can look if any such scheme is available here.

[kunreddy ramakrishna reddy]

@Ambarisha  What Vineet said (NS2 Module Compilation)  seems to be applicable for OS Kernel Also.

 I think it is possible to compile only module instead of recompiling the whole OS.

http://www.cyberciti.biz/tips/compiling-linux-kernel-26.html

Just refer Step-4 in the above Tutorial.

[Aastha Mehta]

@Ambarisha: how about writing a dummy hello world module. and compiling it and trying to load it using modprobe?

[ambarisha b]

But how do I write a dummy hello world module ? I was thinking of
making some change to an existing module.But am afraid I might screw
my working system.

Regards
Ambarisha

[Aastha Mehta]

you could do this:
first write some print statement of your own in some existing module. then try to load and run that specific module and see if your statement got executed or not. if it does then you can go on to writing your own dummy module.

by dummy hellow world module i mean, create an interface (eg same type of signature of the function but a different name) similar to the interface of some existing module. write again a simple print function inside to print hello world. load your new dummy module like other modules and see if it works. if this happens, then you know how to add new kernel modules :)

Aastha.

[Arpan Gujarati]

Currently I am working on WinDbg for windows kernel debugging. I am able to see all the function calls, local variable values, call stack information as the OD boots up. For this I just need to hook up my computer with the target system (where we have the debugee OS) with a serial cable, configure the target system for debugging and must have the source code of debugee.  

 

You can use a similar tool to debug a Linux system (both kernel mode and user mode). Since Linux is open source, this shouldn't be a problem. If anyone can suggest such a tool, good. I'll also try to find out from my colleagues working on linux systems.

--
Arpan Gujarati
4th Year, B.E. (Hons.) Computer Science,
Birla Institute of Technology and Science, Pilani
Pilani - 333031
Rajasthan
(M): +91-9725045119

[ambarisha b]

@Arpan:Yeah, that was my view of things.But I was thinking of running
the debugee OS, as you call it, as a virtual machine because I don't
have access to 2 machines.I can't figure out a way to simulate a
serial connection between this vm and my host os.And offtopic, you
have got access to windows source ? That must be cool.

Anyway , I decided I rather wait till I get the thing working on a vm
than try it out on my present stable one.

//Ambarish

[Arpan Gujarati]

Don't get me wrong when I say debugging. There may not be an error always to perform debugging. Debugging is just an effective way to trace through the execution with the information like function calls, variable values available so that one can better understand the codebase.

 

On Fri, Feb 11, 2011 at 7:55 PM, Arpan Gujarati <arpan44...@gmail.com> wrote:

[Arpan Gujarati]

@ambarish: File systems majorly fall into user mode, so I suppose you may be able to perform debugging on the same OS on which you are working.

 

And I am working on Intel graphics drivers for Windows platform. So I have access to the required "interfaces" (only). Not that cool! :)

[ambarisha b]

So, can I step through a system call like , read() then ?
Of course, I wouldn't expect Microsoft to give the entire source to
each individual developer.But still you get a feel of the quality of
code that makes up windows , right ?

On Fri, Feb 11, 2011 at 8:17 PM, Arpan Gujarati

[Aastha Mehta]

@Arpan: Can you clarify what you mean by "File systems generally fall in user mode". As far as i understand system calls occur in kernel mode. File systems consist of system calls, so usually they belong to the kernel domain too. And it is very tricky to implement/add/modify system calls. I don't know whether tracing is possible in user mode or not. Can you throw some light on it if you know?

[ambarisha b]

Yeah , I too was under the same impression.

@Arpan : Do you mean stepping till the system call or into the call?

[Arpan Gujarati]

First, I said file systems 'majorly' fall into user mode. So they can always be in kernel mode too.

 

We are arguing if file system drivers fall in kernel domain or user. File system is a piece of code which continously interacts with the user actions. At the same time it also needs to access the low level system functions like disk access, IO, buffers. In other words there are interfaces which work at different levels - for example when I mentioned file system, I was referring to a set of file system filter drivers which happen to be in the user mode (in windows). Is the file system majorly in user mode or kernel mode depends on the user-kernel communication model (system calls in linux).

 

I'll try to post detailed info on the topic if I find something.

 

*Filter drivers are an added interface to existing drivers to improve/add features or to make the code more modular. They are optional.

 

@ambarisha - I mean stepping into the call.