[NetBSD logo]    &    [Google logo]

NetBSD-SoC: Improved Writing to FileSystem Using Congestion Control 


What is it?

When the rate of generation of write requests by the kernel or userland processes exceeds the service time of the underlying FileSystem(FS)/disk, write congestion can happen. Such a phenomenon has many undesirable affects manifested in the form of :- (i) unequal sharing of disk bandwidth between  processes, (ii) delayed write (due to lots of request getting piled up in the buffer), and (iii) read requests getting affected due to onslaught of write requests. Hence,  the primary aim of this project is to provide mechanism(s)  that would enforce "fairness" and "rate control" among competiting proceses initiating such write requests.



Mandatory (must-have) components:

Optional (would-be-nice) components:

Technical Details


Caveat: Code released has not been exhaustively tested though reasonable care has been taken to ensure that that the implementation will not cause undesirable affects. In some instances, it has been observed, during heavy write events, that the CCA algorithm makes the system non-responsive to remote login via ssh and in two instances caused the ioflush daemon to pause for long intervals of time.  Also, in the current  release, the rate of read requests by processes are also getting affected while slowing down the rate of generation of write requests. All these are under investigation and will hopefully be addressed in future releases.

Instructions on how to run the code can be found here.  Initial results using the approach can be found here.



SourceForge.net Logo
Sumantra R. Kundu <sumantra@gmail.com>
$Id: index.html,v 2006/05/23 10:44:18 hubertf Exp $