Hi Greg, On Wed, Jan 14, 2009 at 8:53 PM, Greg Freemyer <greg.freemyer@xxxxxxxxx> wrote: > HSM team, > > Reading this thread got me taking a basic look at ext4 for the first time. > > I get the impression from what I read that ext3 is basically feature > frozen for new features and that major improvements like a HSM would > need to go into ext4 if they are to be accepted into the kernel. > > Why are you working with ext2 / 3? Is it just a stepping stone to ext4? > There is no explicit reason to start with ext2/3 but stability was one of the most important reason. Yes, we are definitely looking forward to ext4 for sure. But ya, this is one of the milestone. > One specific ext4 only feature is support for SSD discard commands. > SSDs function optimally when they know which sectors have valid user > data in them. Therefore ext4 calls discard for data blocks that are > part of deleted files. ext2/3 does not have an ongoing effort to do > this and I don't know if anyone will add that functionality to it or > not. > Well, why would I even bother to read the data on a block when the bitmap will suggest me that this is not allocated to anyone anymore. When we delete a file, we mark that bit in the bitmap as unused. Not sure, what you are trying to mean by this ? > FYI: the linux vfat and btrfs filesystems also have SSD discard > functionality. I don't think the others have addressed this new > optimization yet. > Do you want us to look into it ? > Also ext4 has a online defrag capability. (I'm not positive if this > feature is in 2.6.28 or not. I did not look, but as of 2.6.28 ext4 is > "stable".) > Yes, we did look into the patch. It is still under work. I think once this feature is in and it stabilizes with ext4. we can surely look forward to it. Though, the work surely needs to be started much earlier. > Since defrag has to move user blocks around in a live system with > minimal impact on user space, I would think you could look at this > code in ext4 and get another idea of how to do that, and if ext4 is > your target, you might be able to leverage some of the defrag > infrastructure. ie. Maybe just replace the data block allocation > requests with your own. > Well, the first look gives me a similar feeling. Much of the lower level work is already done. Only the allocation policies or mechanisms would differ. Also, Greg, what about the locking, I feel even they do take a lock on the inode and that too for the complete re-org ? > If you try to plug-in to the existing defrag logic, you may even be > able to extend it in such a way that allows HSM to be a module. Being > a module would be very cool. It would allow HSM to be used on distro > kernels that don't want to have HSM in them officially. > Yes, preferably it is always better to be modular. But I also, see that there would be lot of other changes too, we deal with the underlying devices too. We require the complete mapping between the device layer to the file system. As, OHSM claims mapping of files to tiers, which are basically collection of disks). So, achieve that we have to interact at file system level, the device mapper and the block devices. I feel its easy to say that we can make it independent of file system, etc but many a times our design gets restricted by the existing code bases. Then what do you suggest under such circumstances, Should be restrict our design or exploit the beauty of open source and make my private kernel ? > In particular the OpenSUSE kernels allow modules to be added that are > not part of the official distribution. > Thanks a lot for the insights. > Greg > -- > Greg Freemyer > Litigation Triage Solutions Specialist > http://www.linkedin.com/in/gregfreemyer > First 99 Days Litigation White Paper - > http://www.norcrossgroup.com/forms/whitepapers/99%20Days%20whitepaper.pdf > > The Norcross Group > The Intersection of Evidence & Technology > http://www.norcrossgroup.com > -- Regards, Sandeep. "To learn is to change. Education is a process that changes the learner." -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
