On Fri, Oct 1, 2010 at 11:27 AM, JD <jd1008@xxxxxxxxx> wrote: > > > On 10/01/2010 11:00 AM, Kwan Lowe wrote: >> On Fri, Oct 1, 2010 at 2:43 AM, Suvayu Ali<fatkasuvayu+linux@xxxxxxxxx> wrote: >>> Hi everyone, >>> >>> I was wondering if there was any way I could convert an ext4 partition >>> into an xfs partition without copying around files onto a separate >>> partition and reformatting. As other noted shrinking is not an option also it should be noted that block 1 of any file system cannot move without special handling.... >>> I realised, lately I have been dealing with very large files quite >>> often. Ranging from few hundred megs to a gig or two. So I decided to >>> switch to xfs. Do you think its worth the effort? It would require some benchmarking. A lot of people expect a new file system to run faster than the underlying hardware ;-) XFS posted its best results on a system with a very large number of disk controllers i.e. big fast parallel hardware supporting a parallel RAID/LogicalVolume device. >> Curios about this though -- what sort of performance are you seeing >> with ext4 vs XFS? I'm running only a few XFS filesystems but I don't >> see a huge performance increase versus ext4. My workload is typically >> small sound files. It is important to benchmark your raw disk and compare the raw results with the filesystem results. Small sound files are closer to the file load that a mail server might have. There have been numerous discussions in the context of mail and net news.... >> >>> Also since we are talking about my /home here, there will also be small >>> conf/settings files along with the large files. Do you think that would >>> be disadvantageous somehow? Would it make more sense to shrink my /home >>> and have a separate xfs partition in the created space for the large files? I doubt that it would help more than single digit %age one way or another. >>> -- >>> Suvayu >>> >>> > Hi Kwan, > I had run performance measurement for reading and > writing of multimegabyte files in linux xfs and ext3. > There was absolutely no denying that ext3 outperformed > xfs by a long shot. Unfortunately, I do not recall the exact > numbers. Linux and ext3/4 are a good pair. > One thing that is overlooked by a lot of people is that > the linux xfs IS NOT what runs on the big SGI machines. > Many performance features were not ported because the > x86 architecture does not support, for example, huge pages > (64k page). the SGI 64 bit architecture supports these > large pages, and perhaps even larger. There are several > implementations of the MIPS architecture. > SGI was probably using a much more robust implementation. > Also, high performance xfs requires large amounts of ram > for the btrees that are searched for the exact blocks to access. > Most x86 machines, especially laptops, are poor targets > for such an architecture. So, many of the features were > removed or scaled back to a size that effectively killed > performance on pc's. As far as I know the xfs in Linux is the same XFS that Irix had. However Irix was designed to support lots of parallel hardware, I/O and ccNUMA processing. It is the Irix operating system that establishes a foundation that Linux is just beginning to address. Large pages are just one of the core operating system features that made Irix a world class Unix version (still is). > Even on Linux x86_64, with 16GB of ram, the xfs is > still a low performance FS as compared with ext3. > > Perhaps SGI will embark on a new port for the new intel > 64bit architectures (don't hold your breath :) Do not bet on it... the old SGI is gone. > These tests were done back in 2006. > > I am not sure if the linux xfs has undergone > any serious performance improvements since then. You are correct.. yet the design goals of xfs had a lot to do with durability combined with the speed that a journaled FS could have. One of the important design goals of XFS was how it survives the reset button/ power failure and how correctly and quickly fsck would finish on reboot. It should be noted that XFS supports attributes in a way that enables SELinux and Trix (security enhanced trusted Irix) to operate reliably. At the time these were big and landmark improvements in file system technology. Now that these are 'public' open source projects can build on them.... thus ext4 "should" be an improvement but your work load and hardware could tip the scales one way or another. -- NiftyFedora T o m M i t c h e l l -- users mailing list users@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe or change subscription options: https://admin.fedoraproject.org/mailman/listinfo/users Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
