On Mon, Mar 12, 2007 at 11:43:16AM -0600, Mike McCarty wrote: > Roberto C. Sanchez wrote: > > > >I personally am a fan of XFS. However, it is also possible to use ext3 > >on large partitions, as you point out. At work, I have a production > >server (running RHEL, unfortunately) which is serving up a 6 TB > > Why unfortunately? Do Linux fans have to hate other distros as well > as MS? > Ever worked with RHEL or Fedora (or Red Hat before that)? They have their very own little RedHat-specific way of organizing /etc. Many of the things that they do go against pretty much every other distro (except for those which specifically try to emulate RedHat). I don't hate RHEL. I just hate some of the broken defaults. Additionally, they have *Enterprise* right in the name, but don't have out of the box support for any filesystem other than ext2/3 (except for maybe ReiserFS, but I would hardly call that enterprise quality). I currently have a server in production which (because of where it is located and the security policies of the organization/facility where it is located), must run RHEL3 or RHEL4. Before I rebuilt it using RHEL4, it was using RHEL3 to serve up three volumes from two external RAID trays via NFS. It had to be three different volume because under RHEL3, the biggest filesystem that could be supported out of the box was 2 TB (because of 2.4 kernel and some other userland utility limitations). I rebuilt that machine using RHEL4 so that the users would only need to access one volume. Since I knew that right off the bat it would a single volume of about 6 TB and that we would later want to add more storage, I went looking for the XFS or JFS packages on the install CDs. When I couldn't find them, I went into the #rhel channel and asked around in there if anyone knew why RHEL did not support XFS or JFS. The responses I got were along the line of, "ext3 is fine for everything." To which I replied, "what about for filesystems over 8 TB?" Of course, the answer to that was "build a cluster with GFS." Ordinarily, I would just get the sources to the kernel and the associated userland tools and build them myself. But, security at this place would simply not go for it. So, in short, they make the life of the admin exceptionally difficult if you want to do something which they (the RHEL designers/developers) did not think you would want to do. > >filesystem. I took to reading up on ext3 and judiciously set things > >like the block size and some of the other filesystem parameters so that > >crash recovery would not take ages and so that performance would be a > >bit better. Of course, since Debian supports both XFS and JFS quite > > Care to share your insights? Or at least pointers where one may > obtain similar insights? Those of us who use ext3 would appreciate > any distillation of the information. > There is a ton of information about JFS and XFS on the net. All you need to do is check the Wikipedia filesystem comparison page or Google search for filesystem comparisons. The short of it is: ext3 - good general purpose FS (not the best performance, but stable) xfs - excellent performance with huge files and huge filesystems jfs - similar to XFS but I think it has better performance when under heavy I/O load reasierfs - good with lots small files and when you don't really value your data (not that well understood) Regards, -Roberto -- Roberto C. Sanchez http://people.connexer.com/~roberto http://www.connexer.com
Attachment:
signature.asc
Description: Digital signature