large filesystems on s1

[ronnythunder]

while trying to execute my grand plan of using a spare dsr6000 as a media server (tivo uses a drive, b drive is 120gb and used only for media), i've run into an interesting problem. the 3.1x series of kernels used in the series 1 boxes are based on the 2.1.24 linux kernel, which is pretty ancient. as such, there's not a great deal of support for really large (read: >2gb) partitions for anything other than the proprietary mfs partitions.

i'm considering two possibilities at this point: (a) backporting the revision 1 ext2 changes (including sparse superblocks, which is almost required for big partitions (a 120gb partition has over 900 block groups, and it's *way* not practical to have 900 superblock copies), or (b) backporting the fat32 vfat changes. my goal is to use this fs with samba, so it's all going to look the same from the outside.

any comments or observations? my leaning is toward the vfat side, simply because i could probably do all of that with modules and not have to have a custom kernel. however, i wonder if there would be a performance hit with that method (of course, the cpu and memory bottlenecks come into play at some point!).

ronny

[ronnythunder]

a followup on the vfat side: i finally ended up building a custom kernel because i was having problems mounting filesystems that i created with mkdosfs. turns out there's a bug in mkdosfs, but even after fixing it, i can't get an fs built that's bigger than 2gb. so, i'm giving up on that and trying the ext2 route...

on the ext2 side: i tried making an ext2 fs on the 120gb disk, and it took 31 minutes (!) of wall clock time to create the fs. then, the first mount took 7 minutes of wall clock time! this is a little too crazy for me, so i'm persuing the revision 1 ext2 mods. using a binary chop on kernel revs 2.1.25 - 2.1.13x (whatever the end of the 2.1x series was), i discovered that the rev 1 changes were installed between revs 2.1.36 and 2.1.37. i'm going to try to backport those changes and i'll report back.

ronny

[ronnythunder]

well, i think this is the final word (at least from me). if you're considering trying something like this, well, in the words of alton brown, I'd reconsider reconsidering. the kernel is just too damned old in these boxes, unless you want to put some serious work into the port of the newer ext2 fs. i finally got it to mount, but it started getting corrupted soon after starting to put some big files on the disk.

so, i gave up and just put the drive in my pc, where i'll just expose it as a windows/samba share.

perhaps i should try a series2 next time; i just don't have a spare one of those! :)

ronny