2008-08-20 00:05 -!- MaZe(~MaZe@c-24-6-86-168.hsd1.ca.comcast.net) has joined #tux3 2008-08-20 01:17 hey flips 2008-08-20 01:18 oh everybody is asleep now :) 2008-08-20 01:18 ACTION is awake 2008-08-20 01:21 yeah, it was nice meeting you the other day, I hope you had fun with us 2008-08-20 01:23 yeah, it was fun 2008-08-20 01:24 tried to absorb some of your scheduler speak, i don't know much about it 2008-08-20 01:30 there's a lot of activity trying to cross lock the rq so that you can move tasks across to another processor 2008-08-20 01:31 depending on the run category, FIFO or OTHER, you can migrate it directly or use a migration thread to facilitate the move 2008-08-20 01:31 problem with -rt is that the FIFO detection code can be very aggressive about scanning other run queues which can be a bit unbounded 2008-08-20 01:32 that can hold the spinlock protecting for a long time and can cause other kinds of contention if a lot of migration operations are happening 2008-08-20 01:32 if the algorithm is polynomial time this might cause severe contention on those locks 2008-08-20 01:32 have you folks thought about how to do online disk checking yet ? 2008-08-20 01:35 online fsck? 2008-08-20 01:36 yes 2008-08-20 01:37 have you folks thought about extending the page cache to do more sophisticated things like explicit buffer tracking ? 2008-08-20 01:38 i haven't, i believe flips has mumbled about it 2008-08-20 01:39 yeah, I wonder what i know or not is patented already 2008-08-20 02:14 -!- pgquiles(~pgquiles@239.Red-83-41-113.dynamicIP.rima-tde.net) has joined #tux3 2008-08-20 07:01 -!- pgquiles(~pgquiles@239.Red-83-41-113.dynamicIP.rima-tde.net) has joined #tux3 2008-08-20 13:21 the new iterative btree dump is way prettier than the recursive one and shorter 2008-08-20 13:54 shapor, next_key worked perfectly on the first try 2008-08-20 14:51 "Having actually run ZFS in production, there are some serious drawbacks with the remaining features (copy-on-write fragmentation, problems in SAN environments, etc), that may leave one wishing they'd implemented the ZFS features in a more stackable way so you could easily discard inappropriate layers and features" -- znork 2008-08-20 15:27 "You've run ZFS in production, yet you can't see the improvement on Linux's model? You mean the fact that md is completely broken and LVM is unreliable and slow by comparison?" -- outZider 2008-08-20 15:27 "Sir, I wish I had points to mod this up!" -- doomicon 2008-08-20 15:28 "ZFS is really, really nice but it does have some warts and the biggest for many would be that arcane operating system that's dangling off its nutsack" -- Kent Recal 2008-08-20 15:31 lol 2008-08-20 15:31 "Maybe he wants a cluster file system, or one that does HSM. I know I do and ZFS as of today does neither. ZFS is designed for managing a bunch of direct attached hard disks in thumper or similar device. At anything else it is frankly a bit sucky." jabuzz 2008-08-20 16:13 -!- MaZe(~MaZe@216-239-45-4.google.com) has joined #tux3 2008-08-20 23:46 I need a clever 16 bit hex magic number for the volume tree 2008-08-20 23:47 existing magics are 0x1eaf for file index leaves and 0x90de for inode table leaves 2008-08-20 23:47 could be 32 bits, since this table is special and size doesn't matter 2008-08-20 23:50 for now the magic is 0x2008