Linus Torvalds Upset over Ext3 and Ext4

It all started with a request for help from Jesper Krogh in one of the first responses to Torvalds's announcement March 24 of Kernel 2.6.29 on the gmane.linux.kernel mailing list. Krogh reported a significant delay when writing from cache with the ext3 filesystem, despite faster hardware and extensive RAM. Was there a way to autotune it? Ingo Molnar opined that Krogh's wait time of 10 minutes was totally unacceptable, "it is the year 2009, not 1959." His personal "pain threshold" is about one second: "the historic limit for the hung tasks check was 10 seconds, then 60 seconds."

Ted Ts'o, groundbreaking in the filesystem's development, chimed in to the forum. It was just recently that he had been confronted by users over data loss upon installing their apps on the new ext4 filesystem. Ts'o set himself intensely on the problem with the source research and detailed explanation. Again he described the delayed effect in writing data. Synchronization in ext3 occurs every five seconds, whereas ext4 normally writes from cache every two minutes. Ts'o got pretty defensive: "People can call file system developers idiots if it makes them feel better --- sure, OK, we all suck. If someone wants to try to create a better file system, show us how to do better, or send us some patches."

Torvalds, for one, didn't seem too excited about the delayed synchronization. He writes on the mailing list, "Doesn't at least ext4 default to the insane model of 'data is less important than metadata, and it doesn't get journalled'? And ext3 with 'data=writeback' does the same, no? Both of which are -- as far as I can tell -- total brain damage. At least with ext3 it's not the default mode." To avoid the synchronization problem Ts'o had recommended at least temporarily migrating ext4 to a few separate systems only. Torvalds considered this to be "crappy" advice and that "we might as well go back to ext2 then."

In his response, Ts'o fell back on the performance benefits thanks to delayed allocation, as had been allowed earlier under POSIX. By his experience, the difference between five seconds and three minutes "wasn't that big of a deal" in practice, "at least in the days when people were proud of their Linux systems having 2-3 year uptimes." Plus there was a remedy: "For precious files, applications that use fsync() will be safe." If this were a problem for some, they could "turn off delayed allocation with the nodelalloc mount option."

Kernel chief Torvalds is hardly convinced by these arguments. In his view, "if you write your metadata earlier (say, every 5 sec) and the real data later (say, every 30 sec), you're actually more likely to see corrupt files than if you try to write them together... This is why I absolutely detest the idiotic ext3 writeback behavior. It literally does everything the wrong way around -- writing data later than the metadata that points to it. Whoever came up with that solution was a moron. No ifs, buts, or maybes about it."