Techno Musings

There has been a lot of press within the past week about Google’s selection of EXT4 as their new default file system.  While Google’s performance numbers have yet to be published, the basic line of logic was that due to EXT2′s ability (Google’s current file system) to be more easily upgraded to EXT4 and the that the performance of EXT4 was “close to XFS” they selected EXT4.  Further buttressing their investment is the hiring of Ted Tso who has had a heavy hand in the creation of EXT4.

If I interpret into this I would say that XFS actually performed better than EXT4, but as the Google engineer stated the decision was more about the upgrade process than anything else.  I do find it interesting that JFS was the big loser here not even receiving a mention in the final set of two: EXT4 and XFS.  Also as I have talked about before XFS appears to be going about a bit of a revitalization within the kernel community.  Further we took a look in the past at the visualization of various file systems in the Linux source tree in this post.  The important question that I want to bring up from that post comes from the similarity of XFS and BTRFS and of course the EXT2, EXT3, and EXT4 variants.

This sets the stage for the argument/prediction that I will bring up.  I think that for Google’s operations today this was a perfectly logical decision to make.  They had a must have requirement of online updgrade and EXT4 met their goal.  However, I predict that in the long term as BTRFS specifically evolves they will have to consider upgrading to that file system in the future and cope with an out-of-family file system upgrade.  My reasoning is that a lot of the mindshare for doing things like working with Solid State Storage Devices, advanced snapshots, etc. are all being innovated into this file system and not EXT4.  So if Google would like to take advantage of these innovations then they will have to make the switch.  So in that sense I believe that Google made a great short term decision, but missed out on an opportunity to figure out how to do an out-of-family online filesystem upgrade by choosing XFS instead of EXT4.  However, only time will tell and since they hired Tso they could perfect the existing conversion approach to make it drop dead simple for conversion to BTRFS from EXT3/4.

What do you think did Google make a mistake or did they do the right thing, or is it a smart short term decision with painful long term consequences?