Google Switching To EXT4 Filesystem(google.com)
google.com
Google Switching To EXT4 Filesystem
http://www.google.com/url?sa=t&source=web&ct=res&cd=2&ved=0CAkQFjAB&url=http%3A%2F%2Fdigitizor.com%2F2010%2F01%2F14%2Fgoogle-switching-to-ext4-filesystem%2F&ei=oWVQS52SLun_tgebj433DA&usg=AFQjCNFBOHmhTDtyTjz8P-u1NHqTCI9k_w
6 comments
You can hop directly to the original mailing list post here: http://lists.openwall.net/linux-ext4/2010/01/04/8 (the entire thread is a good read for anyone curious about file system benchmarking).
Thanks! Even more valuable tip since the link is a redirect to "digitizor.com" which seems to be currently down.
Why is that link pointing to google in first place??
Why is that link pointing to google in first place??
I wish they'd give us some idea of how they are testing beyond mentioning "customer workload"; maybe even release their own benchmark by simulating some typical customer workload. The dbench bashing seems out of place without some further explanation.
>PS for some reason "dbench" holds mythical power over many folks I have met. They just believe it's the most trusted and standard benchmark for file systems. In my experience it often acts as a random number generator. It has found some bugs in our code as it exercises the VFS layer very well.
I wouldn't really call this "bashing" dbench. He is simply saying that it is a good tool, especially with regard to exercising the VFS layer, but is not complete enough to do everything they could ask for (like emulating their customer workloads). I read it more as a warning to see what it actually does for you and not get caught up in the hype.
I wouldn't really call this "bashing" dbench. He is simply saying that it is a good tool, especially with regard to exercising the VFS layer, but is not complete enough to do everything they could ask for (like emulating their customer workloads). I read it more as a warning to see what it actually does for you and not get caught up in the hype.
Blogpspam. Why not link to the original source instead? http://lists.openwall.net/linux-ext4/2010/01/04/8
It's amazing how little of the blogspam page was the actual content, as opposed to various blinking ads and stuff that your eye naturally skips over.
"It was not until we moved from benchmarks to customer workload that we were able to make detailed performance comparisons and find bugs in our implementation."
So true. I don't know how many times I've been frustrated by people (ab)using benchmarks. The only real benchmark is your actual load. The closer you can approximate that, the better. And no, Dhrystone is nothing like your actual load.
So true. I don't know how many times I've been frustrated by people (ab)using benchmarks. The only real benchmark is your actual load. The closer you can approximate that, the better. And no, Dhrystone is nothing like your actual load.
And not just realistic load, but realistic data quantity and quality too.
Note that Chris Mason used to work on reiserfs, which means he has real-world experience working on actual filesystems. That he is involved in ext4 bodes well (note the CC'ed list on the link that micktwomey posted).
Bigger than that, Ted Ts'o (http://en.wikipedia.org/wiki/Ted_Tso), is a Googler. He probably has some experience with ext2, 3, and 4.
Well, Chris Mason is focused on btrfs, not so much ext4. See, e.g., http://www.insight24.com/webcasts/content-100695_1
From gmane linux-ext4 mailing list
http://article.gmane.org/gmane.comp.file-systems.xfs.general...