BTRFS vs. F2FS on Asus X205TA

Austin Haedicke bio photo By Austin Haedicke Comment

Over the months that I’ve owned the Asus X205TA netbook / EeePC I’ve been logging a good bit of data. You can view the raw data on my shared Google Drive sheet.Specifically for this post you’ll be interested in the “I/O” page.

I ran simple benchmarks for boot time and with hdparm and iostat for both BTRFS and F2FS root filesystems. I also ran the respecive tests across all three mainline I/O schedules (cfq, noop, and deadline). Finally, I checked if there was a differnece in how the filesystems responded to my minimal initrd procee.

Filesystems: BTRFS vs. F2FS:

Most of what I’ve seen on the Phoronix Benchmark Blogs has F2FS consistently edging BTRFS in most domains, though BTRFS is certainly a contender (as is the tried and true EXT4 - not covered here).

You can look at the specifics in the linked spreadsheet, but my results were fairly lopsided. BTRFS was victorious in only 3 of 21 comprisons.
Some of the measures were close, but this is congruent with the above in that (even if the edge is slight), F2FS is persistently out pacing BTRFS. I used pretty basic tools, but for more robust testing I’ll fall back on the linked Phoronix results.

I/O Schedulers: CFQ, Deadline, NOOP:

This is, by far the more interesting data. After my system showed clear favor for F2FS, I took a closer look at the data between I/O schedulers. Similarly, I didn’t chart the raw data, but the rankings (1st, 2nd, and 3rd) for each test.

Similarly, the graph doesn’t tell the whole story, but you can see form the data that some of the results were pretty close. Ultimately in these tests, CFQ seems hit or miss, Deadline was more steady, and NOOP snuck in for the win. Really, NOOP won the (individual) boot time race though it appeared to slow down when loadin the extra initram modules. Also, look those transactions per second!

Looks like it’s F2FS + NOOP for my I/O read and write needs… until the next kernel that is!

comments powered by Disqus