BTRFS Performance

[raydoo]

Did anybody try the BTRFS Filesystem, and has any infos on the Performance ?

[maxbraketorque]

All available information indicates that it will be slower than EXT4. ASUSTOR says this, and comparisons of EXT4 and BTRFS on other systems show BTRFS to be slower. Are you trying to get an idea of how much slower BTRFS will be?

[father.mande]

Hi,

The difference is very limited and depend of your configuration
... ex. in single disk ext4 on benchmark (so not real) is faster than btrfs, in Raid mode (due to a better performance in write ... brtfs is at the same level or better than ext4

So it's very difficult to have a winner, especially if you forgot benchmark to real life ...

BUT brtfs add lot of interesting possibilities compare to ext4
... ex. snapshot ... if you want same on ext4 you need LVM and the performance slow down when it's easy and integrated in btrfs
... so the security is extremely increase compare to ext4 (snapshot, replication, recovery )
... defragmentation, cleaning tools at raid level or file system are better than for ext4

ALL this points take a real advantage to brtfs ... and move you to the future (safely)

At this time for a real usage ... you can't see any difference ... but win in security

Philippe.

[maxbraketorque]

Here are some comparisons that I could find in RAID mode:

https://www.phoronix.com/scan.php?page= ... _fs4&num=1

https://www.phoronix.com/scan.php?page= ... raid&num=1

https://www.phoronix.com/scan.php?page= ... -zfs&num=1

BTRFS is for the most part slower, in some cases quite significantly. It does best in the older comparisons. With that said, these comparisons are all with SSDs. Perhaps it does better with spinning disks?

I like the idea of better data safety, but I'm not sure if it matters in the end. My AS6104T killed my RAID10 volume during the update 3.2.7 to 3.3.0. Don't think BTRFS would have saved my volume. And I also wonder what BTRFS gets me if my NAS is used for backup, its configured as RAID10, and I'm backing up my NAS volume to a USB volume.

[maxbraketorque]

And some more:

https://www.phoronix.com/scan.php?page= ... tems&num=1

https://www.phoronix.com/scan.php?page= ... 2018&num=1

[maxbraketorque]

Since the upgrade to ADM 3.3.0 killed my volume, I've decided to take the opportunity to compare read/write speeds between EXT4 and BTRFS with my 4 drive IronWolf RAID10 setup. EXT4 tests are done. Now waiting for BTRFS volume to finish synchronizing, and then I'll run the same read/write tests. Should have results tomorrow evening.

[Norbyte]

From these tests the btrfs for home use doesn't bring advantages.
When backup (times) really matter, then could the built-in snapshot support give some bonus.

And without migration option the btrfs support is a thanks for the nothing 4 me.

[maxbraketorque]

So here are the performance testing results. I ran two kinds of tests. First was my own homemade test where I copied files from my home computer (MacBookPro with SSD) to and from the NAS. This consisted of copying one large file and then copying a folder full of ~10,000 smaller files. The other test I ran was AmorphousDiskMark which is the Mac equivalent of CrystalDiskMark.

The BTRFS volume did much better than I expected. The only place where it clearly lagged was writing a large number of small files to the NAS. In this case it was ~20% slower than EXT4. Otherwise, BTRFS performance was close to identical to EXT4. There were a few tests where BTRFS performed obviously better, in particular reading many small files from the NAS was ~15% better. AmorphousDiskMark also showed ~10% better read speeds for sequential reads with a large queue depth.

So in terms of performance, it looks to me that BTRFS is pretty good, but having said that, the 20% hit in small file write speeds will likely reduce backup performance by a similar amount, at least for typical desktop backups that would include many small files. For backups of large files, there would be almost no speed hit.

The good performance on large file reads might suggest that streaming performance should be good, but I suppose actual streaming tests would be needed to verify this.

AS6104T ADM 330 file system speed tests_small.png (155.3 KiB) Viewed 2282 times

[Norbyte]

Max,

wasn't the read/write performance bottlenecked by your network?
The RW speeds looks to very close to the Gigabit network speed limit, which is around 110-120MB/s when nothing else use it (and the router is capable to handle).
It could be interesting when you can create just a compare test with Link aggregatednetwork counfiguration (6104T has 2*1Gb net adapter), but that requires router with LA support , and need another device with 10Gb network or also 2*1Gb adapters...)

[maxbraketorque]

Max,

wasn't the read/write performance bottlenecked by your network?
The RW speeds looks to very close to the Gigabit network speed limit, which is around 110-120MB/s when nothing else use it (and the router is capable to handle).
It could be interesting when you can create just a compare test with Link aggregatednetwork counfiguration (6104T has 2*1Gb net adapter), but that requires router with LA support , and need another device with 10Gb network or also 2*1Gb adapters...)

Sustained read and write speeds are indeed saturating the 1 gbps ethernet bandwidth for both EXT4 and BTRFS, but the use of spinning disks in my NAS is also a bottleneck that also prevents showing the unconstrained differences in EXT4 and BTRFS performance. The question comes down to whether to perform an essentially theoretical comparison or a real-world comparison. My test setup is a very realistic one because most people are not using link aggregation or >1 gbps ethernet and are not using SSDs in their NAS. In some regards my comparison is perhaps still not realistic enough because I suspect that many people connect to their NAS via wifi (at least for home users). Ideally I should have done the same set of tests via wifi. And I probably should have done streaming tests. But I had the draw the line somewhere. Perhaps I'll try again later with a spare AS6102T that I have, but for now, I believe that the differences shown in my comparisons are meaningful for many people.