Trying Out & Benchmarking Bcachefs On Linux 6.7

[ipso]

I stopped reading Phoronix a while back because the benchmarks are pretty much useless. When I saw this popup in my feed I figured I would take a look to see if things improved.

Michael knocks it out of the park as usual by benchmarking a file system with debug options enabled. 🤦♂️

Benchmarks using only default options is like taking your fancy car to the track but leaving it in the default ECO mode rather than pressing the Track mode button. Or taking your 4x4 to the mountains but leaving it in the default 2x4 mode. Options exist for a reason, they should be used accordingly.

Linus Tech Tips recently got raked over the coals for poor and misleading benchmark results. Phoronix has never had good results and I would be blown away if Michael has the integrity to add a note to the beginning of the article saying debugging options were enabled, so the comparisons are useless.

[Quackdoc]

I stopped reading Phoronix a while back because the benchmarks are pretty much useless. When I saw this popup in my feed I figured I would take a look to see if things improved.

Michael knocks it out of the park as usual by benchmarking a file system with debug options enabled. 🤦♂️

Benchmarks using only default options is like taking your fancy car to the track but leaving it in the default ECO mode rather than pressing the Track mode button. Or taking your 4x4 to the mountains but leaving it in the default 2x4 mode. Options exist for a reason, they should be used accordingly.

Linus Tech Tips recently got raked over the coals for poor and misleading benchmark results. Phoronix has never had good results and I would be blown away if Michael has the integrity to add a note to the beginning of the article saying debugging options were enabled, so the comparisons are useless.

this is a stupid take, phoronix's job is to be a news source for what people can expect to do, default options are default for a reason, and that is what will likely wind up shipping to people. it's not the distro maintainers job to go through every new feature's flags to enable them and disable them. No, Micheal did exactly his job here. If bcachefs wants to have good preformance, it needs to ship good default options.

[ipso]

this is a stupid take, phoronix's job is to be a news source for what people can expect to do, default options are default for a reason, and that is what will likely wind up shipping to people. it's not the distro maintainers job to go through every new feature's flags to enable them and disable them. No, Micheal did exactly his job here. If bcachefs wants to have good preformance, it needs to ship good default options.

Default settings are the bare minimum so the application will run. Take a database for example, they all have memory limits that can be defined, and the defaults are set way low so the database can start and run on low end hardware.
Using them in production like that will give awful performance.

Even video games, they don't ship with a default resolution of 640x480 so they run fast. They also don't ship with 4K as default as they might run like crap. They ship with sensible defaults that should "work" and give Ok results.

When running benchmarks the whole point is to give a like comparison. If one file system enables checksums, compression and dedup, and another has those features but does not enable by default, what is the point in comparing the default settings?

Just like if a video game defaults to 4K and another to 1080, you aren't comparing apples to apples.

[Quackdoc]

Default settings are the bare minimum so the application will run. Take a database for example, they all have memory limits that can be defined, and the defaults are set way low so the database can start and run on low end hardware.
Using them in production like that will give awful performance.

Even video games, they don't ship with a default resolution of 640x480 so they run fast. They also don't ship with 4K as default as they might run like crap. They ship with sensible defaults that should "work" and give Ok results.

When running benchmarks the whole point is to give a like comparison. If one file system enables checksums, compression and dedup, and another has those features but does not enable by default, what is the point in comparing the default settings?

Just like if a video game defaults to 4K and another to 1080, you aren't comparing apples to apples.

the point of comparing default settings is that is what will ship to the users, so thats what the users can expect

[ipso]

the point of comparing default settings is that is what will ship to the users, so thats what the users can expect

Ya, you're probably right, I'm sure most Phoronix readers buy their hardware from Acer and just use "default" settings. That must be why Linux has just one distro, with one window manager, one web browser and one terminal application... "defaults" are good enough for everyone. There is no way anyone here changes settings in the first five minutes of installing anything.

In fact, I bet that is how Michael got BCacheFS running to begin with, he probably just installed Ubuntu 22.04 and went with "default" settings right out of the box, and it magically worked right? There is no way it required him to compile a new kernel and actually enable it in the kconfig, that would be madness!

[Quackdoc]

Ya, you're probably right, I'm sure most Phoronix readers buy their hardware from Acer and just use "default" settings. That must be why Linux has just one distro, with one window manager, one web browser and one terminal application... "defaults" are good enough for everyone. There is no way anyone here changes settings in the first five minutes of installing anything.

In fact, I bet that is how Michael got BCacheFS running to begin with, he probably just installed Ubuntu 22.04 and went with "default" settings right out of the box, and it magically worked right? There is no way it required him to compile a new kernel and actually enable it in the kconfig, that would be madness!

considering the fact that phoronix is one of, if not the most popular linux news source, I can gurantee you, a lot of users who read these articles will stick with the defaults the distro ships, I highly doubt even a lot of the forum users regulairly build their own kernels

[gadnet]

I don't understand why it's interesting to test the speed of different filesystems. With NVME SSD storage, what does it matter? They are all plenty fast. Shouldn't you pick a filesystem based on its features, ease of use for common use cases?

well if you think performance has no interest at all it just mean energy consumption and money is not an issue for you. For a lot of people energy cost, server cost and response time is important

best regards,
Ghislain.

[zilexa]

well if you think performance has no interest at all it just mean energy consumption and money is not an issue for you. For a lot of people energy cost, server cost and response time is important

best regards,
Ghislain.

Unless you are talking about mass scale storage such as in a datacentre, I have no clue what cost you are talking about?? The filesystems are free to use. And what energy consumption?
I have never seen any such benchmarks since it's not just about speed. For personal use the differences detected in the benchmarks in this article won't be noticed in real life. And for large scale servers, it is all about availability and redundancy with heat being the treshold in the DC. Not speed.