On Sat, Mar 19, 2011 at 4:19 PM, Per Jessen
todd rme wrote:
Todd, thanks your attempt at arguing why we should change the default filesystem. Can someone advocating btrfs as the default filesystem in the next openSUSE please provide an argument along these lines:
"We should make btrfs the default filesystem in openSUSE because the vast majority [= at least 80%] of openSUSE users will significantly benefit from the following new features or qualities:
[please list at least two].
You just said that you didn't want to discuss the advantages of btrfs.
I want to know only about the advantages that are _important_ to at least a _majority_ of openSUSE users. By changing the default we are presumably hoping to make life easier for a big chunk of our target users. (end-users or admins who would immediately change to btrs instead <the current default>).
But why does it have to matter to a majority of users? If it helps just 1% of users, but doesn't hurt anybody, why wouldn't we change? Ignoring the amount of work that it requires to change it, a decision on that would have to be decided on by the people responsible for implementing the change.
It also seems pretty arbitrary to say that every feature of btrfs has to benefit at least 80% of the people, and is an impossible condition to satisfy. We would never have switched to ext4 if we used this criteria.
I know, and we shouldn't have either. It was a purely arbitrary decision. Of course 80% is an arbitrary number, but we are talking about the default. Which (equally arbitrary) number are you expecting to satisfy by changing the default?
More than would be bothered by the change. It doesn't matter if it is a small number, the question is whether overall it would have a net benefit or a net detriment for users. If it has a net benefit, even a small one, then it is still a good idea to change.
A fair assessment would ask
1. Are there any drawbacks of using btrfs compared to ext4 that will effect a significant number of users 2. Are there any advantages of using btrfs compared to ext4 that will effect a significant number of users 3. Do the advantages outweigh the drawbacks for the majority of users? 4. Are there any major flaws in btrfs that should be deal-killers
If the answer to 3 is yes and the answer to 4 is no, then I think we should use btrfs.
Those are very good criteria, I agree. Thanks for elaborating.
My answers (probably irrelevant though) to 1-3:
1. Probably not. 2. Probably not. 3. No. (probably).
And you reasoning for these is?
Currently, issues like fsck and grub I would say are deal-killers. They may not be by the time of the next openSUSE release.
As for features that could be useful to a significant number of people but are not present in ext4:
1. Per-directory compression 2. Per-directory encryption (not implemented yet, but planned and it may be done by the next release) 3. Data deduplication (not implemented yet, but planned and it may be done by the next release) 4. Incremental dumps 5. Combining multiple physical drives into a single directory tree without needing logical volumes
Probably others I am not aware of.
I think perhaps we don't agree on what constitutes "a significant number of users". My impression is that the vast majority of our users are using a laptop or a desktop machine, and I really don't see any of those reasons above as a major advantage to any of those (nor to any of their sysadmins).
Once again, why should it have to affect "the vast majority of users". If the "vast majority" of users see no change whatsoever, but some power users get a benefit, isn't that still a overall benefit? I chose my words very carefully. When I said "significant", I avoided the word "majority" on purpose. I don't think what matters is that the majority benefit. What I think matter is that on average users are better off than before, even by a little bit. If just 1% of users get a benefit, and 0.9% see an equally large disadvantage, then it is still better to make the switch because on average it is beneficial. As for my specific examples: Compressing directories is pretty common if you want to save space. Not on Linux because it isn't available, but it is featured pretty prominently on windows systems. openSUSE already makes it easy to encrypt partitions, but encrypting individual directories would make, for instance, encrypting your home directory much easier (you would no longer need to make a fixed-size encrypted filesystem). Encryption is especially popular with laptops since they can be stolen easily. Data deduplication, if automated by a system tool, would be a great way to save space. It is as much an issue for home users as servers, perhaps more since they don't make as much of an effort to properly organize their files. Incremental dumps makes backups much more efficient and effective. Linux is already suffering from a lack of good backup solutions. And overall it will provide many of the features of LLVM directly in the filesystem. -Todd -- To unsubscribe, e-mail: opensuse-factory+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-factory+help@opensuse.org