Re: File systems [was Re: OpenBSD's extremely poor network/disk performance?]
Hi Karel, Thanks, for the correction... I thought zfs was bigger than that ;) Thanks On Wednesday, 8 January 2020, Karel Gardas wrote: > > > On 1/8/20 12:44 PM, Tom Smyth wrote: > >> As far as im aware there are 2 concerns about ZFS, >> 1) its license is not BSD /ISC you can use it and make money and not be >> sued, >> but it is more restrictive than BSD / ISC >> > > Yes, CDDL seems to be a no go based on past CDDL discussion which is > available for example in Star & OpenBSD thread on @tech: > > https://marc.info/?l=openbsd-tech=110806948606417=2 > >> 2) then there is the Number of Lines of code, which I believe is far >> longer than >> the OpenBSD code base, who and what team would manage the >> introduction of that code >> and the risks that come with that large a code base. >> > > Need to correct you a bit: > > ZFS: ~110k lines > XFS: ~95k lines > Ext4: ~38k lines > > while OpenBSD src/sys alone: > ~3.7mil lines where majority is in dev. But if I subtract drm code which > is probably the biggest contribution in dev (~1.7 mil lines), then I still > get roughly 2 mil lines of code in sys -- which is just part of base. > > LInes counted by sloccount. > > -- Kindest regards, Tom Smyth.
Re: File systems [was Re: OpenBSD's extremely poor network/disk performance?]
On 1/8/20 12:44 PM, Tom Smyth wrote: As far as im aware there are 2 concerns about ZFS, 1) its license is not BSD /ISC you can use it and make money and not be sued, but it is more restrictive than BSD / ISC Yes, CDDL seems to be a no go based on past CDDL discussion which is available for example in Star & OpenBSD thread on @tech: https://marc.info/?l=openbsd-tech=110806948606417=2 2) then there is the Number of Lines of code, which I believe is far longer than the OpenBSD code base, who and what team would manage the introduction of that code and the risks that come with that large a code base. Need to correct you a bit: ZFS: ~110k lines XFS: ~95k lines Ext4: ~38k lines while OpenBSD src/sys alone: ~3.7mil lines where majority is in dev. But if I subtract drm code which is probably the biggest contribution in dev (~1.7 mil lines), then I still get roughly 2 mil lines of code in sys -- which is just part of base. LInes counted by sloccount.
Re: File systems [was Re: OpenBSD's extremely poor network/disk performance?]
Howdy Stuart, On Wed, 8 Jan 2020 at 11:17, Stuart Longland wrote: > > On 8/1/20 1:25 am, Karel Gardas wrote: > > And yes, ffs performance sucks, but nor me nor you provide any diff to > > change that so we can just shut up and use what's available. > > Okay, question is if not ffs, then what? > > - Other BSDs have ZFS… is it viable to port that to OpenBSD? (Maybe > it's been done before? I didn't check.) As far as im aware there are 2 concerns about ZFS, 1) its license is not BSD /ISC you can use it and make money and not be sued, but it is more restrictive than BSD / ISC 2) then there is the Number of Lines of code, which I believe is far longer than the OpenBSD code base, who and what team would manage the introduction of that code and the risks that come with that large a code base. > - FreeBSD has UFS2, DragonFlyBSD has HAMMER… Could we borrow their code? > - If we could clean-room implement a BSD-licensed as a user I would say sweet... but someone more knowledgable / involved in the project would need to see a diff before a determination can be made. > EXT3/EXT4/BTRFS/XFS/JFS/whatever, following style(8), would there be > interest in supporting that in OpenBSD? what is the story with the license? if the license is not ISC / BSD I dont think it would be in base.. as a user I would say more filessytems sweet... but someone more knowledgeable / involved in the project would need to see a diff before a determination can be made. > - Or do we implement yet another file system? (Seems like too much work > for not much gain IMO.) as an OpenBSD user I would say that the performance of Network is dependent on your hardware. / the specific hardware Driver compatibility /capability in OpenBSD, I have had a different performance experience depending on the hardware I was using, and the maturity of the Driver support for that hardware. I have found the em(4) supported nics are pretty good ix(4) has solid performance vmx(4) have been good but it is dependent on the Vmware version you are using , and then others like vio(4) interfaces I have not had as good a performance. but that is more due to the age of the drivers and their capability vs what newer virtio drivers can do. But as a number of members of the OpenBSD Project have said to me Diffs are welcome ... Good Diffs will be considered just bear in mind the the License Requirements and Coding Style KNF when submitting a diff and do it off current... > > Regards, > -- > Stuart Longland (aka Redhatter, VK4MSL) > > I haven't lost my mind... > ...it's backed up on a tape somewhere. > -- Kindest regards, Tom Smyth.
File systems [was Re: OpenBSD's extremely poor network/disk performance?]
On 8/1/20 1:25 am, Karel Gardas wrote: > And yes, ffs performance sucks, but nor me nor you provide any diff to > change that so we can just shut up and use what's available. Okay, question is if not ffs, then what? - Other BSDs have ZFS… is it viable to port that to OpenBSD? (Maybe it's been done before? I didn't check.) - FreeBSD has UFS2, DragonFlyBSD has HAMMER… Could we borrow their code? - If we could clean-room implement a BSD-licensed EXT3/EXT4/BTRFS/XFS/JFS/whatever, following style(8), would there be interest in supporting that in OpenBSD? - Or do we implement yet another file system? (Seems like too much work for not much gain IMO.) There's merit in the third option, OpenBSD already supports EXT2 (which is also 90's vintage like ffs) as there are some platforms (e.g. loongson) that require it. I run BTRFS on a lot of my Linux machines, and aside from some features that are still experimental (quotas being one such issue), it seems to do the job. I've also been a big XFS user in the past. Performance seems good and XFS in particular has seen widespread production use, particularly in high-performance computing arenas. (SGI didn't exactly do things small!) EXT4 is also very widespread and stable, and seems to offer decent performance. ZFS and BTRFS are much newer, and more complicated with software RAID functionality built in. I think these would be harder to implement from scratch. DIY file systems doesn't seem like a good plan for success… it'll be a lot of work, won't be compatible with anything else, and could be as bad if not worse than what we have now, whilst also being untested. ffs is at least mature and stable! Are any of the "modern" file systems (from a design perspective, licensing is a different matter) suitable for use as OpenBSD's root fs? What would be needed? Regards, -- Stuart Longland (aka Redhatter, VK4MSL) I haven't lost my mind... ...it's backed up on a tape somewhere.