|
Replies:
8
-
Last Post:
Jun 21, 2005 5:11 PM
by: sdebnath
|
|
|
Posts:
7
From:
Registered:
6/21/05
|
|
|
|
Posts:
6
From:
Seattle, WA
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 2:57 PM
in response to: geoffbgb
|
|
As a company I am not sure if there is interest to pursue UFS2. ZFS is
our next generation file system and a lot of hard work is being put into it.
However, from an UFS developer's point of view, fixing the
oh-so-many-flaws-in-design issues with the current UFS is definitely a
plus. This would include mucking with locking and other components as
well integrating enhancements from McKusick's paper. Perhaps we can
move towards a protected extended attribute namespace as well? Who is up
for this? :-D
Shawn
Geoff Buckingham wrote:
> Is there any interest in UFS2, I believe Sun may have attempted an implementation in the past, (Solaris 9 timeframe) so there may be code in Sun already.
>
> Anyone curious about UFS2 may find out more here:
>
> http://www.usenix.org/events/bsdcon03/tech/full_papers/mckusick/mckusick.pdf
> This message posted from opensolaris.org
> _______________________________________________
> ufs-discuss mailing list
> ufs-discuss at opensolaris dot org
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
254
From:
DE
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 3:31 PM
in response to: sdebnath
|
|
On Tue, 21 Jun 2005 23:57:28 +0200, Shawn Debnath <Shawn dot Debnath at Sun dot COM>
wrote:
> As a company I am not sure if there is interest to pursue UFS2. ZFS is
> our next generation file system and a lot of hard work is being put into
> it.
quite correct.
> However, from an UFS developer's point of view, fixing the
> oh-so-many-flaws-in-design issues with the current UFS is definitely a
> plus. This would include mucking with locking and other components as
> well integrating enhancements from McKusick's paper. Perhaps we can
> move towards a protected extended attribute namespace as well? Who is up
> for this? :-D
there's a minor quirk, BSD UFS2 depends on softupdates, our UFS
does have logging. I dont think a merger is usefull.
---
frankB
>
> Shawn
>
> Geoff Buckingham wrote:
>> Is there any interest in UFS2, I believe Sun may have attempted an
>> implementation in the past, (Solaris 9 timeframe) so there may be code
>> in Sun already.
>> Anyone curious about UFS2 may find out more here:
>>
>> http://www.usenix.org/events/bsdcon03/tech/full_papers/mckusick/mckusick.pdf
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
6
From:
Seattle, WA
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 3:47 PM
in response to: batschul
|
|
So there are 3 filesystems that are losely based on the same principles:
ext2/3 (very different, but ondisk idealogy remains same), solaris ufs,
and bsd ufs. I think a merger would be quite useful, and would lead
towards an unified file system for the three worlds. Making customers
happy. Yes our version of UFS does have a lot of the features, better
shadow inode storage, extended attributes, and logging, However, none of
them are being fully utilized because of imperfections (si and ext attr,
are not utilized, and logging has design issues). If this project can
aim to solve the general problems and not write a brand new file system,
I see this to be of an immense value to the community. Of course this
isn't easy, but thats the challenge.
Shawn
Frank Batschulat wrote:
> On Tue, 21 Jun 2005 23:57:28 +0200, Shawn Debnath
> <Shawn dot Debnath at Sun dot COM> wrote:
>
>> As a company I am not sure if there is interest to pursue UFS2. ZFS
>> is our next generation file system and a lot of hard work is being
>> put into it.
>
>
> quite correct.
>
>> However, from an UFS developer's point of view, fixing the
>> oh-so-many-flaws-in-design issues with the current UFS is definitely
>> a plus. This would include mucking with locking and other components
>> as well integrating enhancements from McKusick's paper. Perhaps we
>> can move towards a protected extended attribute namespace as well?
>> Who is up for this? :-D
>
>
> there's a minor quirk, BSD UFS2 depends on softupdates, our UFS
> does have logging. I dont think a merger is usefull.
>
> ---
> frankB
>
>>
>> Shawn
>>
>> Geoff Buckingham wrote:
>>
>>> Is there any interest in UFS2, I believe Sun may have attempted an
>>> implementation in the past, (Solaris 9 timeframe) so there may be
>>> code in Sun already.
>>> Anyone curious about UFS2 may find out more here:
>>>
>>> http://www.usenix.org/events/bsdcon03/tech/full_papers/mckusick/mckusick.pdf
>>>
>
>
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
254
From:
DE
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 3:47 PM
in response to: sdebnath
|
|
On Wed, 22 Jun 2005 00:47:03 +0200, Shawn Debnath <Shawn dot Debnath at Sun dot COM>
wrote:
> So there are 3 filesystems that are losely based on the same principles:
> ext2/3 (very different, but ondisk idealogy remains same), solaris ufs,
> and bsd ufs. I think a merger would be quite useful, and would lead
> towards an unified file system for the three worlds. Making customers
> happy. Yes our version of UFS does have a lot of the features, better
> shadow inode storage, extended attributes, and logging, However, none of
> them are being fully utilized because of imperfections (si and ext attr,
> are not utilized, and logging has design issues). If this project can
> aim to solve the general problems and not write a brand new file system,
> I see this to be of an immense value to the community. Of course this
> isn't easy, but thats the challenge.
I agree an the general usefull principle, however I think
a simple port from UFS2 is going to be easier and more
efficient then attempting a merge of the 2 UFS versions.
---
frankB
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
6
From:
Seattle, WA
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 4:16 PM
in response to: batschul
|
|
Frank Batschulat wrote:
> On Wed, 22 Jun 2005 00:47:03 +0200, Shawn Debnath
> <Shawn dot Debnath at Sun dot COM> wrote:
>
>> So there are 3 filesystems that are losely based on the same
>> principles: ext2/3 (very different, but ondisk idealogy remains
>> same), solaris ufs, and bsd ufs. I think a merger would be quite
>> useful, and would lead towards an unified file system for the three
>> worlds. Making customers happy. Yes our version of UFS does have a
>> lot of the features, better shadow inode storage, extended
>> attributes, and logging, However, none of them are being fully
>> utilized because of imperfections (si and ext attr, are not utilized,
>> and logging has design issues). If this project can aim to solve the
>> general problems and not write a brand new file system, I see this to
>> be of an immense value to the community. Of course this isn't easy,
>> but thats the challenge.
>
>
> I agree an the general usefull principle, however I think
> a simple port from UFS2 is going to be easier and more
> efficient then attempting a merge of the 2 UFS versions.
port of bsd's ufs2 to solaris? and shouldn't you be sleeping now? :)
> ---
> frankB
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
817
From:
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 4:56 PM
in response to: sdebnath
|
|
Shawn Debnath wrote:
>
>
> Frank Batschulat wrote:
>
>> On Wed, 22 Jun 2005 00:47:03 +0200, Shawn Debnath
>> <Shawn dot Debnath at Sun dot COM> wrote:
>>
>>> So there are 3 filesystems that are losely based on the same
>>> principles: ext2/3 (very different, but ondisk idealogy remains
>>> same), solaris ufs, and bsd ufs. I think a merger would be quite
>>> useful, and would lead towards an unified file system for the three
>>> worlds. Making customers happy. Yes our version of UFS does have a
>>> lot of the features, better shadow inode storage, extended
>>> attributes, and logging, However, none of them are being fully
>>> utilized because of imperfections (si and ext attr, are not
>>> utilized, and logging has design issues). If this project can aim
>>> to solve the general problems and not write a brand new file
>>> system, I see this to be of an immense value to the community. Of
>>> course this isn't easy, but thats the challenge.
>>
>>
>>
>> I agree an the general usefull principle, however I think
>> a simple port from UFS2 is going to be easier and more
>> efficient then attempting a merge of the 2 UFS versions.
>
>
> port of bsd's ufs2 to solaris? and shouldn't you be sleeping now? :)
yes he should be. But, I agree with Frank on this, a port of UFS2 would
be much simpler. Merging the UFS versions together will likely create
instability.
sarah
*****
>
>
>> ---
>> frankB
>
> _______________________________________________
> ufs-discuss mailing list
> ufs-discuss at opensolaris dot org
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
6
From:
Seattle, WA
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 4:34 PM
in response to: batschul
|
|
An FAQ for those interested...
http://sixshooter.v6.thrupoint.net/jeroen/faq.html
also we could make block allocation a hell of a lot smarter, avoiding
fallocate hacks :)
Shawn
Frank Batschulat wrote:
> On Wed, 22 Jun 2005 00:47:03 +0200, Shawn Debnath
> <Shawn dot Debnath at Sun dot COM> wrote:
>
>> So there are 3 filesystems that are losely based on the same
>> principles: ext2/3 (very different, but ondisk idealogy remains
>> same), solaris ufs, and bsd ufs. I think a merger would be quite
>> useful, and would lead towards an unified file system for the three
>> worlds. Making customers happy. Yes our version of UFS does have a
>> lot of the features, better shadow inode storage, extended
>> attributes, and logging, However, none of them are being fully
>> utilized because of imperfections (si and ext attr, are not utilized,
>> and logging has design issues). If this project can aim to solve the
>> general problems and not write a brand new file system, I see this to
>> be of an immense value to the community. Of course this isn't easy,
>> but thats the challenge.
>
>
> I agree an the general usefull principle, however I think
> a simple port from UFS2 is going to be easier and more
> efficient then attempting a merge of the 2 UFS versions.
>
> ---
> frankB
> _______________________________________________
> ufs-discuss mailing list
> ufs-discuss at opensolaris dot org
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
Posts:
6
From:
Seattle, WA
Registered:
3/9/05
|
|
|
|
Re: Any interest in UFS2 ?
Posted:
Jun 21, 2005 5:11 PM
in response to: batschul
|
|
Frank Batschulat wrote:
> On Wed, 22 Jun 2005 00:47:03 +0200, Shawn Debnath
> <Shawn dot Debnath at Sun dot COM> wrote:
>
>> So there are 3 filesystems that are losely based on the same
>> principles: ext2/3 (very different, but ondisk idealogy remains
>> same), solaris ufs, and bsd ufs. I think a merger would be quite
>> useful, and would lead towards an unified file system for the three
>> worlds. Making customers happy. Yes our version of UFS does have a
>> lot of the features, better shadow inode storage, extended
>> attributes, and logging, However, none of them are being fully
>> utilized because of imperfections (si and ext attr, are not utilized,
>> and logging has design issues). If this project can aim to solve the
>> general problems and not write a brand new file system, I see this to
>> be of an immense value to the community. Of course this isn't easy,
>> but thats the challenge.
>
>
> I agree an the general usefull principle, however I think
> a simple port from UFS2 is going to be easier and more
> efficient then attempting a merge of the 2 UFS versions.
Wow the opensolaris list is slow in processing emails.
Anyways, BSD's UFS2 depends on softupdates, So a port isn't that easy as
well and as Frank said merge not a good idea. Wouldn't importing the
features from UFS2, as described in McKusick's paper, make more sense?
That is after we have cleaned up our subsystems which we previously
couldn't attend to due to lack of business case. We already have a good
base to start on and I think fixing the remaining issues make more sense.
Here's a short list to consider:
1) Redesign logging (tons of work here, smarter deltas: avoid per block
delta, incorporate the concept of a "range", "region"...etc, make the
reservation code smarter)
2) Add 64 bit block pointer support.
3) Look at protected namespace? closed system and open application namespace
4) Shadow inode storage in Solaris is awesome! keep it!
5) Yes, variable block sizes are useful. The fragment concept has been
dragged on for too long and it causes too much pain for only being
useful in the first 12 blocks of a file.
6) redo fsck/mkfs utilities. they need to be smarter...and need serious
code cleanup....Dworkin! :)
7) add more todos
So it seems like we can "merge" the Solaris UFS and BSD's UFS2 *ideas*
and create the next UFS. UFS2 was designed from a BSD's point of view. I
think OpenSolaris gives us a great opportunity to merge our ideas and
come up with a better UFS, making sure that we don't create a whole new
file system, but fix a lot of idiosyncrasies that already exist.
Shawn
_______________________________________________
ufs-discuss mailing list
ufs-discuss at opensolaris dot org
|
|
|
|
|
