[9fans] information about filesystems

[9fans] information about filesystems

[Rudolf Sykora]

Hello,

is there any source where I could read about differences between kfs
and fossil (also I heard about something like cwfs), about reasons why
fossil was designed and superceded, when to use which and alike?

Thanks
Ruda

Re: [9fans] information about filesystems

[Steve Simon]

kfs is a simple, non-backed up filesystem designed for stand alone machines

kenfs is the older main file server, it runs stand alone on a single machine
and can use either magnetic disks or WORM disks and provides access to previous
backups online. it supports a narrower range of hardware than the maion kernel
but has its fans due to its simplicity and reliability.

cwfs is n implemention of kenfs as a user program for plan9, it can read and write kenfs
filesystems and can be used as the main filesystem for a system to boot from (and then serve).

fossil is a write buffer which acts as a file system and can be used to replace kfs
on simple machines (laptops), it can take regular temporary snapshots of the system state and
present them as backups (e.g. every 15 mins), though sadly this part of the code seems to be
broken.

Most often fossil is used in conjunction with venti which provides permenant
online backups. venti works a bit like the back end of kenfs except that it adds
a network interface (so other apps and OSs can access venti) and it does compression
before storing blocks.

both kenfs and venti strip duplicate blocks so as to reduce the size of their backups,
venti detects duplicates using an sha1 checksum so copying files on a venti backed
has almost no overhead (i.e. only the directory entries).

I run two combined file/cpu/auth servers with fossil and venti on both, others run
kenfs on a dedicated machine and still others run cwfs on a file/cpu/auth server.

its a complex debate and I'am sure others will have opinions.

see here for some more info:

kenfs:
	http://plan9.bell-labs.com/sys/doc/fs/fs.pdf
	http://plan9.bell-labs.com/who/seanq/cw.pdf
	http://plan9.bell-labs.com/magic/man2html/4/fs

venti
	http://plan9.bell-labs.com/sys/doc/venti/venti.pdf
	http://plan9.bell-labs.com/magic/man2html/8/venti

fossil:
	http://plan9.bell-labs.com/sys/doc/fossil.pdf
	http://plan9.bell-labs.com/magic/man2html/4/fossil

cwfs:
	http://plan9.bell-labs.com/magic/man2html/4/cwfs

kfs:
	http://plan9.bell-labs.com/magic/man2html/4/kfs

-Steve

Re: [9fans] information about filesystems

[Rudolf Sykora]

thanks for pointers!
R

Re: [9fans] information about filesystems

[Ethan Grammatikidis]

This write-up would make an excellent wiki page, "File System
Comparison" perhaps. Erik's additional notes should probably be
included, or perhaps put in a separate kenfs page. I could make the
pages if desired.

On 13 Aug 2010, at 5:31 pm, Steve Simon wrote:

> kfs is a simple, non-backed up filesystem designed for stand alone
> machines
>
> kenfs is the older main file server, it runs stand alone on a
> single machine
> and can use either magnetic disks or WORM disks and provides access
> to previous
> backups online. it supports a narrower range of hardware than the
> maion kernel
> but has its fans due to its simplicity and reliability.
>
> cwfs is n implemention of kenfs as a user program for plan9, it can
> read and write kenfs
> filesystems and can be used as the main filesystem for a system to
> boot from (and then serve).
>
> fossil is a write buffer which acts as a file system and can be
> used to replace kfs
> on simple machines (laptops), it can take regular temporary
> snapshots of the system state and
> present them as backups (e.g. every 15 mins), though sadly this
> part of the code seems to be
> broken.
>
> Most often fossil is used in conjunction with venti which provides
> permenant
> online backups. venti works a bit like the back end of kenfs except
> that it adds
> a network interface (so other apps and OSs can access venti) and it
> does compression
> before storing blocks.
>
> both kenfs and venti strip duplicate blocks so as to reduce the
> size of their backups,
> venti detects duplicates using an sha1 checksum so copying files on
> a venti backed
> has almost no overhead (i.e. only the directory entries).
>
> I run two combined file/cpu/auth servers with fossil and venti on
> both, others run
> kenfs on a dedicated machine and still others run cwfs on a file/
> cpu/auth server.
>
> its a complex debate and I'am sure others will have opinions.
>
> see here for some more info:
>
> kenfs:
> 	http://plan9.bell-labs.com/sys/doc/fs/fs.pdf
> 	http://plan9.bell-labs.com/who/seanq/cw.pdf
> 	http://plan9.bell-labs.com/magic/man2html/4/fs
>
> venti
> 	http://plan9.bell-labs.com/sys/doc/venti/venti.pdf
> 	http://plan9.bell-labs.com/magic/man2html/8/venti
>
> fossil:
> 	http://plan9.bell-labs.com/sys/doc/fossil.pdf
> 	http://plan9.bell-labs.com/magic/man2html/4/fossil
>
> cwfs:
> 	http://plan9.bell-labs.com/magic/man2html/4/cwfs
>
> kfs:
> 	http://plan9.bell-labs.com/magic/man2html/4/kfs
>
> -Steve
>

Re: [9fans] information about filesystems

[erik quanstrom]

> kenfs:
> 	http://plan9.bell-labs.com/sys/doc/fs/fs.pdf
> 	http://plan9.bell-labs.com/who/seanq/cw.pdf
> 	http://plan9.bell-labs.com/magic/man2html/4/fs

there are some other man pages for ken's fs, too.
the versions i have are a bit different than the sources versions

http://www.quanstro.net/magic/man2html/8/fs
http://www.quanstro.net/magic/man2html/8/fsconfig

- erik

Re: [9fans] information about filesystems

[erik quanstrom]

> is there any source where I could read about differences between kfs
> and fossil (also I heard about something like cwfs), about reasons why
> fossil was designed and superceded, when to use which and alike?

steve's summary is excellent.  a few more words about my favorite,
ken's fs

the "worm" partition doesn't really need to be a worm.  it can
be any sort of device.  i currently use aoe targets.  typically coraid
appliances. but i have a few old fses using scsi disks or ata disks
hanging off ide/marvell/ahci controllers.

the big advantage of ken's fs is speed.  at coraid we top out at
>300mb/s file io on a lowly intel 5000 xeon and a old coraid
appliance.  ken's fs is able to serve so quickly because
- it doesn't waste time compressing (compression is slower than
disks these days!)
- it can have a huge memory cache — nearly all of memory.
- it doesn't need to copy between user space and kernel space;
it also doesn't do anything with the mmu except set up 1024
4mb pages on startup.
- it keeps files less fragmented, due to fact that it doesn't
ensure that blocks are globally unique.

ken's fs also tends to be stable enough to be externally
uptime limited.  things like ups batteries and network hiccups
tend to be more disruptive.

here are a few more ken's fs references.
	http://www.quanstro.net/plan9/disklessfs.pdf
	http://www.quanstro.net/plan9/history.pdf

- erik