[9fans] fs (config) questions

[9fans] fs (config) questions

[Axel Belinfante]

I'm trying to build a file server. The available documentation
gives me a pretty good idea of how to configure it (or so it seems),
provided I know what I want. The question is, given the hardware available
to me, what would be a 'sensible' configuration?

Recent file server discussion suggested to me at least two rules of thumb:
 1) keep it simple
 2) when 'stealing' one (or more) partition(s) from a pseudo cached worm,
    take them from the cache disk, and not from the pseudo worm
 3) go back to rule 1)
Are there other 'rules'? For example, for a cached (pseudo) worm, how big
should the cache be, e.g. relative to worm, or to number of users or ...?

The hardware available to me consists of a pc with two scsi controllers.
The first controller controls a single 4Gb disk.
The second controller controls two 8Gb disks.

My idea was to use the two 8Gb disks as pseudo-worm,
and (part of?) the 4Gb disk as cache.
How much cache would be sensible?
Should I concatenate the two 8Gb disks, or is it better
to interleave them?
Would it make sense to define partitions on the 8Gb disks?
(probably not, because of KISS)

The remainder of this messages describes my attempts so far.
Unfortunately, I did not capture all error messages so it is
a bit vague. I will try again, and come with more concrete
questions and details if I encounter problems.
I hope this still provides enough information for someone to react
on the configurations I tried.

Before the March 27 update I tried to boot the pc with a 9pcfs kernel,
but it did hang even before printing the config: prompt.
With the march 27 update I got to the config: prompt.
I tried as firs configuration (typing from memory, because the actual
attempt I made already a little more than a week ago)
(omitting the ip/ream/end stuff)

  filsys main c[p(w0)0.25p(w0)25.25p(w0)50.25]f[w1.<0-1>.0]
  filsys dump o
  filsys temp p(w0)75.25

but this failed (error message about the config, I think, but I did
not see the what's wrong with it).

Then I tried the simpler

  filsys main cp(w0)0.25f[w1.<0-1>.0]
  filsys dump o
  filsys temp p(w0)25.75

which also did not work (again, I think, I got some error message,
without understanding the error)

The even more simple

  filsys main cw0f[w1.<0-1>.0]
  filsys dump o

seemed to work at first, but this no longer gives me un-dumped temp space.
Moreover, after reboot of the fs it had problems reading the configuration
(I recall error messages like 'filsss unknown command' or something similar:
 an error message containing what seems to be a fsconfig command, but
 with some character duplicated, replacing the original character,
 giving an erroneous command)
Does this sound familiar to someone?i

I intend to (soon) retry building the file server, and then try to have
a laptop connected to the fs as console (the output scrolls too fast,
and, worse, disappears when the fs kernel crashes and reboots.)
I still have to read the man pages, but I recall having seen a 'cons'
command or something similar, and a plan9.ini console config entry.

Thanks in advance for any feedback,
Axel.

Re: [9fans] fs (config) questions

[Axel Belinfante]

To follow up on my own message:
I just more or less completed setting up the file server.
I'm still interested in feedback on the more general question of
what would be sensible filesys configs.

Another, maybe a bit naive, question: the worm-like nature of the
pseudo worm will cause it to fill up eventually.
Would it be possible to plug in another disk and configure that
as part of an existing pseudo-worm, like if you would be adding
a worm disk/platter(?) to a jukebox which always has one more
free slot?

W.r.t. the fileserver I just brought up:
There are probably still some rough edges, but I can boot the
auth/cpu server from it, and connect to it via drawterm.
However, when I ssh to it, it dies:
telnetd 202: suicide: sys: trap: fault read addr=0x4326 pc=0x00004326
telnetd 196:die
Any ideas where to look for that? (after the eastern weekend :-)

Axel.


> I'm trying to build a file server. The available documentation
> gives me a pretty good idea of how to configure it (or so it seems),
> provided I know what I want. The question is, given the hardware available
> to me, what would be a 'sensible' configuration?
>
> Recent file server discussion suggested to me at least two rules of thumb:
>  1) keep it simple
>  2) when 'stealing' one (or more) partition(s) from a pseudo cached worm,
>     take them from the cache disk, and not from the pseudo worm
>  3) go back to rule 1)
> Are there other 'rules'? For example, for a cached (pseudo) worm, how big
> should the cache be, e.g. relative to worm, or to number of users or ...?
>
> The hardware available to me consists of a pc with two scsi controllers.
> The first controller controls a single 4Gb disk.
> The second controller controls two 8Gb disks.
>
> My idea was to use the two 8Gb disks as pseudo-worm,
> and (part of?) the 4Gb disk as cache.
> How much cache would be sensible?
> Should I concatenate the two 8Gb disks, or is it better
> to interleave them?
> Would it make sense to define partitions on the 8Gb disks?
> (probably not, because of KISS)
>
> The remainder of this messages describes my attempts so far.
> Unfortunately, I did not capture all error messages so it is
> a bit vague. I will try again, and come with more concrete
> questions and details if I encounter problems.
> I hope this still provides enough information for someone to react
> on the configurations I tried.
>
> Before the March 27 update I tried to boot the pc with a 9pcfs kernel,
> but it did hang even before printing the config: prompt.
> With the march 27 update I got to the config: prompt.
> I tried as firs configuration (typing from memory, because the actual
> attempt I made already a little more than a week ago)
> (omitting the ip/ream/end stuff)
>
>   filsys main c[p(w0)0.25p(w0)25.25p(w0)50.25]f[w1.<0-1>.0]
>   filsys dump o
>   filsys temp p(w0)75.25
>
> but this failed (error message about the config, I think, but I did
> not see the what's wrong with it).
>
> Then I tried the simpler
>
>   filsys main cp(w0)0.25f[w1.<0-1>.0]
>   filsys dump o
>   filsys temp p(w0)25.75
>
> which also did not work (again, I think, I got some error message,
> without understanding the error)
>
> The even more simple
>
>   filsys main cw0f[w1.<0-1>.0]
>   filsys dump o
>
> seemed to work at first, but this no longer gives me un-dumped temp space.
> Moreover, after reboot of the fs it had problems reading the configuration
> (I recall error messages like 'filsss unknown command' or something similar:
>  an error message containing what seems to be a fsconfig command, but
>  with some character duplicated, replacing the original character,
>  giving an erroneous command)
> Does this sound familiar to someone?i
>
> I intend to (soon) retry building the file server, and then try to have
> a laptop connected to the fs as console (the output scrolls too fast,
> and, worse, disappears when the fs kernel crashes and reboots.)
> I still have to read the man pages, but I recall having seen a 'cons'
> command or something similar, and a plan9.ini console config entry.
>
> Thanks in advance for any feedback,
> Axel.

Re: [9fans] fs (config) questions

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 1155 bytes --]

each time i've considered that question (each time
i've configured a new file server) i've eventually decided
in favour of using the pseudo-worm driver.
it isn't more resilient in the face of certain crashes
but it might be slightly more likely to detect an inconsistency
sooner rather than later (or at least more cleanly).
i could be wrong but i suspect a few supporting commands rely on getting
a read error for blocks not yet written (cwcmd prchain might be one).

i'm glad i did keep the bitmap the last time, for another reason.
the written-block bitmap also turned out to be quite
useful when copying the existing structure to a real
worm jukebox, since it said which blocks had actually
been written, and helped to set precise limits for the copy.

another useful hint when copying a
pseudo-worm to a new file server with a real jukebox
and a new set of cache discs with a different structure is to do
	cwcmd savecache
on the file server, and {disk/exsort -w} before
starting the copy.  that way, cwcmd loadcache
will reload the cache on the new file server
with much less jukebox activity (and that's an understatement).


[-- Attachment #2: Type: message/rfc822, Size: 2273 bytes --]

From: "Russ Cox" <rsc@plan9.bell-labs.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] fs (config) questions
Date: Fri, 13 Apr 2001 14:08:38 -0400
Message-ID: <200104131808.OAA20972@smtp2.fas.harvard.edu>

	yes it is possible, but not quite that easy (yet).  i've done it several times,
	using a specially-modified file server kernel to initialise the
	bitmap for the pseudo-worm area
	(after that the normal code is happy to include and write to the extended space).

is there a reason to use the pseudo-worm driver at all?
as far as i could tell, it's just making sure that blocks don't
get written twice.  that's important, i guess, assuming the
cached worm driver is debugged, i'd expect to be
able to use `cw1w2' to the same effect as `cw1fw2' except
that the former doesn't have the bitmap and thus can be
lifted from one disk onto another without a special kernel.

am i missing something?  (probably.)

russ

Re: [9fans] fs (config) questions

[Russ Cox]

	yes it is possible, but not quite that easy (yet).  i've done it several times,
	using a specially-modified file server kernel to initialise the
	bitmap for the pseudo-worm area
	(after that the normal code is happy to include and write to the extended space).

is there a reason to use the pseudo-worm driver at all?
as far as i could tell, it's just making sure that blocks don't
get written twice.  that's important, i guess, assuming the
cached worm driver is debugged, i'd expect to be
able to use `cw1w2' to the same effect as `cw1fw2' except
that the former doesn't have the bitmap and thus can be
lifted from one disk onto another without a special kernel.

am i missing something?  (probably.)

russ

Re: [9fans] fs (config) questions

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 983 bytes --]

>>Another, maybe a bit naive, question: the worm-like nature of the
>>pseudo worm will cause it to fill up eventually.
>>Would it be possible to plug in another disk and configure that
>>as part of an existing pseudo-worm, like if you would be adding
>>a worm disk/platter(?) to a jukebox which always has one more
>>free slot?

it isn't naive.
yes it is possible, but not quite that easy (yet).  i've done it several times,
using a specially-modified file server kernel to initialise the
bitmap for the pseudo-worm area
(after that the normal code is happy to include and write to the extended space).
see http://www.caldo.demon.co.uk/home9.html and
http://www.caldo.demon.co.uk/work9.html
(the list of machine configurations is a bit out of date).

>   filsys main c[p(w0)0.25p(w0)25.25p(w0)50.25]f[w1.<0-1>.0]

interleaving this way on a single disc always seemed slower to me.
it might make sense across dedicated drives, but i haven't tried that recently.


[-- Attachment #2: Type: message/rfc822, Size: 6739 bytes --]

From: Axel Belinfante <Axel.Belinfante@cs.utwente.nl>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] fs (config) questions
Date: Thu, 12 Apr 2001 21:44:31 +0200
Message-ID: <200104121944.VAA27862@polya.cs.utwente.nl>

To follow up on my own message:
I just more or less completed setting up the file server.
I'm still interested in feedback on the more general question of
what would be sensible filesys configs.

Another, maybe a bit naive, question: the worm-like nature of the
pseudo worm will cause it to fill up eventually.
Would it be possible to plug in another disk and configure that
as part of an existing pseudo-worm, like if you would be adding
a worm disk/platter(?) to a jukebox which always has one more
free slot?

W.r.t. the fileserver I just brought up:
There are probably still some rough edges, but I can boot the
auth/cpu server from it, and connect to it via drawterm.
However, when I ssh to it, it dies:
telnetd 202: suicide: sys: trap: fault read addr=0x4326 pc=0x00004326
telnetd 196:die
Any ideas where to look for that? (after the eastern weekend :-)

Axel.


> I'm trying to build a file server. The available documentation
> gives me a pretty good idea of how to configure it (or so it seems),
> provided I know what I want. The question is, given the hardware available
> to me, what would be a 'sensible' configuration?
>
> Recent file server discussion suggested to me at least two rules of thumb:
>  1) keep it simple
>  2) when 'stealing' one (or more) partition(s) from a pseudo cached worm,
>     take them from the cache disk, and not from the pseudo worm
>  3) go back to rule 1)
> Are there other 'rules'? For example, for a cached (pseudo) worm, how big
> should the cache be, e.g. relative to worm, or to number of users or ...?
>
> The hardware available to me consists of a pc with two scsi controllers.
> The first controller controls a single 4Gb disk.
> The second controller controls two 8Gb disks.
>
> My idea was to use the two 8Gb disks as pseudo-worm,
> and (part of?) the 4Gb disk as cache.
> How much cache would be sensible?
> Should I concatenate the two 8Gb disks, or is it better
> to interleave them?
> Would it make sense to define partitions on the 8Gb disks?
> (probably not, because of KISS)
>
> The remainder of this messages describes my attempts so far.
> Unfortunately, I did not capture all error messages so it is
> a bit vague. I will try again, and come with more concrete
> questions and details if I encounter problems.
> I hope this still provides enough information for someone to react
> on the configurations I tried.
>
> Before the March 27 update I tried to boot the pc with a 9pcfs kernel,
> but it did hang even before printing the config: prompt.
> With the march 27 update I got to the config: prompt.
> I tried as firs configuration (typing from memory, because the actual
> attempt I made already a little more than a week ago)
> (omitting the ip/ream/end stuff)
>
>   filsys main c[p(w0)0.25p(w0)25.25p(w0)50.25]f[w1.<0-1>.0]
>   filsys dump o
>   filsys temp p(w0)75.25
>
> but this failed (error message about the config, I think, but I did
> not see the what's wrong with it).
>
> Then I tried the simpler
>
>   filsys main cp(w0)0.25f[w1.<0-1>.0]
>   filsys dump o
>   filsys temp p(w0)25.75
>
> which also did not work (again, I think, I got some error message,
> without understanding the error)
>
> The even more simple
>
>   filsys main cw0f[w1.<0-1>.0]
>   filsys dump o
>
> seemed to work at first, but this no longer gives me un-dumped temp space.
> Moreover, after reboot of the fs it had problems reading the configuration
> (I recall error messages like 'filsss unknown command' or something similar:
>  an error message containing what seems to be a fsconfig command, but
>  with some character duplicated, replacing the original character,
>  giving an erroneous command)
> Does this sound familiar to someone?i
>
> I intend to (soon) retry building the file server, and then try to have
> a laptop connected to the fs as console (the output scrolls too fast,
> and, worse, disappears when the fs kernel crashes and reboots.)
> I still have to read the man pages, but I recall having seen a 'cons'
> command or something similar, and a plan9.ini console config entry.
>
> Thanks in advance for any feedback,
> Axel.