On Tue, 15 Feb 2000, Benjamin Scott wrote:

> On Tue, 15 Feb 2000, Derek Martin wrote:
> > Ben, PHILISOPHICALLY I agree with you, but PRACTICALLY speaking the
> > problem is that NFS doesn't guarantee that there will be no FS corruption
> > if you use the soft option.  This is a Bad Thing.

>   Hmmmm, by filesystem corruption, do you mean actual damage to the filesystem
> structure itself, or simply lost user data?  I've never had any *serious*

I can't answer that, but does it matter?  All I can tell you is what I've
read, and I can tell you that what I read came out of the O'Reilly NFS
book.  Actually, since I have the book handy, I can look it up and tell
you...

Looks like both. Specifically, with an interrupted write operation, you
can lose data.  If the write was modifying the inode table (I have no idea
how this is handled on NFS...) then I imagine it's possible that you'll
hose the FS.  If the application doesn't check the status of the write
call, you could end up with some serious data corruption.

Now, I imagine the AMOUNT of corruption is largely implementation
dependent, but being unfamiliar with the implementation I'm not willing to
go there... ;-)

It gets worse though... if you're paging in a swapped process on some
systems, they will re-read the text segment from the disk (or NFS share,
in this case), rather than swapping them into swap and paging them back in
from swap (which saves you the time and trouble of writing them to swap,
since it's already on disk...) and if the system loses contact with the
server that the executable was shared from, it will cause system-dependent
problems ranging from swap errors to kernel panic. BAAAAD.

> problems with "soft" NFS on Linux.  Now, if a connection is lost of course the
> program won't be able to complete the I/O it was trying, but that'll happen
> regardless.  And some programs actually check the result of their I/O.

Yep, and some don't...

>   This may have changed with the kernel NFS driver, though.  While the
> usermode NFS driver may have been slower, I really preferred the idea of
> keeping NFS in userland where it can't screw up the rest of the system.

I mostly agree, but I also have to acknowlege that if you use NFS a lot
you might be willing to make that sacrifice. 

>   And as far as NFS on Solaris goes -- *barf*!  For the company that invented
> the thing, Sun does a damn poor job of implementing it.  :-(

What, and NIS (/plus) is better? The only things done well on Solaris
were stolen from someone else (most of the OS... hehehe) 8^)

> > Avoid NFS at all cost, if you can.
>                          ^^^^^^^^^^
>   That's the kicker, isn't it?  Samba and SMB actually seem pretty good, as
> network file systems go (despite the fact that the protocol is rather poorly
> documented and the Microsoft implementations suck (as usual)), but it doesn't
> support any kind of idea of Unix permissions.  :-(

Ya. As much as I hate it, the tradeoff in our current environment isn't
really any more desireable.  The one area where we could get rid of it is
to stop exporting the mail spool, but then all the users would gripe since
they'd need to log into the mail server to read mail. No one's gonna like
that. Oh well, I'm quitting anyway. 8^D

> > Have you tried afs or coda? Anyone?
> 
>   I haven't actually *tried* either of these, but from what I've read, aren't
> they huge overkill for many situations?  All I want to do is share some files

I don't know anything about either other than that they are potential
replacements for NFS. I can't say.

> with other machines on my LAN; I don't need a distributed, caching,
> high-availability, offline-capable, binds-the-world-together-and-cures-cancer
> filesystem.  :-)

In that kind of environment I'd imagine a more robust network FS isn't
necessary, and soft mounts would be fine. But in a heterogeneous
production environment, I'd definitely hesitate.

>   Still, I suppose "too much" is better then "not enough".  Another thing to
> add to my to-do list... >sigh<  8^)

Yeah, me too.  Been there for a while, but "work" keeps getting in the
way... ;^)

-- 
"Quis custodiet ipsos custodes?"    "Who watches the watchmen?" 
-Juvenal, Satires, VI, 347 

Derek D. Martin      |  Senior UNIX Systems/Network Administrator
Arris Interactive    |  A Nortel Company
[EMAIL PROTECTED]  |  [EMAIL PROTECTED]
-------------------------------------------------


**********************************************************
To unsubscribe from this list, send mail to
[EMAIL PROTECTED] with the following text in the
*body* (*not* the subject line) of the letter:
unsubscribe gnhlug
**********************************************************

Reply via email to