[wellylug] Debian & NFS

[Pete Black]

enbd (enhanced network block device) is a similar type of setup to 
iSCSI, and also has a rather interesting RAID personality (fr1) that 
will do RAID-5 across multiple network devices and should one fail only 
reconstructs the 'lost' blocks, rather than resynching the entire disk 
like 'vanilla' RAID-5 does.

I have used it in a test setup to mount a  RAID-1 device on a machine 
that was composed of 'disks' which were actually files on 2 other 
machines that were mounted loopback, formatted as ext2, and exported via 
enbd to the first machine, which seemed to work ok - I intended to use 
this as a 'seamless' network backup solution but never actually got as 
far as implementing it in production.

Obviously disk read/write speed will be bound by the speed of the LAN, 
but it is a pretty neat setup if speed is not your primary concern.

-Pete

>iSCSI looks like a good emerging option - I use it on a big storage
>array used by 3 boxes to share the same oracle DB using RAC
>
>http://www.cuddletech.com/articles/iscsi/
>
>it will allow you to export a disk on one machine and use it on another
>where it differs from NFS etc, is that its disks show up as true block
>devices - eg /dev/sda1
>
>this has numerous benefits. However it may well be overkill for what you
>want to do.
>
>Geraint 
>
>
>-----Original Message-----
>From: wellylug-admin at lists.wellylug.org.nz
>[mailto:wellylug-admin at lists.wellylug.org.nz] On Behalf Of Ewen McNeill
>Sent: Wednesday, 6 July 2005 7:15 p.m.
>To: wellylug at lists.wellylug.org.nz
>Subject: Re: [wellylug] Debian & NFS 
>
>In message <Pine.LNX.4.61.0507061836250.29314 at localhost>, David Antliff
>writes:
>  
>
>>Does anyone know what the Real Deal is with the *two* NFS servers in 
>>Debian? There's nfs-user-server and nfs-kernel server. 
>>    
>>
>
>nfs-user-server is entirely implemented in user space.
>nfs-kernel-server is just a front end to the in-kernel NFS server which
>arrived (IIRC) in either the 2.2 or 2.4 kernels (although it isn't
>always
>compiled in).  From memory nfs-user-server has been in Debian (called
>something like nfs-server originally) since the 1.x Linux kernels).
>
>Generally I'd say use nfs-kernel-server if your kernel has support for
>it.
>It should be faster (fewer context switches), and as you've found there
>are various issues with the nfs-user-server implementation (including
>that it apparently isn't compiled with the large files options
>available/turned on).
>
>  
>
>>User xyz has uid 1000 on hostA, but uid 1001 on hostB. HostA exports a 
>>directory and user xyz on hostB can't access it properly because of the
>>    
>>
>
>  
>
>>uid mismatch. 
>>    
>>
>
>Welcome to NFS.  The traditional solution is to synchronise the uids on
>all the hosts to match, either by hand (eg, /etc/passwd pushes) or
>through some sort of directory system (NIS, LDAP, etc).  For a tiny
>network I generally do it by hand; on larger networks I've used both NIS
>and LDAP successfully.
>
>FWIW, these days I'd probably be pretty tempted to use Samba and the
>CIFS
>client built into the kernel (2.6 kernels anyway; I think there's a
>patch
>for 2.4).  Particularly with the unix extensions (supported by both)
>it should be a fairly good solution.  (Not sure how the performance
>would compare with the nfs-kernel-server, possibly a little slower,
>but that may not matter to you.)
>
>  
>
>>Does Samba still have a 2GiB file size limit?
>>    
>>
>
>I believe Samba 3 (eg, in Debian Sarge) is generally built with large
>file support enabled, and I think the in-kernel CIFS client should
>support large files.
>
>  
>
>>Are there any other alternatives to NFS worth looking at?
>>    
>>
>
>AFS (eg, OpenAFS) might also be worth a look.
>
>Personally I still use NFSv3, in kernel, and sync uids by hand.
>
>Ewen
>
>
>  
>