[Infrastructures] AFS in an infrastructure

Jenkins, Steven JENKINSS@mail.etsu.edu
Mon, 14 Mar 2005 15:59:57 -0500


As someone else mentioned, you have to watch for your AFS servers having dependencies on AFS itself.  However, from a scalability standpoint, a few 'fat' servers are manageable (eg, build them via rdist/rsync/radmind/cfengine from AFS).  You'll want to carefully checkpoint and back up your data, but, in general, your AFS servers don't need a lot of updates.
 
Directory services could be either LDAP or Hesiod...or even NIS.  NIS is the Great Dirty Secret of Systems Administration -- everyone seems to use it, but everyone also disparages it because of the known security and scalability issues.   I've seen places use NIS for their directory services, and it works fine -- you just have to understand the risks and issues, just like any other technology.
 
File replication is probably _not_ going to be 'just copying off AFS'.  I suspect it will be a combination of the replication technology in AFS itself, dump/restores of data (for snapshotting purposes), and 'just copying'.  For some of the issues involved, see the description of the Global File System (VMS) in http://www.usenix.org/publications/library/proceedings/lisa95/gittler.html.  That's a pretty good coverage of what Morgan Stanley did in the mid 90's.  On the other hand, the university implementations I'm aware of don't try to do a multiple cell infrastructure with the illusion of a single cell, so your achitecture might be much simpler than what Morgan Stanley did.  In any case, you'll want to use the 'volume replication' technology in AFS instead of straight copy most of the time.  When I worked at Iowa State, we did straight volume replication for file replication, and it worked fine -- we were a small team and could avoid inconsistencies simply by letting each other know what we were doing.
 
As far as 'Gold Server', you can use a Gold Server as what goes into AFS or your can use AFS itself as the Gold Server.  For the former, you build a Gold Server, and then dump contents into AFS.  For the latter, you don't actually have a Gold Server anywhere.  There are arguments for and against using a Gold Server in that type of environment -- I don't have a personal view one way or another.
 
Steven
 

________________________________

From: infrastructures-admin@TerraLuna.Org on behalf of Sean Kelly
Sent: Mon 3/14/2005 11:36 AM
To: Steve Traugott
Cc: infrastructures@terraluna.org
Subject: Re: [Infrastructures] AFS in an infrastructure



On Mon, Mar 07, 2005 at 10:57:30PM -0800, Steve Traugott wrote:
...
> AFS is a total pain in the tail end, but it's worth it.  Every time you
> run into another shocking AFS misfeature, just keep reminding yourself
> "I like the end result, I like the end result". 

Now I'm trying to vision how all the pieces mentioned on
Infrastructures.Org fit into an AFS world.

* Version Control could be done over AFS. You could store the
  CVS (or Subversion fsfs) repository on AFS.
* Host install images could be stored on AFS and thus be made available through
  many different machines running BOOTP/DHCP/whatever.
* Directory Services could be done using LDAP or Hesiod, depending on which
  route you went down.
* Authentication would be done through Kerberos.
* Network File Servers are AFS file servers
* File Replication Servers are just copying things off AFS
* Client File Access is AFS
* Client O/S Update could be done with tools pulling patches off AFS
...

Am I going overboard with the AFS thing here? Where does the "Gold Server"
fit in when you have AFS and can just store everything in a
replicated/backed up distributed common namespace?

--
Sean M. Kelly
Assistant Unix Administrator/Programmer
Division of Information Technology
Creighton University
(402) 280-2264
_______________________________________________
Infrastructures mailing list
Infrastructures@mailman.terraluna.org
http://mailman.terraluna.org/mailman/listinfo/infrastructures