Neil Neely wrote:
> We're looking at integrating our *nix machines with our AD servers and  
> are trying to find the "Best" way to do this.  In this case I'm  
> finding my google-fu isn't working in my favor... there is no shortage  
> of information.  Every time I think I have a complete grasp of ways  
> this can be done I find one more.  So there are plenty of resources  
> for how to do this using technique X, what I really need is some  
> feedback from people who are further along in this evolution that can  
> give some perspective on which approach they think is the best.
>
> Disclaimer:  I am in the process of learning how these bits fit  
> together, and if I've said something truly bizarre it is likely out of  
> ignorance not arrogance so I really would appreciate being pointed in  
> the right direction.
>
> Relevant background details:
> ~50 production servers that are centrally managed (unified UID and  
> passwords) using homegrown syncing - we would like to move these to AD
> Already have AD infrastructure in place authenticating staff work  
> stations (~50 workstations)
> The servers exist to support our customers (not staff in general)
> These servers do not require shared home directories for staff.
> Staff accessing these servers are all performing some task relating to  
> "administration", though at different levels (tech support through sys  
> admin).
>       * primary concern is not securing these machines against it's  
> legitimate users (so NIS may be acceptable in this environment).
> This economy stinks and doing this without any capital expenses is  
> very important.
>
> Combinations we are seriously considering (in no particular order):
>
> NIS w/Kerberos (via SFU)
>
> Winbind
>
> Likewise Open
>
> We've found various bits and pieces that seemed promising with each of  
> these approaches.  This is our short list of best fit for the problems  
> we've got, but perhaps we've overlooked something.  I would really  
> appreciate any pro's/con's from the trenches on this topic.  "Likewise  
> Open" seems to be the easiest to install at this point, so is slightly  
> ahead in our evaluation.
>
> Thanks for your time,
>
> (sidenote:  AD is being chosen because it is existing established  
> infrastructure here that looks like it will do the job we need,  
> nothing at all against openldap, this is just using the tool that  
> we've got so we can focus on solving other challenges.)
>
> Neil Neely
> http://neil-neely.blogspot.com
>   
For Open packages, look at Luke Howard's PADL products 
<http://www.padl.com>. When I was last working on this (a couple of 
years ago), PADL didn't have caching, but I believe that this has been 
done since then. Luke is very dedicated to the work, and is responsible 
for creating the standard (RFC2307) that is accepted as 'the standard' 
for  Unix to AD connectivity. His company also offers commercial support 
if that is important to your organisation.

The two commercial packages that I've seen that are built to to this are 
Quest(Vintela) and Centrify.

Microsoft's SFU is underpowered, but if you have only a few users and 
machines, you might be able to get away with it for now. I can't 
recommend it for any sizable organisation.

SFU does not scale well as all lookups are done directly to AD - it's 
meant for a large Windows site that has a very small Unix 
infrastructure. Try doing an 'ls -l' in a directory with files owned by 
hundreds (or thousands) of different users - and be prepared to wait....

Quest and Centrify have local caching of AD data, so the load on your AD 
servers is minimal most of the time, and there is no latency on 
obtaining user data.

For 'future-proofing' look for RFC2307bis compatibility (unfortunately, 
this RFC was never brought through for ratification, but it does 
describe Kerberised LDAP compatibility with functionality that scales to 
large infrastructures).

All of the products (including SFU) require extensions to the AD schema, 
and since AD schema changes are irreversible once installled, experiment 
with whichever product(s) that you are interested in on a test domain 
before committing yourself. VMware is your friend - I did a lot of 
playing around with Win2K3 servers running under VMware, using snapshots 
to play 'Groundhog Day' with the AD data (instant schema change reversal 
:-).

AD is a solid back end service. I hammered AD servers during testing, 
and the only way that I could make them crash was to do a nasty DOS 
attack (by accident :-) - I hammered Kerberos authentication requests 
without closing the TCP connections, and eventually ran out of sockets 
on the server... For every other load test that I performed, AD degraded 
'gracefully', i.e., it came up to a maximum number of transactions per 
second, and held it - regardless of additional load applied. Further 
load did not cause it to 'crowbar' (it didn't drop off in the number of 
transactions per second as more load was applied). I was not expecting 
this kind of solid performance from a Windows server (having heard the 
horror stories about IIE and Exchange). AD on Win2K3 also expands nicely 
when you add replicas. It was also 5-6 times faster than the Solaris 
NIS+ servers that we were running at the time when retrieving individual 
results. Where AD (and every other LDAP server that I've seen) falls 
down (compared to NIS+) is when you 'walk a map', since the NIS+ 
protocol is built to deliver the entire map in large chunks, and LDAP 
returns each entry individually. This is why you want caching in the 
client - map walks become 'free', and the load on your AD (or other LDAP 
server) is greatly reduced.

OTOH, I'm an 'OS gourmand' - I eat them all! While I admit to a little 
bias, I'm do try to use the OS/platform/application that does the best 
job for a given service (which also must take into account local 
expertise for maintaining the system). If you have competent Windows 
admins taking care of your AD infrastructure (as my $WORK does), I would 
have no qualms about basing your Unix authentication and authorisation 
on AD. Having dealt with Sun and their NIS+ fiascos for a number of 
years (they didn't really get their act *mostly* together until Solaris 
10), AD on Win2K3 was a pleasant change! (AD on older versions of 
Windows has some performance and configuration limitations that were 
corrected in 2K3.)

- Richard


_______________________________________________
Tech mailing list
[email protected]
http://lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to