[Infrastructures] A few questions

[Michael T. Halligan]

Sean,

Excuse my somewhat rambling response here, I'm very sleep-deprived and 
sunburnt currently.


In response to your question re: RHN, I finished a small (200 server) 
rollout about 3 months ago. I actually
asked about it on this list, about a year ago, and was warned against 
it.. A warning I should have heeded.

The theory behind the product is pretty good. It puts all of your 
configurations into subversion, manages
everything with RPM, a pull-based client written in python,  and an OK 
interface that gives you a flexible
view of your infrastructure, with flexible select tools.

Beyond theory, I'll be brief, It was a trainwreck :

-Sales Engineering sucked. One meeting with the sales engineer, a burnt 
copy of media, he never picked up
his phone again.
- Documentation at first glance is pretty good, but it's about the 
documentation you'd expect from a first-generation
platform rushed to market.  Anything that is not very simple that you're 
trying to do is not documented.  Strong
lack of best-practices documentation, even if you shell out the 2500 
clams to go to training.
- Buggy code. Once every other week we'd run into a bug that they hadn't 
released yet. New versios of the client
came out every two or three weeks.
- Untrained salesmen. The sales people were the best part of dealing 
with RedHat, because they really did try. Unfortunately
nobody seems to be responsible for the RHN product, so the salespeople 
are constantly given (and repeating) very conflicting
information on everything from pricing and licensing to capabilities.. 
They promised a lot of functionality that was two releases
away, not knowing that it was two releases away.. Like being able to 
select multiple servers and apply configuration files to them.

All of this I could have dealt with, it really did save a lot of time 
compared to developing our own system.. You're paying
$180 per box for "rhn support" (which you no longer should need 
technically since you are hosting RHN in your datacenteR)
and $180 per box for a provisioning entitlement.  You're also paying I 
believe $10k for the server software, and anojther $5-$8k
to meet their minimum hardware requirements.  Their minimum hardware 
requirements are about 1/2 what it needs, because there's
a CPU hog built in.. Oracle Embedded.

This system was designed with the thoughts of attaching it to big, fast 
oracle servers. Embedded oracle was at best an afterthought, and
the performance really, really shows.  That's OK, though, because they 
have a $2k proxy software license you can buy, which allows
for a great performance improvement.

All of the negativity I have towards the product, I would still use it 
if there was a strong support organization behind it.  As of two months
ago, there was not.  Their night-time support (which means after 2PM 
PDT) is in australia, where they have one tech who knows Satellite Server..
Unfortunately, he's just a well trained-tech support guy, he's not a 
programmer or an engineer.. You have to wait for morning if anything 
"strange"
breaks. 

The very fast release cycle of this product, with each minor version 
(3.4 -> 3.5 -> 3.6 -> 3.7) adding major features is somewhat
unnerving.  What's more un-nerving is the fact that even though thre are 
quite a few bugs that need fixed, a slew of support processes that neeed
to be ironed out, and a general lack of quality that has to be dealt 
with, RedHat has announced that by the end of the year it will support 
Solaris.

I'm about to start evaluating Novell ZenWorks, and I have a lot more 
interest in ZenWorks, and in Novell's ability to give support.

For now, I think CFEngiine, PXEBoot, Kickstart, Yum, and good change 
management will go a lot further than RedHat's solution will.


Now, for some positive bits about RHN, I'll say that it does have some 
really nifty features. I can click on a server view, select a profile of
a different class of server to sync up to, hit the "provision" button,  
the server will, upon it's next check-in, add or remove rpms in order to
match that new profile.  It also had a nice gui to kickstarting a box, 
where it would, when you choose an existing server to be kickstarted, 
install
a new kernel into grub that is actually a kickstart image, and schedule 
a reboot, making the kickstarting pretty automatic. 

RHN versions everything it seems in one form or another (transactional 
rollbacks in RPM, Subversion for configuration files, etc).. Unfortunately
they missed a few obvious things, like applying their versioning logic 
to their webform for kickstart images.. For that reason, it ended up 
being a lot
better to just put everything in the %post section into a seperate shell 
script, and have the server wget that during %post.


Sean Kelly wrote:

>Seeing as the messages lately seem to be discussing what various
>technologies and methods we are all using, I'd like to throw these out
>there.
>
>What types of distributed filesystems, if any, are people using? I've been
>poking at OpenAFS, GFS, or just old fashioned file replication for our
>solution. OpenAFS looks rather nice, but complicated and very difficult for
>some operating systems such as FreeBSD.
>
>Is anybody out there making heavy use of VMWare? It is being suggested to
>us that we examine the usage of VMWare ESX as a means of virtualizing all
>of our servers, including Oracle. I was curious if anybody out there has
>experience in this area and can give some feedback on this.
>
>And finally, has anybody found the provisioning tools provided by the Red
>Hat Network website to be useful at all? To me, they seems somewhat decent
>but useless when you enter an environment with anything beyond RHEL
>systems.
>
>Thanks for your feedback!
>
>  
>