[NCLUG] Advanced Server 2.1 Woes and Wows
DJ Eshelman
dj at sgc-inc.net
Tue Jul 23 13:53:44 MDT 2002
This message isn't one of those "please help me" messages, more like one of those
"what is
this
crap" letting off steam messages, but also might contain some info for some of you
out
there.
So, we've been deploying Red Hat's Advanced Server 2.1 for the last month in
our
datacenter, with good and bad results.
It's stable as hell, and actually fairly
flexible
if you get used to using Kernel Parameters instead of recompiling the kernel (a
new
one for us,
to be sure).
Once you get it working, it positively screams. We used it
for a pair of new
servers for DB2 EEE, and we're seeing turnaround times on our reports that
are
positively
unreal- but it took a long struggle to get there. Basically these servers
are
awesome pieces of
hardware- dual Xeon 2.2's, 8GB of RAM each (which requires the
enterprise level
kernel to be
installed ortherwise it only sees 4GB- another lesson that
was learned the hard way)
and of
course gigabit ethernet, tons of SCA-SCSI RAID space and
all the trimmings... But
from our
experience let me lay down some ground rules for you if
you plan on using Advanced
Server (in
other words, learn from our mistakes). It could
save you some serious time- in our
case, about
three weeks from initial build to
deployment:
1) Under all circumstances- run, flee,
grab a gun and shoot behind you
before using Adaptec hardware. So far we've lost an
amount of
money I can't even divulge
because we've bought a good lot of Adaptec hardware that
isn't
compatable with Advanced
Server. We're talking RAID controllers, Fibre Channel,
everything- if it isn't on
RedHat's HCL, don't buy it (at least from Adaptec). We
went
(eventually) with 3ware for
our backup server's IDE-RAID, LSI/AMI MegaRAID for our
SCSI
RAID, and are now looking for
a good 2-Gigabit FibreChannel solution. In all of the
above we
first bought Adaptec
hardware because they claim to like Linux. Don't buy it, it's an
outright lie. Remember
when Adaptec was concidered the best? Forget it. Those
days are done-
they are on the
Microsoft Train to HCL Hell.
2) If you fancy a good laugh, take a gander at
RedHat's AS
2.1 HCL. Then take a look at how much support packages cost for AS.
Try to contain
yourself
if you're at work.
3) Once you stop laughing and the pure horror sets in that
nothing you
have is supported, it might be fun to realize that it's not supported
becuase of
the
vendors, not because of RedHat. When you think about it, it makes perfect
sense. Why
would
you certify hardware blindly on your most mission-critical OS? After
extensive
conversations with RedHat reps and techs, the conclusion is that if vendors
don't
give them
drivers to test, they won't step out on a limb about them. I flat got
told
concerning these last
'drivers' that I could in fact compile them into the kernel
and it would probably
work... but
since they hadn't certified them, our Kernel would
become 'custom' and would no
longer be
supported. Bottom line, we need to start pushing
vendors towards this sort of
thing. If we
(the customers) don't demand it, it will never
get done </soapbox rant>
4) RedHat's
support really is stellar. For people like me who
make stupid mistakes, it's great.
But
mostly it's a timesaver. You can spend 5 hours
hacking at a problem, or you can
make a quick call
and usually have it solved or at least on
it's way in less than an hour. That is
pretty cool. But
it does take some dropping of the
pride... For example, when we upgraded our memory
from 4 to
8GB the OS was only seeing 4GB-
wierd stuff. So I spent all this time looking thru
documentation
on kernel parameters,
blah, blah, blah- I ended up calling and right away they said
"Oh, just
install the
enterprise kernel RPM" Lesson: if you paid for it- use it.
5) Don't make
assumptions.
Our very first mistake was assuming this distribution was going to be
updated
regularly-
quite the contrary; Red Hat's goal is to stablize on this kernel (2.4.9-e
or 2.4.9-
enterprise; yes they are different). The good news is that there are plenty
of
optimizations
for multiple situations (big memory, Athlon MP, all sorts of
goodies). The rub here
is that it
often means that a lot of 'standard' RedHat or even Linux
drivers and programs may
not operate
the same way you may be expecting. My point here is
that you shouldn't assume
anything; forget
the concept of recompiling the kernel (if
you want support, anyway) and research
everything
before you sign the PO!
Well,
given the cost I doubt many people will be deploying AS 2.1,
but just in case, I thought I'd
give you all a very small heads up/rant. I have to
say that all in
all we are very happy with
it, but we're forging new territory and that's always a
hard thing to
do.
I've said my
bit- I'd better get back to work. Maybe I'll keep you all posted on the
FibreChannel debacle
if you're interested...
-DJ Eshelman
The IT World's Bitch
More information about the NCLUG
mailing list