A summary of all of the bad Linux advice in the CentOS thread earlier this month
Tyler Morgan
tyler at xmission.com
Fri May 20 18:40:56 PDT 2011
Hi! My boss has one of our email aliases subscribed to get filepro-list
digested to us and, well, I stumbled across this CentOS discussion
earlier this month and am a little scared at some of the bad advice in it:
http://mailman.celestial.com/pipermail/filepro-list/2011-May/026119.html
"Cent OS migration"
This is in no way meant to be personal at all. I have great respect for
anyone who can deal with Filepro on a day to day basis. However, I
(usually) don't have to; I am a Unix admin and have been since I broke
my parents computer with Slackware on a 0.9x kernel around 15 years ago.
Also, to buy some credibility, I recently setup Filepro 5.0.0 on
VMware-hosted 64bit CentOS 5.6 servers and it seems to work fine
according to the coders. Don't ask about the version. I'm more than
happy to answer any questions on how I made this work but honestly it
was really easy overall. We were halfway to making it work in Debian 6.0
(much more challenging because Debian has ditched a lot of legacy stuff
where CentOS and RHEL have held onto it -- termcap, 32bit support come
to mind). In the interest of practicality we switched to CentOS and had
it running surprisingly quickly.
And of course there is a lot of good information in that interesting
thread as well.
Let's talk about distributions of Unix/Linux first. There was a lot of
bad advice regarding specific Linux distributions and downright horrible
advice regarding BSD. I think it is just a lack of perspective.
CentOS is a perfectly fine RPM-based Linux distribution that is about to
get very slowly screwed over by the for-profit RHEL juggernaut. RHEL
gave us a preview of their view of the future of Linux by announcing
that they were no longer releasing their kernel patches as nicely
SRPM'ed up little files. Instead, it is now up to the community to diff
out their code and create their own patchfiles. Conversely, RHEL is free
to pillage anything it wants from the OSS community without any
unnecessary hassles.
Here is one of the many stories online about that with some good links:
http://www.g-loaded.eu/2011/03/02/rhel-kernel-source-released-with-patches-already-applied/
The fact that RHEL is a for-profit company screws up a lot of the OSS
mentality that makes it so great. Want another example? Try installing
Mono on a RHEL box. You can't without about a dozen commands of
trickery. Even though the RPMs exist, they purposefully make it hard
because Mono is a project that competes directly with aspects of RHEL's
for-profit business model.
And don't take their RHEL-certified thing too seriously. Sometimes that
is important (being able to call VMWare because you're working with
certified hardware is really nice), but sometimes it's just marketing.
I'd tentatively argue RHEL-certified falls into the latter camp;
admittedly without having much experience caring about it. HP Proliants
are alright overall. I like their iLO cards and their cute amber warning
lights on every single thing.
Anyway, I use both RHEL and CentOS at work and each has their
application but honestly we are slowly phasing out RHEL for CentOS and,
eventually, if I have my way, will slowly phase out CentOS for my
personal Linux-of-choice: Debian. Everything under VMWare, of course.
For my needs, both at home and at work, Debian has consistently been the
most balanced distribution in the long time I have used Linux.
Really though, Linux-to-Linux, the only 2 significant differences are
the package management system (RPM vs. Deb vs. BSD's ports tree are the
main contenders) and the release cycle which is a to-each-their-own
situation to me. Some people can't stand how slow Debian is to update,
some people hate having to upgrade 15-30 packages on a monthly basis and
love locking their versions down. It all depends on what you need from
the OS and where your system architecture is going. But, no matter what,
if you're not applying security patches and dealing with the occasional
consequences, you're doing it wrong.
The complete discounting of *BSD really irked me. Okay, sure, Filepro
mailing list -- it doesn't work on BSD. Great to know! But let's not
completely discount something like OpenBSD -- by far the most secure and
best documented, full-featured Unix distribution in existence, or the
amount of *BSD code under your Linux distro's hood.
So much code that so many people could not live without was written by
the BSD folks. You know that free IPSec stack you're using for your
OpenVPN sessions you couldn't live without? Or uh, SSH? Written by
OpenBSD in the 90s. OSX on your Macbook? Based on FreeBSD.
And don't even get me started on how awesome at networking BSD is
(OpenBSD in particular). Don't feel like spending $5 grand on a pair of
HSRP capable Cisco routers for your border? OpenBSD was nice enough to
reverse engineer it and then rewrite it as CARP in such a way where they
couldn't be sued for violating Cisco's patents on HSRP. Now you can
spend about $650 on a pair of commodity-hardware 1U rackmount servers
and make a nearly equally capable border router. Never, ever use Linux
as a router. iptables literally pales, falls over dead, and then
decomposes before your eyes when compared to pf.
I'm not saying you should look to any BSD to host your application --
especially if it's Filepro! Linux is just as good in almost every way
and much easier to manage in my opinion, except when it comes to
networking. However, discounting it so quickly with remarks like "less
than 1% of the community uses it" is really going to rub long-time Unix
nerds the wrong way.
Some remarks on LVM were the next thing that caught my eye as bad
information. I can't count the number of times my life has been made
exponentially easier by using LVM2 for everything.
I may be misreading slightly, but it was implied that LVM suffers from a
similar issue as RAID0 -- if you lose 1 physical chunk of it, the whole
array is gone. That is sort of true, but definitely not the whole story.
If you lose any physical volume in a volume group, all of your logical
volumes that belong to that volume group are gone -- no argument here.
This is why you have to plan your (redundant) physical volumes carefully
before putting LVM on top of them.
It's actually a very similar situation to picking what level of RAID you
want. LVM can be configured as a "JBOD" which suffers the same risks as
any other JBOD. This is hardly anything to hold against it. The totally
reliable, sort of magical, well supported, well documented, and easy to
use virtual "partitioning" system LVM provides can make your life really
easy when you're moving stuff around or need to rethink your disk situation.
And that leads me into answering a question someone asked that never got
a reply: RAID5 is still significantly slower for writing than RAID10 (or
RAID0+1, your preference). For reading you can probably make RAID5
faster but honestly I use RAID10 for anything that has to be fast. RAID1
plus the occasional hot spare for anything that can be averagely fast
and is important. RAID5 for large amounts of storage that is expected to
be sort of slow to write to (but pretty speedy to read from!).
Oh Bill. I respect your oldschool ways but /u, really? Don't use /u and
the word competent in the same sentence. This is 100% preference and has
nothing to do with skill level. And I 100% prefer to not make a mess of
my filesystem by unnecessarily rearranging things. As far as installing
to /usr/local/* always... well... I can't say I disagree in general, but
at the end of the day: install to where ever makes it easy.
Again, unless there is a really good reason, I am not a fan of making it
hard on myself. Defaults, defaults, defaults -- they are what get QA'ed
most often! Unnecessary complication based solely on preference can lead
to nightmare webs of pain after a decade of doing it and not cleaning up
after yourself. Or, frankly, not accepting times change.
Speaking of which: "If it's not tape, it's not a backup" -- heh, well,
no. Sure, tape has its uses, but again, there is absolutely no reason to
think enterprise-level magnetic drives are not equally reliable and come
with a whole slew of benefits over tape. Instead of worrying so badly
about what media you're backing up to, why not make sure you're backing
up to multiple locations. And monitor them daily. And test restores
often. And break stuff on purpose occasionally. These are the things
that are important when you're talking about backups -- not the media.
Lot's of /good/ advice on not touching SCO and using real RAID
controllers! They are worth every penny.
Anyway, I know this isn't a Linux or BSD mailing list, but it's always a
best practice to have different perspectives on questionable statements
or advice. I sincerely mean no disrespect and wish everyone well!
Have a nice weekend,
Tyler
More information about the Filepro-list
mailing list