[Subject Prev][Subject Next][Thread Prev][Thread Next][Subject Index][Thread Index]

Re: Free software companies and stock options



Amarendra GODBOLE wrote:

> But rebooting a corporate server, on which say people from 'n' no. of
> cities where the corporation has its branches, might be working
> is _DEFINITELY_ a _BAD_ idea.
> Think of the loss in productivity of, say yahoo, if their servers go down
> frequently.
> If a server has to be reboot to fix the problem, then the very purpose of
> it being a server is defeated. :-)

I'm referring to a corporate intranet server where the number of users
ranges from 100 to 500. Few will notice if it goes offline for a few
minutes.

And rebooting is better than it not working at all.

My point simply was that Linux is *not* as reliable as is made out to
be. Taking an example case:

I setup a mail server for CHIP magazine (the chip-india.com domain) in
around July last using fetchmail 5.0 and qmail 1.03 (based on Red Hat
6.0). This was a multidrop (virtualdomain) solution meant as a
replacement for QuantumLink Communications' shoddy PostMaster product.

Now qmail is very finicky about following SMTP standards. If someone
tries to violate them, qmail will simply refuse to accept the message.
One of these SMTP requirements is that messages should have lines
terminated with CR/LF, not just LF in Unix style.

One mail client, WorldMail standard, IIRC, didn't follow this standard,
and someone using this tool sent a mail to someone @chip-india.com every
few days. fetchmail would download this message then send it to qmail
via SMTP, which would promptly reject the message. And fetchmail would
crash with a SIGPIPE.

The problem confounded me for a few months. I simply couldn't figure out
why fetchmail was crashing with SIGPIPE. I checked updates.redhat.com,
but there no update since fetchmail 5.0. I mailed fetchmail-friends, and
ESR responded with a one-liner that my MTA was broken.

So every few days I had to check if fetchmail was crashing, then telnet
to the POP server, delete the offending message, and start fetchmail
again. I could not take a single day off (including week-ends) during
this period because if fetchmail crashed again, no one else knew what to
do. And this wasn't even my job. I was a writer, not an administrator. I
had volunteered to setup Linux, the "wonderful, robust, scalable"
operating system, and it had in return chained me to the mail server.

Then one day, a few months later, I found an update to fetchmail on
freshmeat (a site I had recently discovered). It turned out that there
had been numerous updates since 5.0 and the bug that my fetchmail had,
had been fixed before I even setup the server (via the "forcecr"
option). But updates.redhat.com had never bothered to carry it. You can
imagine how furious I was then.

The same machine is still the mail server, now controlling all of
jasubhai.com, chip-india.com, crn-india.com, nc-india.com, drchip.com
and lunateks.com. It doesn't have problems anymore, there's been
virtually no maintenance done for many months, and my uptime goes to
about 70 days until there's either a power failure or some moron pulls
the plug (Ctrl+Alt+Del at the console is disabled).

But the point of all this is: Do not expect a smooth ride if you're
planning to move to Linux. No distro I've used so far has worked without
extensive customisation, and I've *never* gotten things running the
first time I tried. My mail server still has problems with IMAP (hint:
if you're serious about IMAP, keep a good distance from UW-IMAP (which
RH comes with). It sucks).

Tools like linuxconf are particularly dangerous. linuxconf has almost
always messed up my system's configuration. I once used it on
zdnetindia.com to change the timezone from EDT to IST. The thing didn't
give me a way to pull down the combo box using the keyboard. I finally
gave up and canceled my way out, and linuxconf went and deleted all the
aliases on eth0. I was left wondering for many minutes why telnet to
port 80 of zdnetindia.com was giving me "no route to host" instead of a
more reasonable "connection refused".

I would still prefer Linux over anything else including Windows
(*horror*), but I've learnt that there is only one way to use Linux in a
corporate environment: OUT SOURCE!

Do not try to be smart ass and get Linux running in your company. Make
sure that you either have a qualified team that will be available *all*
the time, or out source the job to a company like FreeOS.com (Prakash
Advani incidentally, was originally responsible for convincing Gourav
Jaswal to put crn-india.com and later chip-india.com on Linux instead of
buying another expensive license for PostMaster). Trust me, it will save
you a lot of pain.

-- 

Kiran Jonnalagadda
http://lunateks.com

baby.sh: while true; do echo "^G^G^G^G^G"; sed -e 's/food/poop/';
sync; sync; sleep 15; done