Hi guys,
A hacker friend and I are building some apache servers for a company that's hired us. My first ever apache gig! FTW!!! heh
Okay, my overwhelming enthusiasm for the project aside I could seriously could use some help.
So my friend and I came up with a partition scheme that we thought would be sensible for a potentially busy web server. The machines we are using come with 250GB scsi drives. Ample room for the servers we intend to build.
However, whenever we build servers this way they CONSISTENTLY crash. And neither of us have any idea why. Anyway, here's the plan we came up with:
Code:
swap 4096 mb
/boot 25 mb -ro
/ 15 gb
/usr 25 gb
/var/log 25 gb
/home 50 gb
/var/ftp 25 gb
/dev 5 gb
/var/www 100.9 gb
xfs filesystem
The parts are listed in order they are physically created on the drive. IE swap first, /boot second, / third etc.
However... this is the error we get consistently whenever we try to boot a server built this way:
Code:
Starting portmap daemon...portmap: fork: No such device.
Starting NFS common utilities: statd failed!
/etc/rcS.d/S48console-screen.sh: line 40: cut: command not found
/etc/rcS.d/S48console-screen.sh: line 40: [: =: unary operator expected Setting console screen modes and fonts.
/etc/rcS.d/S55bootmisc.sh: line 50: savelog: command not found
[ 14.272718] etch0: Link is Up 100 Mbps Full Duplex, Flow Control: RX/TX
[ 14.272779] eth0: 10/100 speed: disabling TSO
I restart the install process and go with a standard, guided layout everything works fine.
Initially my thought was to setup root and boot in static parts, and everything else in LVM. However I was having trouble with that approach and (for the first time in my apache work) I am on a deadline. So I canned that project and came up with the scheme I show you above.
Ideally, if I had my choice, I would like to go with LVM for just about everything. I like the idea of having flexibility of altering part sizes based on production demands as they become apparent. And the ability to establish quotas and notifications in an LVM scheme is very appealing indeed. Which reminds me, can quotas be setup on static parts?
So when I (out of apparent necessity) went with a standard guided scheme I chose the default LVM layout. Here is how my parts are currently laid out:
Code:
/ 328 mb
/lib/init/rw 1.7 gb
/dev 10 mb
/dev/shm 1.7 gb
/boot 229 mb -ro
/home 222 gb
/tmp 376 mb
/usr 4.7 gb
/var 2.8 gb
all parts xfs
Ok. Now I must admit that while my apache skills were enough to get me the job, I am a novice at much of this. Namely, I never went too wild in my partitioning schemes in my little toy web servers so I don't know for instance why anyone would ever need a separate part for say
or
.
What's so magical about those places that they might need their own partitions? It seems to me that if I am going to start running with the big dogs and not just mucking about with toy websites every once in a while I better RTFM on this stuff and quick! If I knew where to look for the manual on this particular topic, that is! ;-)
So ultimately, and finally, my question is this. WHY on earth did the original partitioning scheme that me and my friend made not work?!?!?!? And secondly, let's say for the sake of argument that our original idea was so flawed that we have to abandon it and go with something different. Any harm in going with the default LVM layout on a fairly busy web server (supporting say 1,000 or so people a month to start)?
Also, for my next trick I am to build a mysql server and a DNS server. How would you approach partitioning differ for these types of servers (assuming a 250 SCSI drive for each).
Thanks! I always get the best advice from you guys and I'm looking forward to hearing your thoughts!
this partition scheme crashes servers. why???
- this would be my guess at the problem.
