|
Overview
|
|
|
|
Common Questions
|
|
|
|
Adobe Acrobat
|
|
|
|
Cisco
|
|
|
|
SGI Scanner
|
|
|
|
Packet Video Conf
|
|
|
|
Small Video Conf
|
|
|
|
New Users
|
|
|
|
Lic. Software
|
|
|
|
Public Domain
|
|
|
|
X-term & CDE
|
|
|
|
AS400
|
|
|
|
Digital Unix
|
|
|
|
Solaris
|
|
|
|
RedHat Linux(Alpha)
|
|
|
|
Silicon Graphics
|
|
|
|
Next
|
|
|
|
Misc Procedures
|
|
|
|
Windows 95, NT
|
|
|
|
Dialup
|
|
|
|
Linux
|
|
|
Setting up Redhat 6.1 from scratch
-
Grab the iso distribution from somewhere. Ga. Tech, North Carolina and sometimes local
cise or circa has these.
-
The Intel compaq servers boot from the cdrom by default.
Don't need to make the floppies.
If you do need to make a floppy get the rawrite and
boot.img for CD based boot or network
boot and use the DOS window to create floppies.
-
Boot it up and select language, keyboard and mouse
type and then the install presents several
options. Server option is not recommended for dual boot configurations as it overwrites
any dos partitions you may have setup before. We select custom
and make our own file systems with a large swap space using
the GUI tool.
-
Allocate at least /, /usr, /var and swap partitions at least. Usually we have
a tmp file system per machine as well on a seperate disk.
Setting up Redhat 5.1 from scratch
-
If the machine is already in SRM console do the firmware upgrade using
bootp just like any other digital unix boxes. ewa0_protocols bootp
and boot ewa0 and update. If you have to switch to the AlphaBios Console, you
need to get the correct zip file from data.microway.com under pub lx164 names
and unzip them. There are two floppies. One switchs from AlphaBios to
Older SRM after which we need to do the bootp to upgrade. The other
switches from SRM to AlphaBios after which you can upgrade using the a floppy.
For some reason bootp doesn't work under AlphaBios console.
Before any attempts to do bootp make sure the CISCO switch port that this
is connected to is fixed at full duplex 100 Mbps. The alpha consoles
either can not or will not do autonegotiate at that prompt level.
-
Copy the latest redhat alpha distribution from sunsite.unc.edu. You need
to use get alpha.tar.gz which combines the directories into one file and
keeps the protections and the links intact. This seems necessary to do the
nfs based installaion later on. Unpack this into /cleo/tmp1/chandra/redhat
directory. It is a good idea to look in the Updates directory also at the
same time and get the packages that are available and unpack them over the
ones that are stock distribution. This will help in not having to patches
after the install.
-
Get three empty floppies.
The first floppy needs to have the MILO boot
disk that works with the SRM console. This is in the form milo.dd at
ftp://ftp.digital.com/pub/DEC/Linux-Alpha/Miniloader/ and write it to
the floppy disk using rawrite.exe (on a DOS PC) or dd if=milo.dd of=/dev/fd0
bs=1K on a linux box. The rawrite is available under dosutils directory
of the redhat distribution.
The second floppy needs to be ext2 formatted on a linux box. To do this,
fdformat /dev/fd0H1440
mount /dev/fd0 /mnt
/sbin/mk2efs /mnt
From the kernel subdirectory of the redhat distribution
cp lx164.gz /mnt/vmlinux.gz
umount /mnt
1:12345:respawn.....
2:12345:respawn:/sbin/mingetty ttyS0
You can comment out the other 2,3,4,5,6 mingetty entries as we don't
have any terminals directly attached to these machines. You have
to kill and restart init which is usally the PID 1 or do a reboot.
You should now see a login prompt on the console.
At the MILO prompt you need add the
boot sda2:vmlinux.gz root=/dev/sda2 load_ramdisk=1 console=/dev/tty0 or console=serial
and it seems to help. We have not tried without the console=/dev/tty0 but
we think it should work.
- We patched up the 2.0.34 kernel with the serial console patch and
rebuit the aboot version of the vmlinux.bootp. It seems to be able to
get to aboot and get to the scsi bus but can not initialize any drives.
We need to look for patches to 2.0.34 on the alphas to support the Qlogic
bus correctly. It appears such stuff exists on the DEC site where
the kernels we got from work on the controllers.
All the sources for kernels, aboot and the downloads chandra did are
non in /usr/src directory of helix14. Please put all the things
you are getting over there so that we can one repository.
Setting up RedHat 5.0 on an Alpha PC164LX
- Currently we are buying our machines from
Microway so the low level setup
of the firmware is currently left out. We do however have specific
requirements for the Microway setup. We require the SRM console to be
used on machines which will not have graphics heads (graphics card, monitor
and keyboard). The ARC console does not allow booting without these things,
and for compute servers there is no desire to have them.
- Disk partitioning is done with the tool "fdisk." Currently fdisk is
limited to disks with 63 or fewer sectors per track. The 18 gig Seagate
drives that we now buy have 237 sectors per track. Microway suggested that
I try an old Intel fdisk trick to get around the limit of 63 sectors/track
(our disks have 237 sectors / track). I did the math and came up with:
Here are the physical parameters of our disks:
512 bytes/sector
237 sectors/track
20 tracks/cylinder
7501 cylinders
Here is what we tell fdisk:
512 bytes/sector
63 sectors/track
255 tracks/cylinder
2213 cylinders
Here is the console output from an fdisk session:
login as root
/sbin/fdisk /dev/sdb
Command (m for help): c
DOS Compatibility flag is not set
Command (m for help): d
Partition number (1-4): 1
Command (m for help): d
Partition number (1-4): 2
Command (m for help): d
Partition number (1-4): 3
Command (m for help): d
Partition number (1-4): 4
Command (m for help): x
Expert command (m for help): c
Number of cylinders (1-[1023]-65535): 2213
The number of cylinders for this disk is set to 2213.
This is larger than 1024, and may cause problems with:
1) software that runs at boot time (e.g., LILO)
2) booting and partitioning software form other OSs
(e.g., DOS FDISK, OS/2 FDISK)
Expert command (m for help): r
Command (m for help): n
Command action
e extended
p primary partition (1-4)
p
Partition number (1-4): 1
First cylinder (1-2213): 1
Last cylinder or +size or +sizeM or +sizeK ([1]-2213): 2213
Command (m for help): w
The partition table has been altered!
Calling ioctl() to re-read partition table.
Syncing disks.
WARNING: If you have created or modified any DOS 6.x
partitions, please see the fdisk manual page for additional
information.
And the command to layout the disk format is:
/sbin/mkfs -m 1 /dev/sdb1
("m 1" reserves 1% of the disk space for root's use)
- In order to allow root logins over the net, rename the file
/etc/securetty file.
- Most of the system setup is done using a GUI tool named "control-panel."
This requires X to be used. If you are not using X then cd to the
/etc/sysconfig directory and edit the files by hand. A good reference for
the /etc/sysconfig directory is here
- To get on the network you will need the file /etc/sysconfig/network
and /etc/sysconfig/network-scrips/ifcfg-eth0. Here is network from a
typical machine:
NETWORKING=yes
FORWARD_IPV4=no
HOSTNAME=helix01
DOMAINNAME=phys.ufl.edu
GATEWAY=128.227.89.1
GATEWAYDEV=eth0
NISDOMAIN=phys.ufl.edu
And ifcfg-eth0:
DEVICE=eth0
IPADDR=128.227.89.226
NETMASK=255.255.255.0
NETWORK=128.227.89.0
BROADCAST=128.227.89.255
ONBOOT=yes
BOOTPROTO=none
USERCTL=no
ypserver ligo
** Note: the NIS server must exist in the /etc/hosts file **
To create the links from /etc/rc.d/rc? to start ypbind the tool "chkconfig"
(same as on SGI Irix machines) is used. Chkconfig gets its information from
the /etc/rc.d/init.d/ypbind script. The correct command to start ypbind at
boot is"chkconfig --level 5 ypbind on" is used, where 5 is
the runlevel to turn this on at. I also turned it on at
runlevel 3 and 4 with similar commands.
Check to see if the directory /var/yp exists. If not create it as ypbind
requires it to start.
You can now start ypbind with the command "/etc/rc.d/init.d/ypbind start"
and then test it with "ypwhich."
- Get network time from one of our servers:
Add the line "/usr/bin/rdate -s ligo" to the end of the file
/etc/rc.d/rc.local
- Use our nameservers for DNS lookups:
/etc/resolv.conf should read:
search phys.ufl.edu
nameserver 128.227.64.2
nameserver 128.227.64.70
- Add the hostname for the local machine, the nameservers, and the NIS
server to the file /etc/hosts. Ideally the name and number of the new
machine should be added to the /etc/hosts file on phys.ufl.edu and then
rdisted over, but this can be done later.
/etc/hosts:
# /etc/hosts file for Physics Department
# For information contact
# Last Edited 11/4/97
#
127.0.0.1 localhost loghost
# Campus name servers
128.227.128.24 name.ufl.edu
192.80.214.100 noc.sura.net
128.227.148.252 cutter.clas.ufl.edu
128.227.16.3 orange.qtp.ufl.edu orange
128.227.192.1 crunch.qtp.ufl.edu crunch
128.227.64.7 neptune-gw.phys.ufl.edu neptune-gw
128.227.24.1 neptune.phys.ufl.edu neptune
128.227.89.55 ligo.phys.ufl.edu ligo
128.227.89.??? newmachine.phys.ufl.edu newmachine
- Edit /etc/fstab and add the NFS filesystems.
Add the user disks and one or more temporary disks and the mail spool area
from the mail server.
/dev/sda2 / ext2 defaults 1 1
# /dev/fd0 /mnt/floppy ext2 noauto 0 0
/dev/sda1 none ignore 0 0 0
/dev/sdb1 /helix01/tmp1 ext2 defaults 0 0
none /proc proc defaults
/dev/sda3 none swap sw
/dev/sda4 /usr ext2 defaults 1 2
#/dev/scd0 /cdrom iso9660 ro 0 0
# NFS mounts:
hee:/heeusers /heeusers nfs rw,bg
hee:/hee/cleo /hee/cleo nfs rw,bg
helix02:/helix02/tmp1 /helix02/tmp1 nfs rw,bg
helix03:/helix03/tmp1 /helix03/tmp1 nfs rw,bg
- Create the directories and mount the filesystems using mount
-a. If this a brand new machine being added to the network,
we will need to add it into a netgroup in the /etc/yp/netgroup
file on the NIS master server and propagate that to the entire network.
It may also be necessary to explicitely export filesystems for the
client to be able to mount them. This is done by issuing
/etc/rc.d/init.d/nfs restart command.
- Make a link using ln -s /local /usr/local if the machine
is going to have its own local. Populate the /usr/local filesystem
from another machine of the same OS and architecture using rdist. You will
need to edit /root/.rhosts file to allow root logins from the
other machine through rsh without password. You can copy it from some other
client. Remember to do chmod 400 /.rhosts after the file is created.
You will then need to login to the other machine as root in
/local/adm/system/rdist directory and use the command
rdist -n -o remove -p /usr/sbin/rdistd -f /path/to/Distfile.usr.local
Check the output of what will be updated.
If it is alright then use
rdist -o remove -p /usr/sbin/rdistd -f /path/to/Distfile.usr.local
The Distfile.usr.local would contain:
/usr/local -> newmachine.phys.ufl.edu
You could also include all of the other files that we have been modifying by
hand, here is the Distfile used to keep all of the High Energy Experiments
Helix cluter in sync:
# We setup helix01 to they way we like and distribute stuff from there
# to all the other helix machines. You need to have /.rhosts
# fixed up on the clients first with helix01 root allowed.
# Also remember to do chmod 400 ~root/.rhosts on the clients
#
#
HELIX = ( helix02 helix03 helix04 helix05 helix06 helix07 helix08 helix09 helix10
helix11 helix12 helix13 helix14 )
ALL = ( ${HELIX} )
#
all: /etc/hosts.equiv -> ( ${ALL} )
all: /etc/group -> ( ${ALL} )
all: /etc/yp.conf -> ( ${ALL} )
all: /etc/resolv.conf -> ( ${ALL} )
all: /etc/shells -> ( ${ALL} )
#all: /etc/securetty -> ( ${ALL} )
all: /etc/printcap -> ( ${ALL} )
#
# When sending out CDE areas watchout for the Xaccess file which
# contains the X-terminals allowed to access the server and
# Xconfig file which determines whether local graphics console
# is present (neutron, proton).
#
all: /usr/lib/X11/app-defaults -> ( ${ALL} )
all: /usr/lib/X11/fonts -> ( ${ALL} )
#
all: /root/.bashrc -> ( ${ALL} )
all: /root/.cshrc -> ( ${ALL} )
all: /root/.login -> ( ${ALL} )
all: /root/.profile -> ( ${ALL} )
all: /root/.rhosts -> ( ${ALL} )
#
all: /usr/local -> ( ${ALL} )
After this is done, remember to add the new client to the Distfile.weekly
script so that /usr/local area gets updated automatically.
- Copy /etc/shells from some other client or add our local
shells to the system shells one per line. This will let our users ftp into
that machine.
Setting up amd
-
It appears that there is a bug in the ypbind on the alphas that
kills the ypbind daemon under redhat linux 5.0 stock distribution. So
the following packages that are related were installed on all helix
machines using rpm. The command to do this is rpm -Uvh package-name
. The only special one is portmap which needs the --noscript
option as it screws up the system area links on the alphas. On the intel
PCs you need to download the .i386.rpm versions of am-utils and may not
need all the ypbind stuff at all if you are not going to be a full
NIS client.
am-utils-6.0a16-1.alpha.rpm
glibc-2.0.7-13.alpha.rpm
glibc-debug-2.0.7-13.alpha.rpm
glibc-devel-2.0.7-13.alpha.rpm
glibc-profile-2.0.7-13.alpha.rpm
nfs-server-2.2beta29-5.alpha.rpm
portmap-4.0-8.alpha.rpm
ypbind-3.3-3.alpha.rpm
- After the am-utils-XXX is installed, create a file
/etc/amd.conf that looks like the following
# Global Options for amd automounter
[global]
# Where the actual mounts occur. This will mimic Digital Unix and SUN Solaris
auto_dir = /tmp_mnt
# Log messages to /var/log/messages through syslog facility
log_file = syslog
# Just enable file maps first.
# Saves time on lookups as it doesn't have to hunt all available maps
map_type = file
# Make the filesystems mounted statically known to amd
restart_mounts = yes
# Supposed to give a listing of available directories for ls (again mimics DU and Solaris)
browsable_dirs = yes
# Directory where we will keep the maps
search_path = /etc/amdmaps
# Extend the umount time to 30 minutes of inactivity
dismount_interval = 1800
# Unmount all filesystems
unmount_on_exit = yes
# One map per machine
[ /maxwell ]
map_name = maxwell
[ /charm01 ]
map_name = charm01
[ /charm02 ]
map_name = charm02
[ /charm03 ]
map_name = charm03
[ /charm04 ]
map_name = charm04
[ /charm05 ]
map_name = charm05
[ /charm06 ]
map_name = charm06
[ /charm07 ]
map_name = charm07
[ /charm08 ]
map_name = charm08
[ /charm09 ]
map_name = charm09
[ /charm10 ]
map_name = charm10
[ /helix02 ]
map_name = helix02
[ /helix03 ]
map_name = helix03
[ /helix04 ]
map_name = helix04
[ /helix05 ]
map_name = helix05
[ /helix06 ]
map_name = helix06
[ /helix07 ]
map_name = helix07
[ /helix08 ]
map_name = helix08
[ /helix09 ]
map_name = helix09
[ /helix10 ]
map_name = helix10
[ /helix11 ]
map_name = helix11
[ /helix12 ]
map_name = helix12
[ /helix13 ]
map_name = helix13
[ /helix14 ]
map_name = helix14
[ /heetape ]
map_name = heetape
The above file is almost identical on all our machines except for
two lines which are missing depending on which machine you are on. For
example this is a file that I copied from helix01 and so has the two lines
for that machine missing. It is important to omit maps for the local file
systems as otherwise amd will umount them from the regular place and
mounts them in the /tmp_mnt area creating problems with exports.
-
Then create a directory called /etc/amdmaps and create files with
the mapnames mentioned above in that directory. For example the
charm05 map looks like the following: These files are very similar to
each other except for the hostname changes. Someday we will figure out
the amdmaps macros fully to automate this to a single file.
/defaults type:=nfs;opts:=nosuid,nodev,rw,grpid
* host!=charm05;fs:=${autodir}${path};rhost:=charm05;rfs:=${path}
-
You then edit /etc/sysconfigtab/amd file to look like
AMDOPTS='-r -w 1800'. The 1800 here is a repitition to the
parameter specified the configuration file but it appears that amd will
umount in 2 minutes without this explicit startup parameter.
- You then edit the /etc/rc.d/init.d/amd file to look like
this. This involves taking out a couple of lines that are there, and
changing a couple of flags and adding a couple of lines to umount and mount
filesystems when amd is stopped and restarted.
The file can be copied from any of our helix machines and looks like this
#!/bin/sh
#
# chkconfig: 345 72 28
# description: Runs the automount daemon that mounts devices and NFS hosts \
# on demand.
# processname: amd
# config: /etc/amd.conf
# Modified by Chandra to eliminate the /etc/sysconfig/amd reading
# and increase the timeouts to 30 minutes
#
# Source function library.
. /etc/rc.d/init.d/functions
# See how we were called.
case "$1" in
start)
echo -n "Starting amd: "
touch /var/lock/subsys/amd
daemon /usr/sbin/amd -r -w 1800 -c 1000 -F /etc/amd.conf
echo
;;
stop)
echo -n "Umounting all filesystems: "
umount -a
echo -n "Mounting filesystems in /etc/fstab"
mount -a
echo -n "Shutting down amd: "
killproc amd
rm -f /var/lock/subsys/amd
echo amd
;;
status)
status amd
;;
restart|reload)
killall -HUP amd
;;
*)
echo "Usage: amd {start|stop|status|restart}"
exit 1
esac
exit 0
The umount and mount were added after we discovered that the mountpoints
did not go away after just stopping amd preventing it from further mounting
filesystem when restarted. This behavious is not seen under Solaris or DU.
Only on linux.
-
You then make sure you machine is the allowed list of machines to mount
the remote filesystems you are putting in your maps. Typically this involves
editing the /etc/exports on the server machine or the
/etc/yp/netgroup on neptune (our Yellow Pages master) and propagating
the information to all the clients. Contact Dan or Chandra if this
step is needed.
|
