• The OS kernel provides interfaces to the following
  • Processes (time-sharing, protected address space)
  • Signals (to/from processes)
  • Interprocess communication (pipes and network connections)
  • Virtual memory (swapping, paging, mapping)
  • Filesystems

• The kernel contains device drivers and the rest is mainly device independent
  • Some kernels, such as proprietary UNIX kernels (Solaris, HP-UX, AIX, Mac OS X, etc.) usually run on hardware that is taylored to the OS, e.g. Solaris on Sun workstations/servers, Mac OS X on Macs.
    • Less portability but greater stability can be expected as a result.
  • Other kernels (Linux kernel, *BSD kernels, Windows) must deal with a wide variety of hardware, especially on Intel-compatible x86 platforms.
    • Greater portability but slighly less stability can be expected as a result.

• Most systems come with generic kernel and kernel configurations that allows an OS to run on most supported hardware.
• What to remove unneeded features and device support
  • not needed with modular kernels (such as that which accompany most Linux distributions)
• Want to reconfigure to make kernel memory footprint smaller
  • Again, modular kernels make this unnecessary
• Want to add third party or experimental device drivers
• Want to add certain features
  • some intrusion detection and resistance systems
  • clustering
  • experimental file systems

• Solaris - /kernel/unix
• HP-UX - /stand/vmunix
• Linux - /boot/vmlinuz-*
• FreeBSD - /kernel
• Windows - ntoskrnl.exe

• Most operating systems provide some way to alter some behaviors of the currently running kernel.

• Example: On linux, kernel behavior can be affected through the /proc file system if the kernel supports it.
  • The files in /proc aren't really files but “backdoors” into the kernel
  • One way to change the maximum number of open files a process can have on Linux:

                # echo 179894 >> /proc/sys/fs/file-max

• Example: On linux, the sysctl utility can be used to configure a running kernel's parameters.
  • Run # sysctl -a to see the kernel parameters that can be altered.
  • Also see /etc/sysctl.conf

• Link to article
• Allows for adding & removing features or device support to a kernel core on an as-needed basis
• Finding currently installed kernel modules
  • Solaris - modinfo
  • Linux - /sbin/lsmod
    • Try 'lsmod | grep fs' on your nfs servers.

• Solaris - add_drv/rem_drv, modload/modunload
• Linux - insmod/rmmod, modprobe (like insmod but also handles module dependencies)

• Kernels usually log messages to the system log
• Example: Linux klogd (Kernel Log Daemon) logs to /var/log/syslog
  • Also, the dmesg command gets info from “kernel ring buffer”
    • Can be useful to run dmesg to verify kernel module loading & unloading

• Usually only concerned about performance tuning if the system or systems seem unusually slow or on really busy web sites whose web servers must handle many web transactions.

• The hardware may be too slow, antiquated.
  • You may have insufficient RAM to support a large number of simultaneous users.
  • You may have an insufficient number of CPUs to support a large number of simultaneous users.
  • You may have slow, highly fragmented or failing hard drives.

• A user or users may be running a very resource-intensive program that puts a large load on the system.
  • May need to set ulimits (man ulimits and /etc/security/limits.conf).
  • Or write cron jobs to find resource hogs and renice them down.

• A file system may be slow.
  • May need to tune FS configuration
    • Most file systems allow turning off access time stamp (atime) updates which can speed up file & directory access.
  • May need to choose a different FS.
    • Some file systems are better suited to accessing very large files, others are better suited to accessing many small files.

• A networked file system (NFS, Samba or other) may seem unusually slow.
  • If NFS or Samba are much slower than ftp, may need to tune configuration.
    • Samba config example: enable “opportunistic locking,” disable “strict locking”

• A network or network applications are slow.
  • Prime suspect: DNS (domain name system)
    • A network name server may have died.
      • rockhopper:/etc/resolv.conf

• Some performance tips for UNIX / Linux that I have seen:
  • Try to put the swap partition as close to the start of the disk as possible (lower cylinder and sector numbers).
  • Try to make sure that the swap partition disk is on a different disk controller than the data disk.
    • or stripe swap across multiple disks for better write performance.
  • Compile the kernel, services and programs with optimizations that take advantage of > i386 CPUs and their specialized instruction sets, e.g. Pentium*, Athlon*.

• Benchmarking
  • Can be ad hoc such as applying the time command to a long process, such as a kernel compile
  • Various benchmarking programs exist
    • Some we can try in lab:
      • Disk I/O - iozone (http://www.iozone.org), iostat
        • Can be used to benchmark local disk or NFS server performance
      • Network - NetPIPE (http://www.scl.ameslab.gov/netpipe), tbench
      • System - nbench (http://www.tux.org/~mayer/linux/bmark.html ), geekbench (http://www.primatelabs.ca/geekbench)

• Steps to set up a local printer on a UNIX system
  • Physically connect the printer to the computer (usually through the parallel port or USB)
  • Configure the proper files on the system to communicate correctly with the proper port
    • kernel needs support for parallel port
  • Create a spooling directory and accounting files for this print queue
  • Set up proper filter programs (if necessary)
  • Configure the printer system to allow printing to your new device
  • Enable printing on the new device
  • Send a print job to test the printer
  • Most of the above steps might be automated using print config utilities
    • printconf on RedHat and derivatives
    • admintool and others on Solaris

• BSD printing
  • The lpr system uses a single printer capabilities file to control printing
    • Which device, what filter to use, baud rates, form feeds, remote host and remote device name, etc
    • The lpr system can work over a network
  • lpr - queues print jobs
  • lprm - removes print jobs form queue
  • lpq - shows jobs in print queue
  • lpc - printer control program
  • lpd - printer daemon
  • /etc/printcap - printer configuration file describing all printers on this host (and possibly other remote hosts)
  • BSD printcap file
    • See /etc/printcap on rockhopper

• System V printing
  • The lp system uses scripts that control the printer devices. (System V.4 adds remote printing and Postscript capabilities)
    • lp - initiates print requests
    • cancel - cancels pending print requests
    • lpstat - to list queue contents and show status
    • accept - allows lp to accept requests
    • reject - prevents lp from accepting requests
    • enable - activates named printer
    • disable - disables named printer
    • lpadmin - configures lp spooling system
    • lpsched - line printer scheduling daemon
    • /usr/spool/lp/model - directory where model files are kept

• Postscript in UNIX printing
  • Most UNIX utilities and applications that deal with output to a printer will output Postscript.
  • See some sample Postscript output with gv or ghostview

• CUPS (Common UNIX Printing System)
  • Dominant printing system on Linux now
  • Also used in Mac OS X
    • CUPS is developed and owned by Apple.

• Download, configure, compile custom Linux kernel in the “Debian” way on your Linux VM.
  • Do this on only one of your VMs.
  • Requires the build-essential, kernel-package and libncurses5-dev packages.
  • Download a kernel-source archive from the instructor-supplied URL.
  • Extract kernel source from the tar.bz2 archive in /usr/src:

       # tar xjf linux-source*.tar.bz2

• Apply existing kernel configuration from /boot:

       # cd /usr/src/linux-source*
       # cp /boot/config-3.2.something* .
       # cp config-3.2.something* .config

• Do minimal configuration of kernel using make menuconfig in the kernel source directory:
  • Change processor (CPU) type
  • Make sure there is kernel support for the following:
    • file system support for: ext2, ext3
    • device driver support for:
      • All related to SATA disk drives
        • See output of lspci -v
  • Save new kernel config as vanilla471

• Compile kernel using make-kpkg:

       # nohup make-kpkg --revision=vanilla471 --initrd kernel_image &

• Install new kernel linux-image kernel package found in /usr/src:

       # dpkg -i /usr/src/linux-image*.deb

• Check GRUB boot loader config (/boot/grub/grub.cfg) to ensure that the new kernel will be booted by default.

• Reboot the VM.

• Set up remote printing to hh310
  • We will set up networking printing to the HP Laserjet 4200 in HH 310 using CUPS, the Common UNIX Printing System, which is now a de facto standard for Linux printing.
  • Ssh to your VM as root using ssh -X (X11 forwarding).
  • Install iceweasel (Debian equivalent of firefox) and hpijs-ppds packages
  • Run iceweasel remotely on your VM.
  • In iceweasel, visit the CUPS administration web interface at the following URL and add a printer with queue name “hh310”:

          http://localhost:631
    
  * The CUPS printer URL for the printer in HH 310 is

          socket://10.8.0.119:9100

           where :9100 is the HP Jetdirect service port for the
           Jetdirect network card in the printer.

• Choose the correct printer model (or as close as possible), and choose the Postscript driver for it.
• Make “hh310” the default printer.
• Send a test page to the printer.
• Try printing something with lpr and a HERE document.

           # lpr << END
           This is a test from root@carson.cs.monmouth.edu.
           END

• Try printing something with enscript (may need to install enscript first) and a HERE document.

           # enscript << END
           This is a test from root@carson.cs.monmouth.edu.
           END

• These are NOT mandatory.

• Start by installing either the bittorrent or bittornado package.
• Try to write a script called btcp that automates the process of:
  • setting up a bittorrent tracker on the source machine
  • creating and uploading a torrent file from the source machine to the target machines
  • seeding the torrent on the source machine
  • downloading the torrent on the target machines

• Install, configure and test spamassassin.
  • Start by looking at /etc/default/spamassassin.

• Try blocking and unblocking outgoing port 80 (www) from your VM.
• Try blocking and unblocking incoming ssh connections from a specific IP address.