BCCD:PowerPC

From Earlham CS Department
Revision as of 19:53, 6 October 2005 by Tmcnulty (talk | contribs) (→‎Setting Up Your Build Environment)
Jump to navigation Jump to search

The BCCD, or Bootable Cluster CD, is a diskless cluster solution "created to facilitate instruction of parallel computing aspects and paradigms." More information about the BCCD can be obtained from bccd.cs.uni.edu, as well as from the BCCD's main wiki, which this page is meant to complement. This page serves generally as an ongoing summation of Tobias McNulty's work on the BCCD for PowerPC.

Building the BCCD on PowerPC

See also: BCCD weblog

Setting Up Your Build Environment

Base System: Debian GNU/Linux testing with development tools. The Debian installer can be netbooted on most NewWorld PowerMacs. See the PowerPC Kernel Archives' Mac netboot page for instructions.

To setup your Debian environment, download bccd-dpkg-selections and run dpkg --set-selections < bccd-dpkg-selections then apt-get dselect-upgrade: 

su -
/* enter root pass */
wget http://hermes.osuosl.org/~tmcnulty/bccd-dpkg-selections
dpkg --set-selections < bccd-dpkg-selections
apt-get dselect-upgrade

Building the BBC-PPC ISO

/* Check out the cvs tree anonymously */
cvs -d:pserver:anonymous@bccd.cs.uni.edu:/ login
/* hit return when prompted for a password */
cvs -d:pserver:anonymous@bccd.cs.uni.edu:/ co bccd

/* OR: Check out the cvs tree with a real user account */
export CVS_RSH=ssh
export CVSROOT=username@bccd.cs.uni.edu:/var/lib/cvs
cvs co bccd

/* execute the build */
export singularity_GARCH=ppc
cd bccd/meta/lnx-bbc
make build
/* wait a long, long, long time */

Building a Package and Adding it to the Repository

The BCCD has a set of downloadable packages, available via the list-packages command (as root) on a running copy of the BCCD. The instructions that follow explain the process of adding a package to the repository that list-packages uses. The BCCD package system simply unpacks the specified tarball in /usr/local, so a "package" can contain anything from a set of binaries to a collection of documentation files in text, html, or some other format (e.g., a curriculum module).

On a running BCCD

These instructions assume that you start as root in ~, on a running copy of BCCD/PowerPC.

  • Build the source:
tar xzf somesource-*.tar.gz
cd somesource-*
./configure --prefix=/usr/local
make
  • Install the binaries in /usr/local, but archive its current contents first so they don't get packaged up too:
mkdir ~/local
mv /usr/local/* ~/local
make install
  • Package up /usr/local, clean up, and move the old contents of /usr/local back where they belong:
cd /usr/local
tar czf ~/somepkg.tar.gz .
rm -rf /usr/local/*
mv ~/local/* /usr/local
cd

gpg --detach somepkg.tar.gz
gpg --verify somepkg.tar.gz.sig # does this do what I think it does?
  • Create the package directory on bccd.cs.uni.edu, upload the package and signature, and add the package to packages.txt:
ssh bccd.cs.uni.edu mkdir /var/www/packages/ppc/2.2/somepkg
scp somesource.tar.gz* bccd.cs.uni.edu:/var/www/packages/ppc/2.2/somepkg
ssh bccd.cs.uni.edu echo "somepkg	some package description" >> /var/www/packages/ppc/2.2/packages.txt

(notice the tab between the package name and description on the last line)

Example: Building GROMACS

  • Setup your environment:
export LDFLAGS="-L/lib/gcc-lib/powerpc-lnxbbc-linux -L/usr/local/lib -L/mpich/lib"
export CPPFLAGS=-I/usr/local/include
export LD_LIBRARY_PATH=/lib/gcc-lib/powerpc-lnxbbc-linux/:/mpich/lib
export PATH=$PATH:/mpich/bin
  • Fix BCCD lib bug (necessary as of 6/20/2005):
mkdir -p /tmp/snapshot.build/staging/singularity/image
ln -s /lib /tmp/snapshot.build/staging/singularity/image/lib
  • Install necessary headers:
list-packages #select fixpackages and hit OK
  • Build FFTW and GROMACS:
cd fftw-* # 2.1.5 works
./configure --enable-float --enable-type-prefix --enable-mpi --prefix=/usr/local
make -j2
make install
make distclean
./configure --enable-type-prefix --enable-mpi --prefix=/usr/local
make -j2
make install
cd ../gromacs-* # 3.2.1 works
make distclean
./configure --enable-mpi --enable-shared --prefix=/usr/local --exec-prefix=/usr/local
make -j2
make install

Now you should have a working copy of GROMACS in /usr/local. Enjoy!

Generating a GPG Key to Sign Packages

Generating the Key

Quick Guide. For the far more detailed instructions from which these commands have been extracted, see How to Generate a GPG key. 

gpg --gen-key
gpg --fingerprint
gpg --export --armor user@host.org > key.gpg.asc
gpg --keyserver wwwkeys.us.pgp.net --send-keys user@host.org
gpg --gen-revoke user@host.org > revoke.gpg.asc
chmod 400 revoke.gpg.asc

When finished, key.gpg.asc is your public key (if you're going to sign packages with this key in the distribution BCCD, add it to the image as a trusted key), and revoke.gpg.asc is your revocation certificate, should you need to revoke this key pair.

Making the Key Authoritative

To make your new key authoritative in the BCCD image, you need to append it to bccd/archive/garpkg/files/gar-keys.asc.

  • First, grab a copy of your key fingerprint:
tmcnulty@bccd:~$ gpg --fingerprint
/home/tmcnulty/.gnupg/pubring.gpg
---------------------------------
pub  1024R/825ADB6E 2005-06-14 Tobias McNulty (BCCD) <tmcnulty@ppckernel.org>
     Key fingerprint = 47BB 9126 6FE0 866D 11C0  AB3E 499D 4547 825A DB6E
  • Next, append the line starting with "pub" to gar-keys.asc:
tmcnulty@c15:~/bccd/archive/garpkg/files$ cat >> gar-keys.asc

pub  1024R/825ADB6E 2005-06-14 Tobias McNulty (BCCD) <tmcnulty@ppckernel.org>

<control-d>
  • Now, append gar-keys.asc with the contents of key.gpg.asc:
tmcnulty@c15:~/bccd/archive/garpkg/files$ cat ~/key.gpg.asc >> gar-keys.asc
  • You'll need to update the checksum for gar-keys.asc:
tmcnulty@c15:~/bccd/archive/garpkg/files$ md5sum gar-keys.asc >> ../checksums
tmcnulty@c15:~/bccd/archive/garpkg/files$ vim ../checksums #remove the old checksum line

Now, rebuild the ISO, upload some signed packages to the repository, and give list-packages a try!

Regression Testing

Executing the Tests

To execute the regression tests included with bccdrt on a running copy of the BCCD, simply checkout the bccdrt module from CVS on bccd.cs.uni.edu and start runtests.py: 

export CVSROOT=username@bccd.cs.uni.edu:/var/lib/cvs/var/lib/cvs # you aren't seeing double and this isn't a typo!
cvs co bccdrt
cd bccdrt
./runtests.py # run list-packages and install python first if need be

System Overview

This section exposes the architecture behind bccdrt, which refers generally to the collection of Python and PHP scripts than handle the execution and data extraction (bccdrt.py), recording (post-results.php), and analysis (show-results.php) of BCCD regression tests (the scripts were written and tested on PPC, but can be used on any architecture the BCCD supports).

htdocs

This directory houses the PHP scripts that record and analyze test data.

rt.d

This directory houses individual test scripts and the bccdrt library, which provides several routines to the individual tests (see bccdrt.py below).

runtests.py

This script calls some generic bccd cluster initialization routines (bccd-allowall, bccd-snarfhosts, and bccd-checkem machines) and then executes each of the tests in the rt.d directory. No per-test modification of runtests.py is necessary.

bccdrt.py

This library provides several routines to individal tests, including system(...), runcmd(...), and postresult(...). It resides in the rt.d directory, but contains no actual test code itself.

  • system(...) takes a single argument in the form of a shell command, executes that command, and allows the output to print to the console.
  • runcmd(...) does the same thing, but returns the output (both stdout and stderr) as a string.
  • postresult(...) takes two arguments: the test name and an associative array of your test data (column) names and values. postresult(...) adds the generic test data, like machine name, BCCD release, etc., encodes the data in XML, and posts it to post-results.php via HTTP.

To define a new test, write a new script in Python that executes whatever programs on the BCCD the test involves and then extracts the relevant data from them (e.g., with regular expressions). There is an sample test in rt.d named gromacs.py. The only call your test function must make is to postresult(...).

Example: Creating the associative array and calling postresult(...) 

import bccdrt

result={}

result["molecule"] = os.path.basename(moleculedir)
result["np"] = np
result["walltime"] = Real
result["psNODEhr"] = psNODEhr

postresult("gromacs",result)

post-results.php

This script is generally responsible for recording data as posted by bccdrt.py. This involves the following:

  • receiving the XML-encoded data from bccdrt.py
  • determining the SQL table format from the given data names and values
  • creating the SQL table if necessary
  • inserting the given data into the table

What does this mean for you? post-results.php is entirely automatic; that is, it requires no per-test modification to operate.

show-results.php

This script handles the analysis and display of the regression test data that resides in the SQL tables created by post-results.php. Per regression test, it operates with the folllowing parameters: (a) a table name; (b) an array of check columns, i.e., columns that contain the actual test data that we're concerned with; and (c) an array of control columns, i.e., columns whose values determine the grouping of data within each test table, e.g., if one's control columns are 'processor' and 'node_count', then show-results.php will discover every possible combination (between columns) of distinct column cell values and use that set of combinations to group test data prior to analysis and display. Each grouping combination includes exactly one value from every control column. This is easier understood when seen than when described (due in a large part to my inability to describe the system at hand), so visit the actual script and see what it does for yourself!

So, to make show-results.php aware of your newly-defined regression test, simply create an associative array containing your table name (tbname), array of check columns (check_cols), and array of control columns (control_cols) and append it to the $tests array. The table name is the test name prefixed by rt_, and the column names are exactly the same as those passed to postresult(...) in the original Python.

Example: Defining a test in show-results.php 

$tests = array( array(  "tbname" => "rt_gromacs",
                        "check_cols" => array("walltime","psNODEhr"),
                        "control_cols" => array("machine","molecule","np") )
               );

show-results.php then iterates through $tests in the body of the HTML document, calling showResults(...) with the parameters defined in each element.

Recap: Creating a New Test

That's a lot of talk for a relatively simple process. Here's the basic outline once more of what creating a new regression test involves:

  • 1. write the actual test in Python and call postresult(...) to send the test data to the result server
  • 2. drop your test in the rt.d directory
  • 3. define your test's table name, check columns, and control columns in show-results.php
  • 4. commit your changes to CVS, update the CVS module on the result server, and execute the tests.

FAQ

How does show-results.php analyze test data?

The analysis script calculates the standard deviation of each grouping of each check column and from that determines, on a per-cell basis, outliers in the data set, i.e., possible test failures (1*S). Cells that pass the test are given the CSS class rt_pass, and cells that fail rt_fail. Currently, the stylesheet marks passes green and potential failures red.

TODO

  • Implement early failure detection and reporting (e.g., "this command doesn't even run")

Conclusion

That brings a close to our discussion of BCCD regression testing. Please direct any questions, comments, or suggestions to the system's author, Tobias McNulty.

Fossilizing the BCCD

This section outlines the steps required to disassemble a BCCD ISO, manifest it on a hard disk drive, and boot from that hard drive. Most or all of this must be done as root.

Mount the Images

The Basic Images

cd /mnt # or where ever
mkdir bccd
mount -t iso9660 -o loop bccd-ppc-2005-08-30T00-0500.iso bccd

# on PPC
mkdir initrd
gunzip < bccd/boot/root.bin > initrd.ext2
mount -t ext2 -o loop initrd.ext2 initrd

# on x86
mkdir lnx
mount -o loop bccd/lnx.img lnx
mkdir root
gunzip < lnx/root.bin > root.ext2
mount -o loop root.ext2 root

The singularity

First, decompress the singularity with the cloop utility extract_compressed_fs: 

wget http://developer.linuxtag.net/knoppix/sources/cloop_0.66-1.tar.gz
tar xzf cloop_0.66-1.tar.gz
cd cloop-0.66
vim Makefile # add APPSONLY=1 at the top
make zcode
make extract_compressed_fs
./extract_compressed_fs ../bccd/singularity > ../singularity.romfs
cd ..

The latest currently-available version of cloop (2.01) doesn't work for this purpose; others might (I didn't experiment), but 0.66 definitely does.

Next, mount the singularity (you must have romfs support compiled into the kernel): 

mkdir singularity
mount -t romfs -o loop singularity.romfs singularity

Extract the singularity

cd singularity
tar cf - . | (cd /path/to/destination/partition;tar xvf -)

Create a working initrd

Create an initrd for fossilized booting with the linuxrc at http://ppckernel.org/~tmcnulty/bccd/linuxrc: 

cd /mnt/root # or where ever you mounted root.ext2 (from root.bin)
wget http://ppckernel.org/~tmcnulty/bccd/linuxrc # replace the existing linuxrc
chmod a+x linuxrc
cd ..
umount root
gzip < root.ext2 > /path/to/destination/partition/boot/root.bin

Configure the bootloader

Configure your bootloader (e.g., yaboot, lilo, or grub) as follows:

  • boot the kernel /boot/vmlinux on PowerPC or /boot/bzImage on x86
  • use the initrd /boot/root.bin
  • execute the init script /linuxrc.

Here is a sample lilo.conf.

TODO

  • fix the mounting commands so that / is only mounted once
  • decide how to handle directories like /etc that are mounted in ram at /dev/rw/etc and populated with items from /etc.ro (leave as is, or create a script to simplify the setup for hard disk booting?)
  • modify init scripts to make them appropriate for hard disk booting (e.g., remove the "Enter a password for the default user" prompt)

Good luck! Direct questions and comments to tmcnulty@ppckernel.org.

TODO List for BCCD/PowerPC

Fix Runtime Build Environment

1) /etc/ld.so.conf

change /lib/gcc-lib/i386-lnxbbc-linux to /lib/gcc-lib/powerpc-lnxbbc-linux

bccd/special/bbc-provided/files/ld.so.conf

2) LD_LIBRARY_PATH in /etc/bashrc

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/gcc-lib/i386-lnxbbc-linux/ should be: export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib/gcc-lib/powerpc-lnxbbc-linux/

bccd/special/bbc-provided/files/bashrc

3) Fix library install dir:

ld looks at "/tmp/${USER}.build/staging/singularity/image/lib" to find the libraries that are in "/lib":

/bin/../lib/gcc-lib/powerpc-lnxbbc-linux/3.2.3/../../../../powerpc-lnxbbc-linux/bin/ld: cannot find //tmp/gray.build/staging/singularity/image//lib/libc.so.6 collect2: ld returned 1 exit status

Not sure what needs to be changed here.. (the quick fix on the image is mkdir -p /tmp/gray.build/staging/singularity/image && ln -sf /lib /tmp/gray.build/staging/singularity/image/lib)

4) CPP defaults to /lib/cpp, but cpp is at /bin/cpp and /usr/bin/cpp

5) MPICC

export MPICC=/lam-mpi/bin/mpicc

so mpicc is found?

6) LDFLAGS - may be fixed once above issues are addressed

Maybe this will come automatically once LD_LIBRARY_PATH and/or ld.so.conf is fixed, but:

export LDFLAGS="-L/lib/gcc-lib/powerpc-lnxbbc-linux"

is currently necessary while building FFTW: 

checking whether the Fortran 77 compiler (f77 ) works... no

configure: error: installation or configuration problem: Fortran 77 compiler cannot create executables.

(can't find -lfrtbegin)

Build and Test Packages

To Build and Test

perfex

To Test

gromacs - test multinode runs
Retrieved from "https://wiki.cs.earlham.edu/index.php?title=BCCD:PowerPC&oldid=653"