Recently, Alex Fukunaga started up a blog about Lisp. One of the things he has done is make CFFI bindings for mpich2. Here is an introductory post about those bindings with a link to his CL-MPI site.

Today, I have been working on getting his bindings up and running under Ubuntu Linux and Mac OS X.

So far, it has been a nightmare trying to get mpich2 compiled on either platform. It compiles famously if you just run

% ./configure --prefix=/usr/local

However, that doesn’t build shared libraries. Without shared libraries, the CFFI stub library happily links against the static library and happily imports zero of the symbols. After sifting through configure –help, I decided that I should add –enable-dynamiclibs. Doing so yells at me that I must also specify –enable-sharedlibs=type. So, I sifted through the help again and decided to try:

ubuntu% ./configure --prefix=/usr/local --enable-dynamiclibs --enable-sharedlibs=gcc
macosx% ./configure --prefix=/usr/local --enable-dynamiclibs --enable-sharedlibs=gcc-osx

Both compiles then bomb out saying MPIU_CALL_MPIDI_CH3 is undeclared. Yay.

Searching the web on the relevant errors, I discover that I do not need to enable dynamiclibs to get shared libraries. Wheee. Now, mpich2 is built on both architectures.

I added this target into cl-mpi/mpich2-stub/Makefile:

libmpiskeleton.dylib: mpiskeleton.o
        mpicc -dynamiclib -o libmpiskeleton.dylib -dylib mpiskeleton.o

and changed the bottom of cl-mpi/cl-mpi-configure.lisp to this:

(eval-when (:compile-toplevel :load-toplevel :execute)
  (pushnew :mpich2 *features*) ; Use MPICH2                                     
  (defvar *mpi-header-file* "/usr/local/include/mpi.h")
 
  ;; For MPICH2, Need to load a special stub shared object, and not the MPICH s\
hared library directly                                                          
  (defun load-mpi-foreign-libraries ()
    #-darwin
    (cffi:use-foreign-library "/usr/local/asdf-install/site/cl-mpi/mpich2-stub/\
libmpiskeleton.so.1.0.1")
    #+darwin
    (cffi:use-foreign-library "/usr/local/asdf-install/site/cl-mpi/mpich2-stub/\
libmpiskeleton.dylib")
    ))

After sorting out some svn merge problems in mpi-test.lisp, it appears as though it all works wonderfully on MacOSX. Most of it seems to work under Ubuntu for me, but I have to look into some error messages more closely.

When I first noticed it, I tried searching around with iostat and top to see if I had some process run amok. Finding nothing, I tried rebooting.

The magical reboot was not so magical. My machine was still merrily hitting the disk every couple seconds.

I searched around the web for “disk activity” linux and “disk activity” ubuntu. I found recommendations that I:

• Kill processes that poll for new media
• Eliminate access time tracking on my filesystem, and
• Set syslog not to flush some of its logs

All of those seemed bogus to me. None of them seemed like anything that would have changed yesterday or today. But, the last one got me thinking. What logs do I have that do get flushed on every message? Are any of them going batty today?

Sure enough, /var/log/auth.log was getting hit every two or three seconds. Someone was trying to guess logins on my machine.

May 28 04:57:57 evariste sshd[6966]: pam_unix(sshd:auth): authentication failure
; logname= uid=0 euid=0 tty=ssh ruser= rhost=nitro.xyzdns.net  user=backup
May 28 04:57:59 evariste sshd[6966]: Failed password for backup from 209.51.159.
194 port 54874 ssh2
May 28 04:57:59 evariste sshd[6968]: Invalid user guest from 209.51.159.194
May 28 04:57:59 evariste sshd[6968]: pam_unix(sshd:auth): check pass; user unkno
wn
May 28 04:57:59 evariste sshd[6968]: pam_unix(sshd:auth): authentication failure
; logname= uid=0 euid=0 tty=ssh ruser= rhost=nitro.xyzdns.net 
May 28 04:58:01 evariste sshd[6968]: Failed password for invalid user guest from
 209.51.159.194 port 54992 ssh2

Looking back through the last week, I’ve had a few spats of bogus login attempts. Most of them were just four or five attempts in a row. This one today had gone for twelve hours.

I threw that IP address into /etc/hosts.deny. Now, all is quiet. Well, except my need to make sure I don’t let things go for twelve hours again….

I have a PC that I can boot into either Ubuntu Linux or Windows Vista. Of course, I have a variety of services running under Ubuntu Linux on that box that the rest of my network would rather have around. As such, I would rather never boot that machine into Windows. So, I thought I’d give VMWare Fusion a try.

Actually, I thought I would try both VMWare Fusion and Parallels. Alas, Parallels lets me get my virtual machine set up, but will not let me run it without a license. VMWare Fusion lets me play for 30 days before buying a license. From what I’m seeing from VMWare Fusion’s performance, I can’t imagine dropping $80 on Parallels just to see if its virtual machine can outperform what I’m seeing from Fusion.

Right now, I am in the process of moving over the PC’s Windows stuff to my laptop so I can try running Vista through Fusion. While I was waiting for that, however, I installed Ubuntu under Fusion, updated a ton of packages, installed emacs, sbcl, slime, etc.

For comparison, I took some lisp code that runs in just under 11 seconds on my laptop. I ran the same code under Ubuntu in Fusion on the same laptop. It ran in just under 12 seconds. Some of that may also be that I am using an older version of SBCL under Ubuntu than I am native.

I have some more testing to do to make sure that cl-opengl will perform as well. But, I am quite pleased.