Viraj's Weblog

This weekend I finally installed Leopard on my macs. I have an old Powerbook G4 15" and a Mac Mini. Both of them run it great (I have at least 1G RAM on each). At home I also have an Ubuntu server running on a laptop which I use for various Linux things. I've been sharing drives via Samba on it and hacked up my own backup scripts.

But with Leopard, there is this cool Time Machine thing I can use right? Well I wanted to use a network share via Samba with Time Machine. Turns out it is not so straightforward. OS X by default won't let you use the network drive. This post describes how to get around that. There is also a wealth of information here. Great, now I can select the network drive in Time Machine.

But now I was getting an error about creating the disk image. I watched a bit what it was actually trying to do on the Samba share and noticed that it was using the machine name in the filename being created, which is a 'sparsebundle' file. My machine name was something like 'Powerbook G4 15"' and I had a feeling the '"' or some other character in the filename was not playing well with Samba. So I changed my computer name (via Preferences -> Sharing) to some very simple name. Still no go .

After some searching I came across this post which seems to describe my problem. It generally suggests to create the disk image file manually and then copy it to the network share. But I wondered, why not try creating this directly on the network share? I should have the same problem Time Machine gave but perhaps I can get more information and debug further. So I tried it, but unfortunately the command given in the post was wrong. A few comments below describes the correct one (wtf don't they fix this?). Here is the command I ran:

hdiutil create -size 50g -fs HFS+J -type SPARSEBUNDLE -volname "Backup of powerbook" powerbook_000d93b43026.sparsebundle

And I got the unsexy error:

hdiutil: create failed - Operation not supported

Fooey. Ok, I didn't get much debug info there. It sounds like some operation is being tried on the Samba share, but what? Next I followed the rules and created this image on a local drive. That worked fine. Now the post states to copy this image over, but I instead tried to 'mv' it. This should try to replicate everything at the destination and perhaps give me more info. It did:

mv /tmp/powerbook_000d93b43026.sparsebundle .
mv: chown: ./powerbook_000d93b43026.sparsebundle/bands/62: Operation not supported
... lots more
mv: chown: ./powerbook_000d93b43026.sparsebundle: Operation not supported
mv: /bin/cp: terminated with 1 (non-zero) status: Cross-device link

Interesting! So it looks like the real culprit here is 'chown'. OS X is trying to chown these directories on the Samba drive, which of course is not supported. Looking at the files created by hdiutil, it looks like it is trying to chown the group 'wheel':

powerbook_000d93b43026.sparsebundle valankar$ ls -lRa|more
total 16
drwxr-xr-x@ 6 valankar wheel 204 May 24 10:27 .
drwxrwxrwt 10 root wheel 340 May 24 10:27 ..
-rw-r--r-- 1 valankar wheel 498 May 24 10:27 Info.bckup
-rw-r--r-- 1 valankar wheel 498 May 24 10:27 Info.plist
...

Whatever, knowing this is not helping me much. In fact, I just found another post where someone digged further. After copying it over (actually the 'mv' copied it just fine but bombed on the chowning), I enabled Time Machine and it generally worked. But it worked SLOW.

As an aside, I've generally had some slowness using Samba when it involves lots of intense disk access. Say I'm watching a video as well as recording something with EyeTV on the Samba share. It generally will clip the video pretty badly. No amount of Samba tuning on my part could improve this and I eventually gave up. But Time Machine was ridiculously slow, and it drove me to wonder, "WTF am I using Samba for anyway?" I mean seriously, I have no Windows boxes here. Why aren't I using Apple File Sharing? Surely Linux supports it right?

Thus began my journey into setting up netatalk (AFP) on Ubuntu. It was a short one and generally without hiccups. This page has some pretty good docs, however with Ubuntu there is one stickler. SSL support is not compiled into netatalk, and this is required for Leopard. The workaround is described here. Sheesh, why even include the package if it's essentially unusable? Well it works with Panther, albeit with a warning. But with Leopard you just get an error trying to connect.

Now how do I connect to my share? Well I have yet to figure out how to show my shares in Finder. Instead I have to go directly to Go -> Connect to server. In there I type the full path of the share, e.g.:

afp://system76/timemachine_powerbook

At this point I came across a problem. In my /etc/hosts file on the server I had something like:

127.0.1.1 system76 ...

And when I attempted to connect, it hung at trying to connect to 127.0.1.1. Evidently this IP is passed along over AFP. I fixed my /etc/hosts so my hostname was not pointing to a loopback (no idea why it was in the first place).

Voila, I have a network share now. Sadly, neither hdiutil nor Time Machine were able to create the disk image on the network drive, so I still needed to create the image locally and copy it over. But once I did it worked like a charm, and much faster than Samba. It's still pretty slow for the initial backup of about 20G (in fact I think it took like 12 hours), but after that it worked pretty well. I'm a bit worried how it will deal with large backups though, as it could be lots of traffic over the network.

Another thing I have to worry about is resizing the disk image. This post describes it somewhat, and it's likely I'll have to do that at some point.

I setup the same thing on my Mac Mini, so now I have 2 Time Machines going to the same share. Plus I have AFP setup and no more need for Samba. Good riddance!

One caveat. My syslogs are filling with:

May 26 09:02:33 system76 afpd[3651]: bad function 4F
May 26 09:03:05 system76 last message repeated 89 times
May 26 09:03:25 system76 last message repeated 163 times

Turns out this is some undocumented AFP call.

Update: And now I think I've lost faith in Time Machine network backups altogether. There is a great post describing the pitfalls of such a solution. Maybe it's time to move back to rsync snapshot backups.

I had an urge to record some of my borrowed acoustic guitar and here is what I ended up with. Unfortunately the acoustic is drowned out by everything else. Again, it was all created with open source software . I wanted to get progressively harder towards the end but it ended up just getting really noisy. Anyway it was a fun track to spend my weekend on.

For this recording I was kind've inspired by a track called Hovercraft by Space Mtn. Some of the bass line is similar. I'm getting a bit better with mixing sounds though I can only go so far with the electronic drums. Still I'm impressed with how clean of a sound Hydrogen and Ardour produce.

Today I picked up a bass guitar, the Yamaha RBX170. It's a pretty low end guitar , and I generally wanted to try to create some songs with a bass loop in them.

Well here is my first try. I played a whopping 2 notes on the bass . It will definitely take me a lot of practice to play it as it's a different beast from the regular electric guitar.

I haven't been practicing guitar as much as I'd like, but every now and then I spend a few hours attempting to play. Generally what I enjoy is trying to just guitar solo over some song played from my ipod.

This weekend I played with 2 open source tools: Hydrogen, a drum machine, and Ardour, a recording tool similar to Pro Tools. I'm using them on Mac and unfortunately some of the support is lacking. However I was able to record a tune. I created everything from scratch and I have to say I really love these free tools. I had the displeasure of using Pro Tools and just thought the whole thing was a racket intended to get people spending gobs of money on the software and training.

There's a really interesting article in the April 2008 issue of Discover magazine about the time before the Big Bang. All of the common theories seek to explain what happens right after the Big Bang, but what about before it? There are a couple of new theories floating about. One of them has to do with entropy. A condensed universe has very little entropy, but its explosion, and all that comes of that, including life, is the increase of that entropy. From the article:

In any such large groups of objects, the system tends toward equilibrium. Physicists use the term entropy to describe how far a system is from equilibrium. The closer it is, the higher its entropy; full equilibrium is, by definition, the maximum value. So the path from low entropy (all the molecules in one corner of the room, unstable) to maximum entropy (the molecules evenly distributed in the room, stable) defines the arrow of time. The route to equilibrium separates before from after. Once you hit equilibrium the arrow of time no longer matters, because change is no longer possible.

"Our universe has been evolving for 13 billion years," Carroll says, "so it clearly did not start in equilibrium." Rather, all the matter, energy, space, and even time in the universe must have started in a state of extraordinarily low entropy. That is the only way we could begin with a Big Bang and end up with the wonderfully diverse cosmos of today. Understand how that happened, Carroll argues, and you will understand the bigger process that brought our universe into being.

...

These high-entropy universes would be boring and inert; evolution and change would not be possible. Such a universe could not produce galaxies and stars, and it certainly could not support life.

This furthers the idea that life as we know it is simply the dissolution from a high-energy, low-entropic state to the dispersal of that energy. It's as if life is actually decay, or rather the steps along the way as energy decays towards the equilibrium state. Life is like the burning of a flame, and the process is really just these molecules and atoms heading towards the high-entropic state. In other words, it seems life is really the process of death! Death of energy.

Another intriguing idea in the article is that there is no time at all. Our idea of time is simply the result of different arrangements of matter. That is, at one point there is one configuration of matter in a universe, then what we define as a second, or minute, or hour later is another configuration of matter. If that configuration were not changing, there would be no time. This actually makes sense to me, because if you think of life a progression of energy, if that energy is static, you would not age, no cells would change in your body, and time would stand still.

In Platonia all possible configurations of the universe, every possible location of every atom, exist simultaneously. There is no past moment that flows into a future moment; the question of what came before the Big Bang never arises because Barbour's cosmology has no time. The Big Bang is not an event in the distant past; it is just one special place in Platonia.

Our illusion of the past comes because each Now in Platonia containts objects that appear as "records," in Barbour's language. "The only evidence you have of last week is your memory - but memory comes from a stable structure of neurons in your brain now. The only evidence we have of the earth's past are rocks and fossils - but these are just stable structures in the form of an arrangement of minerals we examine in the present. All we have are these records, and we only have them in this Now," Barbour says. In his theory, some Nows are linked to others in Platonia's landscape even though they all exist simultaneously. Those links create the appearance of a sequence from past to future, but there is no actual flow of time from one Now to another.

Very interesting stuff!