Solarizer

Last night at the LOSUG there was a great talk by Sun’s Jarod Nash on ZFS “Under the hood”.

There must have been 40 people there, which was a great turnout. I suggested the meeeting was mentioned on the zfs-discuss mailing list - given the topic, but surprisingly to me that only brought in one new face. So much for good suggestions

Jarod was enthusiastic and definitely knew his stuff. So much so that there was actually a handout! The flip-side had a list of the acronyms used in the ZFS architecture block diagram on the front, and throughout the evening there was a gentle wafting of air from people flipping the handout over and over…

What I still haven’t figured out (see the title) is why the handout was laminated.

There’s a really badly written article called Sun hopes for Linux-like Solaris over at ZDnet in Australia.

It would appear from the article that Sun can’t use “Linux tools” such as bash, because they’re GPL and Solaris isn’t. For one - bash isn’t a Linux tool. Secondly, Sun ships bash. What sort of morons write this garbage (the article, not bash)?

There’s a kernel (no pun intended) of truth in the article, though you have to think a bit laterally sometimes to detect it. It seems to be talking about the currently Sun-internal Project Indiana.

Sun’s packaging (which they inherited from SysV) indeed sucks and needs fixing. But Linux has sucky packaging systems too - step forward rpm and say corrupt DB environment.

Sun’s contributed massive amounts of code to GNOME, and GNOME ships with Solaris 10. Nothing to change there then.

Sun’s hardware support for x86 isn’t good enough, although it is getting much better. Sun’s hardware compatibility list is however a poor indicator of what works and what doesn’t - it doesn’t include SATA chipsets that Sun use in the Thumper for instance.

It is hard to see how they’d be able to merge any code (ignoring licensing issues) from Linux in to address considering the different driver models and Linux’s cavalier approach to interface stability. But they have got versions of FreeBSD’s drivers in Solaris 10 already, so perhaps it is possible. More drivers would be good.

I think there’s some changes in Sun Studio 12 so that the tools support gcc flags. I’m not overly convinced that’s a good idea, but they’ve not broken any existing argument handling, so that’s fine by me.

What things do you think Project Indiana should do to improve Solaris?

Jonathan showed off “first silicon” of the Rock multi-threaded multi-core CPU yesterday. It looks very cool, and will be interesting to see it perform inside actual tin.

But looking at all those 2395 pins, it reminded me of one of those combs used on pony manes.

No, I’m definitely not thinking of this

I’ve been trying to install OpenSolaris on my Mac.

Short summary - arrrrrggggh!

Right, that’s off my chest now. Now I’ve “only” got a G4 Powerbook, i.e. it won’t run Intel binaries directly, like OpenSolaris x86. The only way to run Intel binaries is to use a CPU emulator, of which there are basically two.

1. Microsoft Virtual PC 2004

Although end-of-lifed, apparently Microsoft are still selling it for Macs. Bless their little capitalist socks. But you can get it for free via MSDN, so I tried the latest and “greatest” version 7.0.2. It doesn’t work.

Problem #1 is that it can’t use disk images bigger than 2GB, so forget about using that DVD disk image.

So you burn a DVD and try again. Actually, it starts pretty fast. But unfortunately early on in the OpenSolaris kernel initialization it panics. No way around this. Bug entered at opensolaris.org.

2. QEMU

This is an open source emulator, and there’s a nice-ish Mac front-end to it called Q.app.

It boots. It is pretty slow. But it boots the Solaris Express (OpenSolaris b60) disk image! W00t! It starts installing!

A day later (I kid you not) it finishes installing. It reboots. It fails to start because it can’t read the filesystem on the disk image properly. Try rebooting again, and now it can’t even find the Solaris kernel. Jeez.

3. VMware Player

OK this was cheating, but I was desperate. I installed Solaris Express on VMware Player on a Windows box. It looks pretty good. I found a tool that would convert the VMware virtual disk image to a Virtual PC virtual disk image, and copied that across to my Mac.

Tried booting that in Virtual PC - identical kernel panic to the one described above, which wasn’t surprising.

Tried converting it to a QEMU disk image. Tried starting it in QEMU - no dice; same sort of wierd disk errors as before.

Arse.

The only solution seems to be to get a new Intel machine just to run OpenSolaris.

Though come to think of it… isn’t there a PowerPC port of OpenSolaris around? Hm…

Here’s a curious bug report on OpenSolaris: “zfs_rmdir() can return EEXIST on an empty directory when spotlight is messing with it”

This looks like a slightly better indication that ZFS really is coming to OS X, than some mockups from IB

More at Ars Technica.

The votes are counted, and the results are in. The new governors are:

• James Carlson
• Alan Coopersmith
• Casper Dik
• Glynn Foster
• Stephen Lau
• Rich Teer
• Keith Wesolowski

Congrats to you all! Perhaps you’ll allow more people to vote next time?

Voting is open for positions on the OpenSolaris Governing Board.

But don’t bother unless you’re certain they’ll let you! Lowly committers don’t seem to be permitted to express themselves at the polling booth.

All the cool kids are moving to OpenSolaris

So we’re trying to get Sun Cluster 3.1 failing over an instance of M-Vault (our X.500/LDAP server) between two Solaris 10 (x86) machines. If you recall I mentioned way back when that we’d added SMF support so we were good Solaris citizens…

But it doesn’t work. Run the start script by hand, which pretty much just does a svcadm enable eddy and it works. The stop script does what you might expect, and also works.

Copious debug-by-printfs later, we decide that the cluster manager’s starting & stopping the server several times, not involving the probe command, and then just giving up.

A little googling… Ah, how to use Sun Cluster 3.1 with a service managed by SMF is mentioned in the clustering release notes. Click link. Here we go. To do that:

1. Don’t use SMF.
2. …

Round objects.

But apparently Cluster 3.2 integrates heavily with SMF, so our work wasn’t wasted

One thing I think Solaris needs badly is some GUI tools for managing the machine. There’s an OpenSolaris project called Visual Panels which is intended to address this, but it uses Java and Swing and looks like an interface only a programmer could love

As I was noodling away on my little Cocoa “Core Data” app on my Mac, I remembered that of course Solaris used to have a version of OpenStep, which was the set of UI frameworks and tools that was part of NextStep (the OS) and is now part of Mac OS X.

There’s a screenshot of OpenStep on Solaris over here.

Cocoa produces beautiful looking and powerful applications, and is very easy to write to. It would be wonderful if we could get OpenStep back on Solaris, if only for the ease of development it brings.

Maybe Jonathan knows what happened to it?