Stuart Herbert | Gentoo

For the last week or so, I’ve tried running Linux natively on my laptop. VMware’s a great piece of software, but the overhead of running Gentoo as a guest O/S gets to be a problem at times. Performance for CPU-intensive tasks (like compiling) is good, but anything that’s disk I/O intensive (like booting, or cvs up) takes a lot longer. Memory usage is also an issue, because some of the RAM has to be left for Windows XP. (Well, a bit more than “some”, alas).

The reason I have a laptop is that I’m never sat at a desk at home. I probably spend more time online from wireless hotspots than I do via wireless @ home. For me, wireless has to work, otherwise I can’t get things done.

And, although it’s definitely getting there, one week with wireless on Linux has convinced me to switch back to VMware for now.

Forgetting driver issues (everyone knows about those, and why card manufacturers are reluctant to release open-source drivers), there are three issues that have convinced me to switch back:

• Unreliable connection
• Trouble with hotspots
• Gentoo baselayout and daemon support

Unreliable Connection

I’ve repeatedly had problems with connecting to access points in the first place. wpa_supplicant often needs several tries before it can authenticate (whenever this happens, it does this thing where it tries authenticating against a bssid of 00:00:00:00:00:00). For open & WEP-protected networks, switching to wireless-tools avoids this problem. I don’t have a problem w/ having to use the wireless-tools (instead of wpa_supplicant) for open & WEP-protected networks, but unfortunately I’ve had the wireless driver crash a couple of times when using them, forcing a reboot to fix things.

It’s not just me; I watched on Saturday as Tim was also having problems connecting to a wireless network.

The situation in Windows actually isn’t much better in general. At work, we’ve had lots of trouble trying to get all the laptops working with the same encryption scheme at the same time. (Our experience is that Sony laptops are extremely temperamental on this. YMMV). But I’m lucky (I guess) because wireless under Windows works great with my Dell Inspiron.

Trouble With Hotspots

Unreliable I can live with (hey, I put up with evo for a year <grin>), but not being able to get onto the Internet via wi-fi hotspots is something that I can’t live with.

I have a BT Openzone account. Thanks to the Wireless Broadband Alliance, it works on plenty of other networks too - including the T-Mobile hotspots that you can find in pretty much any Starbucks. If you’ve never used a hotspot, they do this thing where, once you have an IP address, when you open up your browser you’re redirected to their login page. Only after you’ve logged in can you then access the wider Internet.

And, unfortunately, I haven’t once managed to get connected to a T-Mobile hotspot using Linux natively. The wireless interface reports that it is associated with the hotspot, but the DHCP client can’t get an IP address.

When I have some free time, I need to use Ethereal on Windows to see how Windows connects to the hotspot, and then compare that with the process under Linux. That should show me what I’m not doing right, with a bit of luck.

But me and free time are creatures that have fleeting meetings at best

Gentoo Baselayout and Daemon Support

If you’re a roaming user, the idea of starting up a wireless connection when you boot the box has limited appeal. The process Windows uses makes a lot of sense:

1. Start up the wireless interface
2. Connect to any “preferred” wireless network, if one is available
3. Assign an unroutable IP address if no “preferred” wireless network is available

As best as I can tell, our current stable baselayout is missing step 3. If wlan0 is not associated with an access point, it has no IP address. Which leads on to the problem with daemons …

The problem with our daemons is that many of them require networking to be started (’use net’ in their init script’s depend() function) otherwise they won’t start. Stable baselayout doesn’t seem to consider starting lo as satisfying the ‘net’ dependency, so stuff doesn’t start.

Having the wireless always considered to be started (even if it’s using the fallback, unroutable IP address) would solve this problem nicely.

Update: see the comments for how to configure baselayout to treat ‘lo’ as satisfying the ‘net’ dependency, and how to fallback to a static IP address if wireless cannot connect. I’ll give these a try the next time I boot natively.

The Advantage Of Using VMware On Laptops

The nice thing about running Windows XP as the host o/s on a laptop is that all of these problems simply go away. It handles all the wireless for you; your Linux guest o/s simply sees what it thinks is a wired network connector. No more problems with wireless drivers under Linux either (which probably goes a long way to the stability).

And you get added benefits like suspend to disk Some high-profile kernel folks seem believe in a world where suspend to disk isn’t a feature that Linux needs. Personally, it’s an important feature for me. I don’t use suspend to RAM, and it’s not a feature that I personally find remotely useful.

I’ll give Linux wireless another go at some point in the future, but, tbh, once I get my Merom-based laptop, I’m expecting VMware’s performance to be good enough that I probably won’t need to think about native booting anyway. (It’s more likely to be running under Parallels than VMware, fingers crossed).