[plug] Linux Desktop Market

[Craig Ringer]

Chris Watt wrote:

> What I'm curious about finding out from people (probably in the
> archives, but fresh conversation is essential in this industry) is
> whether or not people on this list, as professionals in the IT
> industry believe that desktops and laptops with Linux preinstalled and
> configured for best use of the system would fly as a retail product.

My experience so far has not been great. I've been running Linux thin 
clients for several years now, and the longer I use them the less happy 
I am with them. The main reason, I think, is that Microsoft, and to a 
lesser degree Apple, have largely ceased to suck, and by comparison my 
Linux thin client deployment looks worse and worse.

Overall, they're still doing a rather good job. All the apps are 
improving too. Plus, there's always the no-license-fees issue to 
consider. It's just that there are always enough niggling problems to 
make things really, really annoying.

Why? Because the apps are all just broken, weird, and buggy enough to 
make the experience continuously frustrating. It's little things, but 
there's an endless stream of "little things" that begins to get really, 
really aggrivating. I can report bugs, and these days sometimes send in 
patches, but it's still a large and frustrating waste of time for me and 
the users. Anyway, for anything I resolve or that's fixed by an updated 
version, new problems are guaranteed to appear.

OO.o, Thunderbird and Firefox's single-instance-per-user stuff is all 
slightly broken. All these apps will occasionally just refuse to start 
up because they think they're already running, or because there's a 
dead-to-the-world left-over process that failed to exit. This often 
happens after an unexpected X server disconnect, like the user turning 
off their client without logging out. Similar issues seem to exist with 
some of the persistent GNOME session processes. I now have a cron job 
that sends a SIGTERM, then SIGKILL, to all "user" processes at 3am every 
night, and that cuts out most but not all of the problems.

Mozilla Thunderbird likes to ignore the user's umask. It sometimes just 
doesn't feel like opening a message until it's closed and relaunched. 
It's progress indication is flakey, and it seems to have some sort of 
race that causes it to save attachments from the IMAP server as 
zero-byte files sometimes.

I could try Evolution, but my server only has 4GB of RAM. Anyway, Evo 
2.0 has some "interesting" ideas about how default apps and MIME types 
should be configured, and it seems to like to forget the settings 
periodically anyway. Oh, yeah, see "single instance per user," too. Did 
I mention I only have 4GB of RAM?

File and print dialogs. I don't need to say any more about that.

GDM seems to get confused when it's been running for a very long time, 
and eventually just stops serving XDMCP sessions. It seems to get 
somehow hung up on an old gdm instance for an X server that was closed 
down weeks or months ago. Finding the oldest running gdm process and 
killing it gets it all going again, but it happens enough that I've 
moved to XDM. Recent GDM versions seem to do this less often, and 
eventually recover on their own, but that's still twenty minutes of the 
terminals not accepting new logins...

I've even had the odd kernel-level (or otherwise very low level) issue 
that's just magically gone away with a reboot. The most recent one was a 
run of firefox processes that just won't die and sit hogging CPU 
(reported as "system" time) until I rebooted the server. I felt like I 
was running Windows NT.

The list goes on. It's little things, but there are enough little 
things, and they're disruptive enough, that I'd currently be somewhat 
reluctant to recommend a Linux deployment to anybody, even for "basic 
needs" users. Sure, I can fix these issues, at least some of them. 
However, i'd have to:
   - Hold up the user while I attempt to debug the issue. They just want
     to keep on working, but in many cases fixing the problem destroys
     the information required to debug it. A core file is often not going
     to be sufficient.
   - Run debug binaries on my production server. Fun, especially with
     OO.o and the Mozilla apps, or with GNOME.
   - Spend enough time working on these issues that I'd never get any
     other work done.

It's that latter point that's the killer. If it wasn't for the CALs I'd 
be very, very tempted to jump ship to Win2k3 terminal server right now, 
as I've had some recent experience with them that was actually rather 
darn impressive. So long as you only run MS and Adobe software, anyway 
:S . Given the CAL cost, I'm sticking with our current setup, but it's 
hardly all peachy.

It's not just thin clients either, though they make the problem worse by 
providing a very long running server and a rather higher than usual rate 
of unexpected X server disconnects. I also hit quite enough issues in my 
day to day Linux use, especially now that I pay attention and /notice/ 
the problems instead of just unthinkingly working around them. It's a 
real experience to work on your system for a few days "thinking like a 
user".

That said, I'm increasingly wondering if I should take advantage of my 
recent move to Xen and replace Sarge with a more desktop-oriented, 
better integrated distro for the GUI terminal server. I'm running into 
more and more issues on Sarge that I just don't seem to hit on my FC4 
box at home.

I should note at this point that there's no way in hell you could stop 
me from using Linux, especially Debian, for the core network operations 
- general network services, file serving, email, etc. Print services 
(because CUPS is pure evil) and terminal services look like increasingly 
attractive candidates for another platform though. I'm sure it's just 
the grass looking greener on the other side of the fence, but the grass 
on this side has a few too many thorns for my liking.

Despite all that, I'm still a happy Linux desktop user. I'm just less 
and less inclined to see it as a good idea for business deployments or 
for deployment to non-technical users without a geek in the house. Not 
unless you want to take on the role of a distro builder / customiser / 
fixer and software QA team, anyway, or you want to sign up for something 
like RHEL.

Even so, overall the thin clients at the POST do a good job. There are a 
lot of frustrating problems, but then, well, I didn't have to pay for 
the software, it's virus free, and I'm not paying for CALs, annual 
maintanance, etc. That counts for a lot. Alas, none of that really helps 
address the "keeping the sysadmin sane" issue.

Given all that, I'd be very, very reluctant to recommend Linux to a 
business or home user without ready access to technical aid.

> I read a lot about Linux installed systems selling at higher prices
> than Windowze installed systems

I'd be inclined to say "offered for sale at higher prices than," at 
least if you're talking about the big OEM offerings. The reason for that 
price discrepancy is more likely to be to do with contracts between the 
OEM and their dominant software vendor than anything else.

It's also worth noting that most disclaim any support obligation for the 
"OS-less" machines beyond hardware warranty.

> So, does anyone think that a company could be founded on the ability
> to offer premium computers (so computers ready for the internet,
> office apps, etc.) could be sold to the public or do you think there
> are still factors hindering the adoption of this system, esp. by
> schools, seniors and first time users?

I think there are a lot of factors against. It'd be possible, but you'd 
probably need to repeat "will not run Office or Windows software, and 
won't work with a lot of off-the-shelf hardware" to the point where you 
tattoo it to the customer's forehead. I wouldn't want to be the poor 
person handling support, either. What distro? What app versions? OK, 
open a terminal and....

My personal impression is that the current ideal area for Linux 
preinstalls is in the workstation market, and that's already fairly well 
catered for between the big OEMs and the whitebox crowd.

I also think the direction being taken at CA is very interesting, in 
terms of targeting users with very basic needs who have no preconcieved 
expectations in terms of software. I'll confess to being a little worred 
with regards to the robustness issue I've mentioned above, but honestly, 
I don't think it's as much of an issue when rebooting the whole box will 
generally fix the problem. Anyway, the CA folks are doing a lot of work 
into customising the distro to avoid many possible problems.

--
Craig Ringer