I delayed installing 11.04 for some time due to bugs with kernel power management and the GNOME classic desktop, but this week I finally decided it was time to upgrade.

After fixing the usual minor breakage from the upgrade, one thing became apparent – video playback was fucked using the free software radeon driver. Since practically everything revolves around watching anime  on this machine, that was completely unacceptable. Tearing and stuttering was obvious during pans (mplayer wasnt complaining that the cpu was too slow etc, and since I started in 9.04 video playback has been fine), making anime unwatchable.

Long story short, I read up on Phoronix (http://www.phoronix.com/scan.php?page=article&item=amd_r600g_june10&num=1) to get some familiarity with the current state of X windows rendering and did some work on how to test X configurations without logging out of your main session (which is unacceptable as this machine is essentially a server running some gui apps).

It turns out that the ‘Switch Users’ feature actually allows you to spawn multiple sessions of X running under different users – and when you login as a new user, of course the new X session’s configuration is generated/read in without affecting your normal X session. With this and after reading the radeon manpage, I was able to create the following ‘/etc/X11/xorg.conf’:

Section “Device”
Identifier “devname”
Driver “radeon”
Option “SwapbuffersWait” “off”
Option “ColorTiling” “on”
Option “EXAVSync” “on”
EndSection

SwapbuffersWait and ColorTiling are set as Phoronix said they would make 3D things faster (Minecraft is still shockingly slow – 9-15fps – so I havent really seen that), however the important option is EXAVSync.

This has stopped tearing in video playback, however I still get nasty judders visible during pans. Testing on a separate X instance showed that with the above settings, playback was almost perfect – so I assume that my normal 2D desktop session is too much for radeon to cope with any more (…).

At least I can switch over to a separate user to watch videos with this, I’d rather stick to free software and not use the proprietary driver.

If anyone has ideas of how to fight this issue further, I’m all ears (enabling the xorg-edgers PPA with the above xorg.conf actually lead to performance worse than the stock 11.04 configuration, which worries me greatly for the future).

When Tools -> Folder Options -> View tab -> Advanced settings: Remember each folder’s view settings is enabled, Explorer saves folder display options (the View setting, sort option etc) in ‘ShellBag’ entries in the registry, see ‘HKEY_CURRENT_USER\Software\Microsoft\Windows\ShellNoRoam\Bags’ and ‘HKEY_CURRENT_USER\Software\Microsoft\Windows\Shell\Bags’. There is a maximum number of folders that can have these settings saved for, and this is located at ‘HKEY_CURRENT_USER\Software\Microsoft\Windows\ShellNoRoam\BagMRU Size’ – this defaults to 5,000 in Windows XP SP2 (remember to edit it as a Decimal value).

NirSoft has a useful tool that allows you to see all of the ShellBags currently defined, including their actual path and the ability to hop directly to the associated registry values, called ShellBagsView. Find the folder you are interested in and hop to the registry – the sort method defined for the folder is maintained in the ‘Sort’ key, which can take the following values:

0: Sort by name
1: Sort by size
2: Sort by type
3: Sort by modified

Hopefully this will get Explorer to behave.

However, this didn’t do the trick – Windows still managed to be ~2 minutes slow even after a successful update the prior day duing testing, which is completely unacceptable (its my main time source for catching the bus in the morning).

The lastest attack on this issue revolves around a recent discovery of Windows Time settings in Group Policy:

1. Run gpedit.msc (Group Policy Editing MMC snap-in).
2. Local Computer Policy -> Computer Configuration -> Administrative Templates -> System -> Windows Time Service -> Time Providers:
Enable Windows NTP Client: Enabled.
Configure Windows NTP Client: Enabled.
Type: NTP (NT5DS means that the client attempts to synch with a Domain Controller’s time service, which doesn’t exist on my network).
Special Poll Interval (the setting to determine update frequency): 3600 (seconds, hourly).
EventLogFlags: 1 (log time jump changes – this should give me a proper audit record if the NTP client is doing its job).

See Windows Time Service Tools and Settings, search for Group Policy – provides some documentation on these settings and their associated registry keys.

After making the changes, the Internet Time tab of Date and Time Properties correctly acknowledged that time is synched hourly, so I have my hopes up (I think the default is weekly).

I tried the RCs etc, but still the instability and bugs made it unviable. I was planning to look into uTorrent’s built-in functionality to filter based on IP addresses as a second-best alternative, but haven’t got round to it yet.

Thankfully, some kind people have forked Peer Guardian 2, creating PeerBlock! (found via the Phoenix Labs RSS feed). So far this has successfully installed on all of my machines, and I hope it will continue behave itself and be worthy of standard install status like Peer Guardian 2 used to be.

However, as usual, it brings its own problems…

When I wanted to enumerate internal Windows services with sc.exe (a commandline program that allows you to create, delete, list etc services), it failed to run. WTF?

In fact, it did execute, but promptly terminated itself. Running the exe on another machine worked, and XP’s sc.exe worked on the Server 2003 machine as well. So why would this fail?

Process Monitor showed that sc was searching for  w03a3409.dll in various PATH places, but not finding anything. Everything (a fantastic file indexer and instant search tool) confirmed that this dll didn’t exist. Searching through the registry however showed it was associated with two hotfixes:

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Updates\Windows Server 2003\SP3\KB923561

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Updates\Windows Server 2003\SP3\KB956572

The latter was key – it talked about sc.exe in one of the FAQ answers. I tracked down the installer through Microsoft Download Center (thanks for not directly linking to the installer in the article Microsoft) and installed it on the server (why didn’t it complain that the hotfix had been installed before?).

Low and behold, even before restarting the machine as it requested, sc suddenly worked. Presumably this is another M$ fuckup then… Both before and after this issue, the machine checked out as being completely uptodate according to Microsoft Update.

This time my secondary machine died – I host my database server, wiki and issue tracker on it (and I use the latter two a lot to record issues and try to record and keep progress permanent).

First, the PSU started to whine – prolonged bouts of  a very high pitch tone, just loud enough to hear even with ear defenders. I remembered this had happened in the past, so I put up with it for a while… except this time it didn’t go away. I finally decided to turn the machine off for 5 minutes to give it a ‘rest’ (which means its failed its job of being a permanently-up server), however the POS didn’t turn back on again.

I swapped out the shitty FSP Group PSU for a spare Antec and tried to start the machine up – but no luck, except for one instance where I quickly managed to create fresh images of the drive partitions. Further tests after that failed.

Inspection of the motherboard revealed I had a few leaky capacitors – so presumably it had been afflicted with the Capacitor Plague, which does fit given that it is a 2002-3 era machine. Two more additions to the shitlist in the same day then (FSP for PSUs, MSI for motherboards).

Windows systray popup detected:

System Event Log
================
Type: Information.
Timestamp: Fri Jul 10 00:12:05 2009.
Event: 0.
Source: Application.
Category: Event Log Corruption.
User: SYSTEM.
SNARE Event Count: 1.
Description:
The event log is corrupt and requires immediate attention

This is an excerpt from an email I was sent by SNARE / Kiwi Syslog Daemon when it apparently happened (SNARE and Kiwi are services I run that allow me to get emailed when important events happen on my machines). The SNARE service promptly died after.

I vaguelly remember having this issue in the past – presumably the only way to ‘solve’ this issue is to clear the event log and start afresh, which I have done. Good thing I don’t usually use old event log data.

Blowfish-tunneled X11 was definitely faster, but still not fast enough. During testing, the uTorrent window managed to get stuck whilst rendering the General tab of a large torrent – it was literally finding it so hard to update the piece graph/bar that it refused to move to another tab!

That was clearly no good, so I had a look into NX Server.

Installation on the server side was pretty convoluted – NX Server, NX Client and NX Node all need to be installed for the server to work, and client authentication is multistage, requiring sshd_config changes.

It seems the NX client installed locally attempts to:

1. Access the SSH server first under the ‘nx’ username, authenticated via PKC and presumably a default DSA key installed with the server (that you can change).
2. Somehow authenticate the user you want to access under with SSH via PKC (this user must be recorded in SSHD’s AllowUsers so I assume SSH is involved here even though NX has already gained access).
3. Use the password submitted by the client to login again? This could just be an optional NX Server authentication step mind. In my case I disallow password authentication for SSH in general.

When all this is done, NX launches a GNOME/KDE/etc desktop as requested, or any command you want with a custom session.  So, have it run Gnome Terminal/Xterm etc, then  screen and run any individual GUI apps you want in the screen virtual terminals (gnome-session to generate your own GNOME desktop amongst others e.g.).

In my case, outside of testing and curiosity I wanted to see how uTorrent and GNOME System Monitor fared. Both cases speed-wise were a success! NX Server/client was definitely fast enough to maintain speedy responses when using both applications (and indeed whole desktops)… however the CPU usage associated with these applications was excessive (~25-50% on my Athlon XP 2.25GHz). No CPU overhead was experienced when the windows were minimised, but that is beside the point – I want these windows constantly on display.

VNC proved to still be faster than NX, and used almost no CPU to render a complete desktop including a busy uTorrent window maximised… having a VNC window up is almost the same as a properly integrated uTorrent window, so for now I’ll stick with VNC for remote control/monitoring status.

Perhaps this needs reinvestigating when I make the next generation machine.

Unfortunately later on I noticed that 1.8.3 seemed to be causing a new problem on my system. The torrents I was seeding should be generating 1.5-2MB/Sec of constant u/l bw (bandwidth), however the u/l speed had gone to hell, with small infrequent spikes:

Sometimes seeding would resume, go fine for a while and then suddenly die again:

Note that no odd WINE errors were being reported when uTorrent was run via a terminal, however some uTorrent verbose logging was throwing up 10047 (an address incompatible with the requested protocol was used) errors during the problem periods, presumably resulting from the real issue behind this.

Reading this version’s official announcement thread, I noticed that some people had issues with seeding and uTP. So, I googled how to disable it – change the Advanced setting ‘bt.transp_disposition’ to 5.

After restarting uTorrent, sure enough, seeding returned to normal. Infact  I got a nice steady ~4MB/Sec upped thanks to my efforts. More happy leechers.

Lets hope its a WINE issue that will eventually be fixed – don’t want to be shut out from the future here.

Sending windows that are supposed to exist on your Linux server to your local Windows desktop? Awesome, but surely unpossible?

At least, it was at the start…

My first task was to get Cygwin/X working with a current Cygwin install on my XP machine. Presumably once I had mastered that, I could then pull it off remotely from my Debian server. Unfortunately, whenever I tried this before I constantly got bs errors back saying that Cygwin/X couldnt connect to the display – I remember this put me off when I first discovered X11 forwarding last year. Determined to make a difference this time, I played with an XP VM. I currently use Sunbelt Personal Firewall (SPFW), and from previous googling I suspected that it might be interfering with things.

After uninstalling it on the VM, low and behold, Cygwin/X managed to connect to the 127.0.0.1:0.0 display straight away, and I saw my first rootless X windows ever (I had already been able to render X windows in a virtual desktop window, for some reason). As much as I love SPFW, this is pretty poor. Obviously Cygwin/X was allowed full in/out access on Trusted Networks, which of course the localhost is. Still, its going to need a lot more than that for me to ban this FW, so I left the investigations here for Cygwin/X (if anyone has any ideas as to what to do here, I would much appreciate them! Perhaps there is an obscure setting somewhere that I’m not aware of).

Continuing with tests on my main machine, I next moved on to Xming, which is supposed to be a better implementation of X Windows for Windows. Sure enough, my first experiences with it resulted in GPFs. However, using the XLaunch helper app with Xming is more flexible than Cygwin/X. I set the display to 1 instead of 0, and along with setting up Putty to forward X11 stuff to ‘localhost:1′ (it seems the fact that SSHD defaults to ‘localhost:10.0′ does not affect this), and I finally got my windows coming through SSH (hello xeyes).

Playing around with stuff like gnome-system-monitor and a full desktop via gnome-session showed that this method was very slow and CPU intensive (Athlon XP 2.25GHz FTW), and therefore not really practical for constant use.

Still, it was a proof of concept and suggested that there were better things to come – more posts will follow.