... Resnet, FATMAN and JANET - (A Love Story). Or I think that's what the title of last night's talk was. Thankfully, the talk itself was more memorable than the title. There was certainly a fair amount I learned about the whole setup and history of St. Andrews connection to the world. That combined with a few pieces of kit being shown off at the end kept the attention of the audience (ooh... look at the shiny!).
A video of it has been recorded and if it makes an appearance online, I'll be sure to post.
Oh, and the talk was followed by an EGM. We now have the same person as last year looking after the website and I get a vote on the executive committee. :)
Friday 31 October 2008
Tuesday 28 October 2008
Wired Nosedives Again
The other day, wired crashed on us again. This time, no warning, some services still running. Rather odd. So I get called over to take a look at the physical box. Jaunty is there with the screen out and wired taking a rather serious hexdump:
Of course, the keyboard is not responding so we cannot scroll up to get any other messages. All we could do was reboot and investigate based on this photo and the logs (which were pretty bear).
Not good and we still can't pinpoint the issue. However, today wired went down again. This time the more regular cannot access services type of going down.
However, things were not quite so straight forward. Both the physical and virtual machine were still running! It turns out that the virtual network interfaces had been dropped and came back after a networking restart.
I really think with all these issues, the migration to the other machine can only make things better.
Monday 27 October 2008
Listeners Revisited
As I stated in a previous entry, STAR's listener figures were never known for being great. Usually only one or two people through the day maybe peaking at around 5-10 in the evening before getting nobody overnight.
However, we've just have just gone back on air. Though this time things changed a little:
The peak listener figures have shot up to what is possibly a STAR high. 47 simultaneous listeners - wow. Although things did die off towards the end of the week, it's still fairly impressive.
It should be noted that when reading the graph, 1 should be removed from any number to account for the union bar machine. However, that does still mean no overnight listeners.
However, we've just have just gone back on air. Though this time things changed a little:
The peak listener figures have shot up to what is possibly a STAR high. 47 simultaneous listeners - wow. Although things did die off towards the end of the week, it's still fairly impressive.
It should be noted that when reading the graph, 1 should be removed from any number to account for the union bar machine. However, that does still mean no overnight listeners.
Wednesday 22 October 2008
Changing Services
We're having a bit of a changed on the wired servers. We've just picked up another physical box, allowing us to segregate the services a bit better (we've not had much luck with virtualisation).
The current plan is to move the services in the wired VM onto the new physical box. This has been completed but is not yet in use as we're testing it (OMG wired testing something!?!). This new box is also running a VM that will only provide shell services. The benefit being that if the shell system is compromised (e.g. fork bomb, root exploit) the rest of our services should remain up.
But why the move from smoked to the new box? Well, the plan is to used smoked for storage services only (e.g. databases, nfs). So we have a split between services and data.
The migration will NOT be completed until we have tested the new setup. This will involve poking the committee to use the new services. And in Kieran's words "don't just log in, type ls and exit".
We did have a bit of fun with cron jobs running across two machines. Due to one updating the pid file then the other, we ended up with a huge number of GLaDOS instances on the IRC server.
The current plan is to move the services in the wired VM onto the new physical box. This has been completed but is not yet in use as we're testing it (OMG wired testing something!?!). This new box is also running a VM that will only provide shell services. The benefit being that if the shell system is compromised (e.g. fork bomb, root exploit) the rest of our services should remain up.
But why the move from smoked to the new box? Well, the plan is to used smoked for storage services only (e.g. databases, nfs). So we have a split between services and data.
The migration will NOT be completed until we have tested the new setup. This will involve poking the committee to use the new services. And in Kieran's words "don't just log in, type ls and exit".
We did have a bit of fun with cron jobs running across two machines. Due to one updating the pid file then the other, we ended up with a huge number of GLaDOS instances on the IRC server.
Wednesday 8 October 2008
Virtually Broken
The other day saw wired's VM just decide it didn't like executing anything useful and just eat CPU (we're talking much more than usual). So, I kill the server (you can't use it) and attempt to restart it.
Unfortunately this fails. Not because somebody screwed up with tar and root like last time but because the kernel module decided to unload itself. This was actually nothing to do with the crash but did give some rather concerning error messages.
If this is a one off, we have no problems but if it happens again, we should probably consider other options to KVM. We've tried Xen which seemed to "pause" execution and resume it after an arbritary period of time. As it didn't crash out, we had no error logs and could find no similar situations on the internet. So that option is also out.
VMWare has been toured as an option as well as OpenVZ. As we only have on VM, we could actually just run wired on the bear metal. The last option seeming the most stable as the host OS has not actually crashed out yet.
Unfortunately this fails. Not because somebody screwed up with tar and root like last time but because the kernel module decided to unload itself. This was actually nothing to do with the crash but did give some rather concerning error messages.
If this is a one off, we have no problems but if it happens again, we should probably consider other options to KVM. We've tried Xen which seemed to "pause" execution and resume it after an arbritary period of time. As it didn't crash out, we had no error logs and could find no similar situations on the internet. So that option is also out.
VMWare has been toured as an option as well as OpenVZ. As we only have on VM, we could actually just run wired on the bear metal. The last option seeming the most stable as the host OS has not actually crashed out yet.
Wednesday 1 October 2008
Numbers, Numbers, Numbers
[caption id="attachment_109" align="aligncenter" width="700" caption="One week of listener figures."]
[/caption]
It's very well known that STAR does not attract huge numbers of listeners. In fact, you'll be hard pressed to find a student radio station (online only) that does. In previous years, our listener numbers have been derived from parsing the Icecast logs with something along the lines of awstats (in fact it was usually awstats). This creates some rather high listener figures (up to x hundred listening at once) due to counting each "hit" in the log as a listener.
This year, we made use of MRTG polling the Icecast server. This provides us with rolling statistics and some rather pretty pictures. Below is the graph for Fresher's week 2008:
Not bad for a one week stint. We peaked about 12 listeners.and averaged about 1 through the night. So adjusting for the Union bar computer, that's 11 during live periods and nobody through the night. Let's hope these start to go up in the future.
[/caption]It's very well known that STAR does not attract huge numbers of listeners. In fact, you'll be hard pressed to find a student radio station (online only) that does. In previous years, our listener numbers have been derived from parsing the Icecast logs with something along the lines of awstats (in fact it was usually awstats). This creates some rather high listener figures (up to x hundred listening at once) due to counting each "hit" in the log as a listener.
This year, we made use of MRTG polling the Icecast server. This provides us with rolling statistics and some rather pretty pictures. Below is the graph for Fresher's week 2008:
Not bad for a one week stint. We peaked about 12 listeners.and averaged about 1 through the night. So adjusting for the Union bar computer, that's 11 during live periods and nobody through the night. Let's hope these start to go up in the future.
Subscribe to:
Posts (Atom)