Summary Strategy This file is currently a mix of "which kernel to run" and "longterm computer strategy." 20 February 2005 Over Xmas, I set up the Opteron storage server (trevi) to
capture live television and to serve as a workstation for students in
my classes who need access to the captured material. Trevi thus
runs the following services:
This kind of distributed and removable storage is very attractive and I may get more of those enclosures -- they cost around $30 I believe, so you can literally store the files in these removable drives that can be plugged in at a moment's notice. Label each enclosure with the contents of the drive and pile them up. If all of them need to be connected, you can daisy-chain them on a fast firewire card -- this is really very attractive. The USB 2.0 drive enclosure kits are cheaper ($23 vs $39), but so far USB is messing up and Firewire is working. The only issue is file system -- but there are no good cross-platform alternatives, and ext2 is fairly fully implemented on OSX (ext2fsx lets you read, write, and format; there's a residual problem still with some USB drives) and readable on Windows (free explore2fs and ext2fsnt or proprietary ext2fs anywhere with free read-only support). You can also serve them though Samba or NFS on Linux. So don't worry about the file system; ext2 is good. I've now discovered that I can indeed daisy-chain the external firewire drives! That's a first for me. Having firewire is obviously a plus for interactivity / sharing with Macs. A problem remains that you can't reattach a drive once it's been removed, without rebooting the machine -- see firewire for details. The other machines currently have these functions:
Long-term plans:
26 November 2004 update The Opteron storage server was converted to Debian amd64 in September 2004 and runs the following services:
The other machines have the following roles:
Video lab design 19 November 2002: Now that you have a dynamic and simple system -- Debian, through Libranet -- you should plan to switch everything to this distribution and maintain several very similar machines. Check out Thing's instructions on HA, a heartbeat cluster method. There's also firewire networking (Oracle released some patches). Consult with Marsha Smith about furnishing the video lab -- you might even get some funds from Neil, though at the cost of opening it for others in the program. Anyway, ask Marsha, just for yourself -- Linux hardware, nothing complicated, ability to transcode and render, hooked up to TV and VCR, easy access on a list of tasks:
XFS and LVM If you want the changes present in pre10 _and_ xfs _and_ LVM, get the -aa kernel. Version pre10-aa2 has xfs included. You can get the original at http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/ or a copy at: http://giga.cps.unizar.es/~magallon/linux/kernel/. Narrative on 18 November, after the installation of Libranet 2.7 and Win98 on VMware The VMware solution was installed this morning, with Dreamweaver now directly available in Linux. I don't see any drawbacks -- the response time is excellent. This has a very solid feeling of a profound solution to the long-term project of setting up a good computer system. Narrative from 15 October 2002-- Computer plans It looks like 2.4.19-ac4 is fine. What I want is a laptop for my main work -- but if that's so, then you'd want to have it at home and take it in only occasionally, such as for teaching. That would mean that the laptop replaces my current laptop. Now, recall that with the laptop, you'll be able to use Dreamweaver in OSX, and also run Debian... This is fine, but not really my priority -- that is to say, it's my priority for teaching. So what you can do is migrate your windows box to Debian and run applications in OSX. What's left now is what? I think Agfa has an Apple printer driver, or you leave a Windows machine in the lab that runs it. Video is all better on Linux and Mac. MS Office is on OSX. OCR sounds like it's been ported to OSX, and Tim is getting a sheetfeeder. So what you do is move gubbio to the lab and convert Spello -- which will become the new gubbio -- to Debian. That should do it for now. Or -- you move gubbio home. Or you donate gubbio to UCLA LUG. Or you do this: remove the add-on CPU and replace the CD drive. Turn it into a dumb web server and leave it just for that -- it could even be configured to be a firewall? This is what you do. You don't need the CPU speed. You'll also need to peel off all the Windows stuff -- spend some time tidying that stuff up. The question is whether it makes sense to get a second machine right away -- a nice and quiet desktop for the office, with a 1GHz CPU that doesn't need a fan, and not much hard drive space -- just enough to run an OS, and then mount the rest to the lab. That way your three machines can all go to the lab, where they will be replaced over the next year or so and passed to UCLA LUG. This is the way to go. Get a TiBook and get a quiet PC, and move everything into the lab for now. In the lab, move towards putting Debian on everything, so the machines are as similar as possible. Get Codeweavers Office. Narrative from the installation of 2.4.20-preX in mid-August 2002 Note on recent patches: The 2.4.19 kernel may be fine -- I haven't tried to run it. The pre2 patch has a problem with CPiA, but otherwise runs fine. The pre3 patch doesn't boot. The pre4 patch may still have some difficulties with IDE -- it's worth waiting, perhaps until 2.4.20 is official. No point taking chances with your IDE drives! The main reason to get a new kernel (my current 2.4.16 is running great) is to get dv1394 for exporting back to camcorder. However, even this is frankly not a feature I urgently need. The main use would be for TV programs that you want off the hard drives. So it's worth getting, but there's no hurry. There is also a virtue in keeping a kernel for a long time -- it creates a stable and reliable system. Many Linux boxes are still running the 2.2 kernel -- in fact, this is still the default Debian stable kernel. 2.4.16, patched with the gigadriver patch from Andre Hendrick, seems to be an excellent kernel. The quality of kernels is uneven, and people in the know avoid some of them and favor others. For digital video, there is still work adding new features to the kernel, so this is worth getting once it's stabilized. The great thing about the current setup is that it's stable and works fine. It's really a production system -- I can take a tape, grab the content, compress it to divx, write two or three tapes to a DVD, and then delete the original dv to make room for more. In compressed form, an hour of video takes up about one Gb, so you can have a hundred hours on each hard drive (120 and 150 hours). This is not exaggerated, but it's also not too much -- you'll accumulate that amount of video. The originals you then keep on tapes, carefully marked. This is already a fully satisfactory system for your purposes. What you may also want to do is pick out interesting sequences for talks before you delete the dv files, since you cannot edit the mpeg4 files (at least not at the moment; kino may acquire the capability). Add this material to a separate file called something like video strategy and place it under the Research directory on grain -- you can link to and from the Linux pages.
|
|
|
|
|||||
Maintained by Francis F. Steen, Communication Studies, University of California Los Angeles |