| 07:54.59 | *** join/##pxe Norky (n=dan@milton.ocsl.co.uk) |
| 08:58.14 | *** join/##pxe nami1 (n=test_acc@83-71-69-75.b-ras1.prp.dublin.eircom.net) |
| 09:03.58 | *** join/##pxe AndyS (n=tonline@p54AA6BDD.dip.t-dialin.net) |
| 09:29.19 | AndyS | Hi. I have played around with existing PXE installations before (Skolelinux), and now I wondered if it would be possible to solve a specific problem using PXE: At our institution (It's a school) we usually "clone" the harddrive of a PC to use that image to restore a broken PC. It works great using Knoppix and ntfsclone, and the stuff you actually have to do by hand (like, changing the computer name of the windows machine) is not very time consum |
| 09:29.20 | AndyS | ing. |
| 09:29.25 | AndyS | What I wondered about, though, is the possibility to automate the image creation and application process using PXE and a specialized linux which automatically does the image creation or application thing on startup. I would love to have a server application where I can choose a PC, automatically start it up, tell it to start (via PXE) the linux which does the image creation thing and shut it down again. Then I'd start up the "broken" PC (again |
| 09:29.25 | AndyS | via wake-on-lan), tell it to start (via PXE) the linux which does the image restoration thing and shut that one down, too, when the work has been done. |
| 09:29.33 | AndyS | From my understanding I'd need to have all PCs to boot via ethernet to make this work totally automatically, as I don't want to have to mess with BIOS settings every time a computer breaks (After all, it should be totally automatic). The normal machines would get a GRUB boot-image starting the windows on their local harddrive, whereas PCs needing to be fixed would be given one of the specialized cloning/restoration images. |
| 09:29.33 | AndyS | Now, the questions: |
| 09:29.33 | AndyS | 1. Can this be done? |
| 09:29.34 | AndyS | 2. Is it possible to have the PXE server tell the clients that they should not care about network booting this time so they just boot from their harddrive? (Thus making it unnecessary to bring GRUB into the game) |
| 09:29.37 | AndyS | 3. If 2. is not possible, how slow would you imagine such a system (like, booting grub to boot locally) to be if I only got a 100mbit line from each whole room of 18 PCs to the server. There'd be 10 rooms total, all connected via 100Mbit ethernet to a gigabit switch which the PXE server is connected to with gigabit lan. |
| 09:29.41 | AndyS | Thanks for your time. I'm very happy about partial anwers, too! |
| 10:56.07 | Norky | regarding 2, yes, look for the localboot option in the pxelinux config |
| 10:59.01 | *** join/##pxe namit (n=test_acc@83-70-43-68.b-ras1.prp.dublin.eircom.net) |
| 11:03.16 | Norky | AndyS, I'd suggest not relying on any kind of disk booting - otherwise what would happen if the disk bootsector were corrupted? Though I imagine you have already thought of this |
| 11:04.18 | AndyS | The disk booting is a necessarity here because the IT staff is not allowed to do anything Thin Client like you could say |
| 11:04.52 | AndyS | furthermore, our network isn't really good enough for it, either :) |
| 11:05.18 | AndyS | thanks for the answer to nr2 :) |
| 11:06.17 | AndyS | it's a bit of a strange way to utilize ethernet booting, but it would be very convenient and not a big deal to maintain if I weren't around anymore |
| 11:06.20 | Norky | I mean for recovery purposes |
| 11:07.06 | Norky | don't make your recovery procedure rely on working disk bootsector, though it sounds like you haven't |
| 11:07.25 | AndyS | ohh, now I get it |
| 11:08.09 | AndyS | that surely is a problem if I use ntfsclone for the imaging process |
| 11:08.15 | AndyS | haven't thought of it before, thanks |
| 11:08.39 | AndyS | I suppose I have to clone that one, too, then, using dd or something |
| 11:09.16 | AndyS | The machine which will be cloned will be of the "fully working" kind, hopefully :) |
| 11:09.50 | AndyS | but with a room of 18 identical PCs we have never run into the problem that no PC is working anymore, yet. |
| 11:09.52 | Norky | you could configure each machine's BIOS to boot off the network first. The default pxelinux configuration would be then just to localboot (hand control straight back to the disk) |
| 11:11.06 | Norky | then for broken machines have a different pxelinux in place which starts teh recovery procedure |
| 11:11.08 | AndyS | So practically the PC doing network booting would just ask the server "Do the default?" and if the server replies "Yes", no other network traffic would be necessary? |
| 11:11.25 | AndyS | ahh |
| 11:11.29 | Norky | that way all configuration could be done on the server, |
| 11:11.53 | AndyS | disregard the last sentence |
| 11:12.22 | AndyS | but then I'd have to transfer the pxelinux image everytime on every machine booting, right? |
| 11:12.42 | Norky | or you can set up an simple menu system (syslinux project includes 2 menu systems) that give you the option of "boot local harddisk" and "restore this machine from an image" |
| 11:12.53 | Norky | yes, but it's tiny |
| 11:13.21 | AndyS | I see |
| 11:14.31 | AndyS | Though as full automation and very limited network traffic is my aim, I'm not sure if this is the best solution for me |
| 11:14.34 | Norky | alternatively, some PCs (HP machines for example) can be made to boot from teh network just by pressing a single key at the POST display |
| 11:14.50 | AndyS | yeah, most pcs can do that here I believe |
| 11:15.19 | AndyS | but that would require physical presence |
| 11:15.29 | Norky | indeed |
| 11:15.58 | Norky | how are teh machine normally left? At the logon prompt or powered off? |
| 11:16.07 | AndyS | can I tell the pxe deamon on the server to tell the clients to do a harddrive boot? |
| 11:16.21 | AndyS | powered off |
| 11:16.31 | Norky | yes, as I said, the "localboot" option |
| 11:17.23 | AndyS | I'm a little confused |
| 11:17.23 | Norky | http://syslinux.zytor.com/faq.php search for localboot in there |
| 11:17.33 | AndyS | pxelinux is a way to boot linux over the network, right? |
| 11:18.09 | AndyS | and then there's this pxe daemon running on the server, which tells the clients to load pxelinux |
| 11:18.10 | Norky | pxelinux is a network-based bootloader, primarily for booting Linux, buit it can boot all kinds of other stuff now |
| 11:18.44 | AndyS | I was wondering if the pxe daemon can tell the clients to boot locally directly, without using pxelinux at all |
| 11:18.46 | AndyS | ahh |
| 11:18.55 | AndyS | well, I suppose it's small enough |
| 11:19.25 | AndyS | thanks alot :) |
| 11:20.05 | Norky | erm, I suppose you could have the DHCP server not give a "filename" option to machines that you want to locally boot, though that may confuse their PXE stacks. |
| 11:20.34 | AndyS | completely different question: there isn't something like "wake-on-lan-and-boot-from-ethernet", right? |
| 11:21.07 | AndyS | hmm, I should try that |
| 11:21.22 | Norky | erm, some BIOSes I have seen will, or can be configured to boot from teh network first if, and only if, they have been WoL'ed |
| 11:21.51 | Norky | but this is a feature of certain BIOSes, it's not part of the WoL spec |
| 11:22.47 | AndyS | I suppose that not all PCs we own/we will every buy will have that feature, but that's pretty good to know :D |
| 11:23.07 | AndyS | you've helped me alot, thanks again :) |
| 11:24.11 | Norky | consistently controlling the booting of ordinary desktop PCs relies on consistency of behaviour and features of those PCs, i.e. it'll never happen because PC BIOS vendors are very inconsistent |
| 11:24.50 | Norky | things are better if all PCs you have are recent models from the same big name vendor |
| 11:25.27 | AndyS | yeah, but nowadays it's not so easy to buy a model that will be available in the next 5 years |
| 11:25.59 | AndyS | and we would quite certainly like newer ones much better |
| 11:26.22 | Norky | but even then, you're often at the mercy of whatever your mobo vendor chose to pay the BIOS writer for and how much coffee the BIOS engineers were drinking ;) |
| 11:26.43 | Norky | not quite what I mean |
| 11:27.17 | AndyS | yeah, but a possible, if bad, solution to the inconsistency |
| 11:27.52 | Norky | in recent years the big vendors have been making desktop machines that all tend to have similar BIOS features, so any, say, IBM desktop machine bought in the last 4 years should behave pretty similar to any other |
| 11:28.21 | Norky | I'm talking in generalities of course |
| 11:28.22 | AndyS | it would actually be possible, after all we are only a school so we don't have many programs that need fast PCs |
| 11:28.33 | AndyS | I see |
| 11:29.15 | AndyS | but as this school focuses primarily on teaching programming it's not really nice to have to wait 5 minutes for visual studio to come up |
| 11:31.12 | Norky | I'm not suggesting that you go out and buy a bunch of new machines, just to hopefully save yourself a bit of struggling with a handful of machines that behave slightly differently to the rest of your 180 |
| 11:31.47 | AndyS | yeah, although that problem would solve itself in about 10 years if we keep by that buying sheme |
| 11:32.13 | Norky | just warning you that you might see a few differnt machines that make completely automated operation impossible :) |
| 11:32.32 | AndyS | always getting HP machines instead of some cheaply put together ones done by a local company |
| 11:33.12 | AndyS | It's true, you can't even be safe from differences between the same models |
| 11:33.20 | AndyS | we recently got 18 new pcs |
| 11:33.43 | Norky | HP have been guilty of a few cheap and cheerful; machines in their low-end line, but generally they are pretty good, you can rely on a dc7100 being exactly the same as any other dc7100 |
| 11:33.44 | AndyS | and one of them always starts automatically when the energy is switched on in the room |
| 11:33.48 | AndyS | the other 17 ones don't do anything |
| 11:34.10 | Norky | that is a BIOS-configurable option on most server-class hardware |
| 11:34.19 | AndyS | at first I thought it was a different bios setting, but I compared two and they are the same |
| 11:34.49 | Norky | I work for a company that sells HP machines, though usually servers and HP9000 and itatium kit |
| 11:35.20 | Norky | it might well be a NVRAM setting, just not one that the BIOS-configuration routine exposes to the user |
| 11:35.57 | Norky | or possibly a different revision of PSU or mobo |
| 11:36.11 | AndyS | could be, I haven't looked into it that much |
| 11:36.14 | Norky | what manufacturer are these 18 machines? |
| 11:36.19 | AndyS | I guess it has something to do with the PSU |
| 11:36.31 | AndyS | local manufacturer, ASUS motherboard I think |
| 11:37.01 | AndyS | they build them themselves |
| 11:37.13 | Norky | as soon as the PSU has power, it should be supplying a small charge to the motherboard (standby), the motherboard can then 'decide' to power on the machine, or not |
| 11:38.58 | AndyS | so it could only be the PSU if the other 17 PCs didn't get standby power |
| 11:39.13 | AndyS | but they do, so it has to be a mainboard issue |
| 11:42.34 | AndyS | It will be pretty cool if this automatic cloning thing using PXE works good enough, maybe I'll make a website/wiki entry about it |
| 11:43.24 | AndyS | possibly even with the needed slax linux images doing the automatic cloning/recovering thing |
| 11:46.06 | Norky | have you heard of g4u? (ghost for Unix) |
| 11:47.16 | Norky | I've not used it myself, it's a cloning tool like norton ghost, I believe it copes with NTFS volumes, and will clone the bootsector and partition table not just the filesystem |
| 11:50.19 | AndyS | nope, haven't heard of it |
| 11:50.38 | AndyS | but ntfsclone is pretty nice (it's in the ntfsutils I think), and nearly does the same |
| 11:50.51 | AndyS | for the MBR I can just use DD as it's always at the same place I guess ^^ |
| 11:53.18 | AndyS | I should look into it, though, maybe it does exactly what I want |
| 11:53.24 | AndyS | thanks for the pointer |
| 11:59.08 | Norky | actually, I have an idea that windows XP doesn't write a bootsector to the MBR, just to the start of its partition, which shoudl be enough if the partition table has the partition marked as bootable, or at least it doesn't need to have the bootsector in the MBR, which is how dual-booting where GRUB chainloads NTLDR is abel to work |
| 11:59.19 | Norky | but anyway, I digress |
| 12:02.56 | AndyS | this whole bootloader thing is pretty confusing in my opinion, but I suppose it's normal nowasays to have the real software on the partition itself and to only have a pointer to it in the MBR |
| 12:03.03 | AndyS | *nowadays |
| 12:38.04 | *** join/##pxe AndyS2 (n=tonline@p54AA6BDD.dip.t-dialin.net) |
| 14:23.20 | *** join/##pxe highvoltage (n=jonathan@ubuntu/member/highvoltage) |
| 14:33.13 | highvoltage | anyone around? |
| 15:17.19 | *** join/##pxe nami1 (n=test_acc@83-70-163-211.b-ras1.prp.dublin.eircom.net) |