On Wed, 2 May 2007 14:40:13 +0200 Jan Seeger <[EMAIL PROTECTED]> wrote:
> Ah yes sorry there. Then I cannot help you, since I have no experience > using netboot/pxe. Sorry > Greetings Jan Seeger Well, dave, I look to be one of the few who have done this. I think I can give you a hand. > 1. I can still compile/emerge on the frontend but its very slow as its > on NFS. Can I compile on the server by chrooting? Basicallly I want to > know if I can compile on EMT64 targeting AMD64. The frontend was > installed with stage 1 so I guess the compiler won't even run in a > chroot'ed environment. Various NFS tweaks can help with this. The biggest difference I noticed was allowing async NFS mounting -- and a substantial difference it was. During an emerge (genlop in this case) My sync-mounted router 'davey' gets about 2M or less per second on the server's hard drive, whereas the async-mounted 'slim' achieves speeds more on a reasonable scale of 5M or more; lag is significantly lower as well. After emerging I wanted to show what genlop said of the install speeds... | davey linux # genlop -i genlop | Global build time: 32 seconds. compared to | slim ~ # genlop genlop -i | Global build time: 7 seconds. I didn't think such a small package was a very good example, so a dependancy, DateManip, is also shown: | davey linux # genlop DateManip -i | Global build time: 7 minutes and 31 seconds. compared to | slim ~ # genlop -i DateManip | Global build time: 27 seconds. Here's the server export options from exportfs: | /nfsroot/davey 192.168.1.1(rw,wdelay,no_root_squash) | /nfsroot/slim 192.168.1.86(rw,async,wdelay,no_root_squash) And from fstab on the clients: | 192.168.1.87:/nfsroot/slim / nfs rw,async,rsize=32768,wsize=32768 0 0 | 192.168.1.87:/nfsroot/davey / nfs | noauto,wsize=1024,rsize=1024,sync,noatime 0 0 Although the async mount was scary at first, I haven't had any problems. slim is my media PC. >2. The Gensplash is compiled into the kernel. It appears at startup >with the 'Initialising kernel' screen but the progress bar never >appears. I'm not a gentoo spash user, but I'd wager you need to provide the splashscreen images and whatnot to the kernel. The kernel and any initrd/initramfs images must live on the TFTP server to be provided to PXE, and I would imagine the same thing would be true of a splashscreen. >3. My DHCP server issues static routes, but since I've made the >frontend boot over PXE (static IP address) it never gets the routes. >Any ideas? Contrary to popular opinion, there's nothing wrong with using dhcp to assign ip information -- At least, that is one of the few things that my net-booters haven't suffered from. My kernel invocation is something like | DEFAULT /bzImage.davey.2.6.19.new | APPEND ip=dhcp root=/dev/nfs nfsroot=192.168.1.87:/nfsroot/davey As you can see, I use PXE-Linux and I recommend it. (I would be interested to hear from you if you got PXE-Grub working because I never did.) the useful /usr/src/linux/Documentation/nfsboot.txt suggests that: | ip=<client-ip>:<server-ip>:<gw-ip>:<netmask>:<hostname>:<device>:<autoconf> | | This parameter tells the kernel how to configure IP addresses of |devices and also how to set up the IP routing table. This option might solve the problem, but I'd think you'd have better luck using dhcp as I do. Best of luck, Dan -- [EMAIL PROTECTED] mailing list