2

I'm looking into changing how to deploy network boot for clusters of servers. My worry is network transfer since not all boxes will be close to boot servers and network might be slow.

The machines have ordinary disks etc, we are storing data locally on them (database content etc).

Is there some mechanism to avoid copying full images if not necessary, like cache files locally and only validate checksums etc.?

Any pointers or ideas appreciated!

2 Answers 2

3

Let's assume you boot Linux boxes using PXE for loading kernel/initrd and NFS for mounting filesystems.

You can cache PXE files (configuration, images, kernel, initrd) on other tftp servers closer to machines which boot via PXE. Then you will only need to rsync when something changes.

For NFS filesystems, you can rsync them from time to time to local storage and use e.g. aufs to merge local and remote filesystems into one.

2

My opinion is, that PXE is part of firmware of the network card, and then it know nothing about disk, filesystem and so on. Then caching of booting files locally is not possible.

You can use extra network for booting and parallel network for computing, when you have more than one network card and you are worried about a slow network responses. But I think, that files for booting is not as big for normal network and therefore there will be no problem to transfer files needed for booting.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .