I'm on fedora core 8, and I may have a need to upgrade to the latest, v12, because of an issue I'm encountering (described in "some background", below, but my main query is here).
If I understand the issue I'm having correctly, what will fix it is a kernel ugprade. I've attempted to upgrade the kernel to latest (2.6.31 at the time) but I've run into some problems that will require troubleshooting, and given that I extremely rarely upgrade kernels, it could take me a while (fwiw, I tried the kernel ugprade on a vmware instance, don't know if that can lead to kernel upgrade problems that I otherwise wouldn't have).
So before going down that path, I would like to know:
1) If there are any chances that kernel compatibility with certain userland tools could be broken (once the kernel is correctly booting, that is) given the "large" version jump from 2.6.23 to 2.6.32 (might as well update to latest). For example, would some commands for traffic control (tc) not work anymore?
2) If it actually would be simpler & safer to just upgrade the distribution? (as long as the paths to the various network scripts haven't changed, mostly around interfaces, VLANs, etc)
In terms of keeping the various tools on the machine compatible with the kernel, I am tempted to go with #2. However, how safe is it to jump directly from fedora core 8 to fedora core 12? I guess most upgrades are tested from FC version N to N+1. I would jump 4 versions directly.
For what it's worth, I have an image of the machine running on vmware, so I could attempt the ugprade there first.
Any suggestions as to which path I should take? Isolated kernel upgrade or upgrade the distribution?
Some background (if you care about the context)
We use a fedora core 8 box as a router, mostly to shape traffic for network simulation purposes. We have many projects (games or games-related, actually) that use real-time network communication, and employ various network models. Via the use of VLANs, we allow for machines anywhere in our studio to hop on this router, at which point we apply selective traffic shaping rules. So, we have a large number of interfaces (due to the large number of VLANs), and in order to shape traffic bidirectionally, we also use the ifb device which adds even more interfaces. So:
- We have lots of interfaces: 52 (26 VLAN interfaces, 26 equivalent ifb interfaces)
- We mostly use the token buffer filter (tbf) and netem queueing disciplines
- We use the mirroring action to redirect ingress traffic from the regular VLAN interfaces to the IFB devices
Every now and then, when replacing/removing/adding qdiscs on the fly (and I think it only happens when a tbf qdiscs is involved), the kernel locks. No crash, just a lock, seemingly stuck in an endless loop.
There seems to have been some work done on around endless loops in the kernel network scheduler around nov 2006. Kernel 2.6.23 (which I have) was released around sept 2007. Maybe it takes that long to promote some changes from dev to release. Hopefully that's what I'm hitting. And hopefully by doing a kernel upgrade it will go away.