schoebel
commented
4 years ago Hi Frederik,
much of your question is answered by my last presentation at LCA 2020 (the Kernel Miniconf presentation, not the Sysadmin Miniconf one) this spring, immediately before Corona made travel to Australia / NZ impossible. AFAIK the videos from both my presentations are in the LCA archives, and on Youtube. Look for the right presentation about kernel upstreaming.
Yes, I can fully understand your point. It is even on my 1&1-internal roadmap, but so far many other high-workload tasks are taking precendence all the time, and I am alone (one-man-show). Pretty much a similar situation to your company.
What you or others could do in the meantime: create and maintain rpm and dpkg packages for the major Linux distributions, e.g. in the OBS or similar systems (I don't have time for such tasks). This would be appreciated by many people I know. AFAIK up to now nobody has invested the time to maintain both a pre-patched kernel and a MARS module package (or both in a single package, which would be easier to install).
I hope to get the time for kernel upstreaming some day.
Cheers,
Thomas
fgbreel
FelixSchwarz
Hi @schoebel while migrating an old Proxmox Installation with DRDB 8.4 to a six years newer Proxmox release (6.2) for one of our SME clients and after having almost entirely grey hair on finding a reasonable solution with DRBD 9 and Proxmox 6 i all of a sudden passed by your repository and work on MARS.
Great work and absolutely amazing that it runs that big workloads over at 1&1 production servers. Even if we finally decided to go with DRBD 9 in our current case, MARS seems to be exactly the use case one would search for if he runs HA nodes that are distributed in broader distributed networks without direct GBit connections. However, one major showstopper to me always is if there is no reliably addressable source of recompiled binary packages (officially package repository or PPA) but "only" a properly described way to compile it your own.
Now to my question: Wouldn't it be of pretty high importance also for 1&1 to get the entire work upstream and maintained within the scope of the ordinary kernel update processes? Did you or you team mattes ever attempt pushing or say proposing your work upstream ? I know that this would probably fail at the first attempt and having the restively few adopters it has by now, but looking at the technology and the architecture itself that would make a whole lot of sense to me technology-wise and it seems to fill a reasonable gab between lowlevel local technologies as DRBD and higher level distributed software defined storages with CEPH.
Unfortunately at my company we are not really working in the core C development of the Kernel and also lack expertise and seriously large workloads that would make up a good test bed for your technology, but still i wanted to just post my admiration for you having pulled this off. Great job and hope to see that merged to mainline one day! I hope your employer admires your work as well at gives you freedom trying to propose it upstream.
Best Frederik