When I was testing the unfinished virtio-9p driver for FreeBSD and fixing it, it turned out to also have performance issues, with a kernel build inside the VM taking forever, spending the vast majority of the time contending on locks in the 9p driver. This post made me wonder if the QEMU host was itself contributing to the slowness, but as it’s from 2022, the patch has been merged a long time ago now..
This is a great post of how great software engineers are supposed to work. Find a problem then dig down to the bottom.
This new world where data scientists stop at sql, full stack stops at js is terrible because people make assumptions about underlying tech and build fragile high level workarounds instead bothering to find the deeper issues.
I am all for specialization, but don’t limit yourself by “not my area of expertise” cos eventually your area will go obsolete and you will be out of a career.
I’ll omit the details of the hour of frustration that ensued, but it turns out that the /proc which ps and friends were reading from was still that of the root PID namespace — even though they were no longer running in it! This might cause some funny unexpected behaviour when using tools like pkill… But that’s a problem for another day.
I believe this is because you have to remount procfs.
Oh… this is from 2022.
When I was testing the unfinished virtio-9p driver for FreeBSD and fixing it, it turned out to also have performance issues, with a kernel build inside the VM taking forever, spending the vast majority of the time contending on locks in the 9p driver. This post made me wonder if the QEMU host was itself contributing to the slowness, but as it’s from 2022, the patch has been merged a long time ago now..
Hi, author of the post here :)
It’s quite possible that there’s another easy win (or 3) for performance hanging out in the codebase. Try throwing a profiler at it! :)
Good perf write up! once again C is hamstrung by its lack of basic data structures
Aw c’mon, we got both data structures–integers and pointers!
Hope rants are ok here.
This is a great post of how great software engineers are supposed to work. Find a problem then dig down to the bottom.
This new world where data scientists stop at sql, full stack stops at js is terrible because people make assumptions about underlying tech and build fragile high level workarounds instead bothering to find the deeper issues.
I am all for specialization, but don’t limit yourself by “not my area of expertise” cos eventually your area will go obsolete and you will be out of a career.
I believe this is because you have to remount procfs.
Yes, the author said they did that in the very next sentence.
So they did