On Sat, Oct 29, 2016 at 05:12:51PM +0200, Jonathan Schleifer wrote: > Another thing I noticed: > > When running dd if=/dev/zero of=foo bs=65536, my SSH connection gets extremely > laggy. If I open 4 more in parallel, all go down to KB/s of writes, and SSH > becomes unusable. Now unusable as in things need forever to start. Unusable as > in I press a key and it takes forever to print that letter. > > Is this supposed to be like this, or is here something seriously wrong? Is > OpenBSD only using a single core for the kernel and thus once that core is > busy, the entire system starts to become unusable slow?
I run a T1000 which is segregated into a couple of LDOM guests (about 10). Some of the guests use softraid crypto inside. The host does not. Disk i/o is slow across the board, and when one guest upgrades (extracts sets) or makes builds, the other guests experience slowed down i/o as well. It's fine for network-bound tasks but I would not run a database or mail server on it, if that's what you were planning to use this box for. In the near term I am planning to replace the disks with SSDs to see if that helps. This requires the 2.5" disk frame which isn't easy to track down, or some self-built disk frame which holds a 2.5" disk. I've never used this system with just the host. The OpenBSD kernel is mostly giant-locked so having many CPUs for one kernel doesn't make sense. If you're running your tests without any LDOM guests configured, your system is probably using something in the order of 32 CPUs. In which case reducing the number of CPUs on the host by assigning some CPUs to guests might help a bit. See 'man ldomctl' for details about setting up guests.
