andyf isn't that dangerous ?

Any suggestions on how I could profile a 'zfs send' process (OmniOS) to see what the send is doing, where it is spending the more cpu time?

Being a Linux monkey and banging around in OmniOS having not touched Solaris since the 1990s is a hoot

Is there a way to push an OmniOS box to run processors at higher or highest p-state (clockspeed)? I'm watching powertop while running a 'zfs send' that is very slow and the cores never kick up beyond the lowest p-state (800Mhz). I've Googled my *ss off and I cannot find a command within OmniOS to stick a processor at a higher p-state. Linux has tuned-adm profiles as well as cpupower and even writing to sysfs cpu pointers.

aeonjeffj believe it or not, ChatGPT is pretty good. lemme ask GPT4

ChatGPT told me poweradm which is a Solaris command that doesn't appear to exist or be an addon package to OmniOS or IllumOS

same here

and there's pooladm

but not sure about that

maybe others know better.

Hello! Yesterday's "guest72" and "guest28" here, thought it'll be more consistent if I create an account. Sorry for asking some questions again, I tried to boot up the SCSI image on the virtual machine we talked about yesterday, but unfortunately I encountered some weird behavior I couldn't track down. After turning on verbose mode and booting the

system normally (option 1 in bootloader screen) instead of cloud-init initializing the instance, I got the x86_feature: list normally, then at the end this is the last message after which the system just hung without any further message, for about an hour when I killed it: "mem = 1042984K (0x3fa8a000)" since there are no error messages I couldn't

even get some ideas what could go wrong. Does anyone have any idea what could be done to figure out what's causing the problem? Thank you in advance!

hellllo

how is everyone? :)

I asked a few questions here yesterday. Didn't get answers but I eventually figured it out so I'm posting them here in case others come looking. ZFS send/recv over 25GbE link had terrible performance. Turns out OmniOS kernel was controlling PCU p-states and keeping clock speed at lowest, 800MHz. Removing that control by editing /etc/power.conf and changing "cpupm" to "disable" and adding /etc/system.d/zfs file with ZFS tuning

zfs_vdev_(async/sync)_(read/write)_max to a higher number equalling number of drives in the zpool tripled ZFS send/recv performance while using mbuffer as a network data pipe. /etc/power.conf: cpupm to disable, cpu-threshold to 10s. /etc/system.d/zfs: set zfs:zfs_vdev_async_read_max_active = 0xf (repeat for async/sync and write/read values).....and of course reboot to take effect.

Apologies if any of that is rudimentary to anyone here. I Googled and used ChatGPT4 and could not find this information.

aeonjeffj maybe you should blog about that :)

might

That feels like a surprising default and set of behavior about being kept low while in the cpupm mode. Probably something we should dig into a bit more.