1 aa513bf8 2023-01-22 jrmu version=pmwiki-2.2.130 ordered=1 urlencoded=1
2 aa513bf8 2023-01-22 jrmu agent=Mozilla/5.0 (X11; OpenBSD amd64; rv:82.0) Gecko/20100101 Firefox/82.0
4 aa513bf8 2023-01-22 jrmu charset=UTF-8
6 aa513bf8 2023-01-22 jrmu ctime=1621396839
7 aa513bf8 2023-01-22 jrmu host=125.231.29.99
8 aa513bf8 2023-01-22 jrmu name=Vmm.Optimize
11 aa513bf8 2023-01-22 jrmu text=[@%0a11:53 %3c mlarkin> so jrmu the first thing I would do is crank host HZ to 1000 %0a or 2000%0a11:53 %3c jrmu> I'll take your word for it then%0a11:53 %3c mlarkin> that will make an immediate improvement%0a11:53 %3c mlarkin> leave the guests at 100%0a11:53 %3c jrmu> ok so I'd do that with the BIOS settings right?%0a11:53 %3c mlarkin> no%0a11:53 %3c mlarkin> param.h%0a11:54 %3c mlarkin> rebuild host kernel%0a11:54 %3c mlarkin> sorry param.c%0a11:54 %3c mlarkin> try 1000%0a11:54 %3c mlarkin> that usually always works%0a11:54 %3c mlarkin> my success rate with 2000 is more hit/miss%0a11:55 %3c mlarkin> and whatever you do, dont come to tech@/misc@ and complain %0a that HZ=1000 is breaking things.%0a11:55 %3c mlarkin> this is just "hallway conversation"%0a11:55 %3c brynet> I think Misha had a page for tuning things, talk with him. %0a Also would recommend checkout out Dave Voutila's vmm_clock %0a and virtio_vmmci drivers.%0a11:55 %3c jrmu> :) ok thank you%0a11:56 %3c brynet> https://github.com/voutilad%0a11:56 %3c mlarkin> HZ=1000 allows for more efficient servicing of guest clock %0a interrupts%0a11:56 %3c mlarkin> if 1000 works for a while with "lots" of vms, try 2000%0a11:56 %3c brynet> (If you're planning on Linux VMs, that is.)%0a11:57 %3c mlarkin> brynet: HZ=1000 helps with any guests%0a11:57 %3c brynet> right, meant dv@'s drivers. :)%0a11:57 %3c mlarkin> anything > 100%0a11:57 %3c mlarkin> ah%0a11:57 %3c mlarkin> yes%0a11:58 %3c mlarkin> cheloha@ is working on a better clock subsystem that will %0a fix a lot of this but it is a ways away still. to give you %0a an idea how hard that is, he has been continuously working %0a on it since 2019%0a11:58 %3c mlarkin> for now, cranking HZ is the only solution we have%0a11:58 %3c jrmu> thanks, that is interesting, I'll research into it more later%0a11:58 %3c jrmu> I'll put this in my notes%0a@]
12 aa513bf8 2023-01-22 jrmu time=1621397080
13 aa513bf8 2023-01-22 jrmu author:1621397080=jrmu
14 aa513bf8 2023-01-22 jrmu diff:1621397080:1621396839:=1d0%0a%3c [@%0a39d37%0a%3c @]%0a\ No newline at end of file%0a
15 aa513bf8 2023-01-22 jrmu host:1621397080=125.231.29.99
16 aa513bf8 2023-01-22 jrmu author:1621396839=jrmu
17 aa513bf8 2023-01-22 jrmu diff:1621396839:1621396839:=1,37d0%0a%3c 11:53 %3c mlarkin> so jrmu the first thing I would do is crank host HZ to 1000 %0a%3c or 2000%0a%3c 11:53 %3c jrmu> I'll take your word for it then%0a%3c 11:53 %3c mlarkin> that will make an immediate improvement%0a%3c 11:53 %3c mlarkin> leave the guests at 100%0a%3c 11:53 %3c jrmu> ok so I'd do that with the BIOS settings right?%0a%3c 11:53 %3c mlarkin> no%0a%3c 11:53 %3c mlarkin> param.h%0a%3c 11:54 %3c mlarkin> rebuild host kernel%0a%3c 11:54 %3c mlarkin> sorry param.c%0a%3c 11:54 %3c mlarkin> try 1000%0a%3c 11:54 %3c mlarkin> that usually always works%0a%3c 11:54 %3c mlarkin> my success rate with 2000 is more hit/miss%0a%3c 11:55 %3c mlarkin> and whatever you do, dont come to tech@/misc@ and complain %0a%3c that HZ=1000 is breaking things.%0a%3c 11:55 %3c mlarkin> this is just "hallway conversation"%0a%3c 11:55 %3c brynet> I think Misha had a page for tuning things, talk with him. %0a%3c Also would recommend checkout out Dave Voutila's vmm_clock %0a%3c and virtio_vmmci drivers.%0a%3c 11:55 %3c jrmu> :) ok thank you%0a%3c 11:56 %3c brynet> https://github.com/voutilad%0a%3c 11:56 %3c mlarkin> HZ=1000 allows for more efficient servicing of guest clock %0a%3c interrupts%0a%3c 11:56 %3c mlarkin> if 1000 works for a while with "lots" of vms, try 2000%0a%3c 11:56 %3c brynet> (If you're planning on Linux VMs, that is.)%0a%3c 11:57 %3c mlarkin> brynet: HZ=1000 helps with any guests%0a%3c 11:57 %3c brynet> right, meant dv@'s drivers. :)%0a%3c 11:57 %3c mlarkin> anything > 100%0a%3c 11:57 %3c mlarkin> ah%0a%3c 11:57 %3c mlarkin> yes%0a%3c 11:58 %3c mlarkin> cheloha@ is working on a better clock subsystem that will %0a%3c fix a lot of this but it is a ways away still. to give you %0a%3c an idea how hard that is, he has been continuously working %0a%3c on it since 2019%0a%3c 11:58 %3c mlarkin> for now, cranking HZ is the only solution we have%0a%3c 11:58 %3c jrmu> thanks, that is interesting, I'll research into it more later%0a%3c 11:58 %3c jrmu> I'll put this in my notes%0a
18 aa513bf8 2023-01-22 jrmu host:1621396839=125.231.29.99
