[9fans] interesting qemu problem

[9fans] interesting qemu problem

[ron minnich]

I'm now running an upgraded qemu:
QEMU PC emulator version 0.11.0 (qemu-kvm-0.11.0), Copyright (c)
2003-2008 Fabrice Bellard


And have noticed that an old image I use for qemu is going astray.
Same kernel as it has been for quite some time, but the load is pegged
at about 2500 at all times.

I'm not sure where to start; any suggestions here?

I'm using qemu in this case because I don't quite have a working tiny
arm board and I want to beat on the kernel source proper for
something.

thanks

ron

Re: [9fans] interesting qemu problem

[erik quanstrom]

> And have noticed that an old image I use for qemu is going astray.
> Same kernel as it has been for quite some time, but the load is pegged
> at about 2500 at all times.

2.5, i assume.  what are the 2-3 processes running?

- erik

Re: [9fans] interesting qemu problem

[ron minnich]

On Tue, Mar 16, 2010 at 9:48 AM, erik quanstrom <quanstro@quanstro.net> wrote:
>> And have noticed that an old image I use for qemu is going astray.
>> Same kernel as it has been for quite some time, but the load is pegged
>> at about 2500 at all times.
>
> 2.5, i assume.  what are the 2-3 processes running?

Never mind, this fools me every time. It was so long since I booted
this thing that it had to do a snap, and after a replica -s, well,
that takes a long time.

ron

Re: [9fans] interesting qemu problem

[ron minnich]

I was wrong. I built a new kernel from sources and performance is
still very bad, with a load of 2500 minimum.

Also, venti, on this little machine, is a bit hungry for memory.
venti...2010/0316 20:31:06 venti: conf.../boot/venti: mem 1,048,576
bcmem 140,753,578 icmem 211,130,368...httpd
tcp!127.1!8000...init...icache 211,130,368 bytes = 3,298,912 entries;
16 scache
sync...2010/0316 20:31:12 arenas01: indexing 430 clumps...
announce tcp!127.1!17034...serving.
fossil(#S/sdC0/fossil)...fsys: dialing venti at tcp!127.1!17034


This image formerly ran in 256M, now requires 512M, because venti
footprint is 140+211+211 ... wait, how does it ever fit in 512 anyway.

I think qemu changed too. An old kernel has the same problems. Not
sure what is up here. but for now qemu on unongo 9.10 is useless for
plan 9 from what I can see.

ron

Re: [9fans] interesting qemu problem

[erik quanstrom]

> This image formerly ran in 256M, now requires 512M, because venti
> footprint is 140+211+211 ... wait, how does it ever fit in 512 anyway.

swap?  this would answer two questions.

- erik

Re: [9fans] interesting qemu problem

[ron minnich]

On Tue, Mar 16, 2010 at 3:07 PM, erik quanstrom <quanstro@quanstro.net> wrote:
>> This image formerly ran in 256M, now requires 512M, because venti
>> footprint is 140+211+211 ... wait, how does it ever fit in 512 anyway.
>
> swap?  this would answer two questions.

I should mention that another person here tried qemu recently and
commented that it was dog slow as well.

Something changed in qemu I think and it's affecting plan 9. That was
a very old qemu image and it was peppy in the old days.

ron

Re: [9fans] interesting qemu problem

[Eric Van Hensbergen]

On Mar 16, 2010, at 6:04 PM, ron minnich wrote:

> I was wrong. I built a new kernel from sources and performance is
> still very bad, with a load of 2500 minimum.
> 
> Also, venti, on this little machine, is a bit hungry for memory.
> venti...2010/0316 20:31:06 venti: conf.../boot/venti: mem 1,048,576
> bcmem 140,753,578 icmem 211,130,368...httpd
> tcp!127.1!8000...init...icache 211,130,368 bytes = 3,298,912 entries;
> 16 scache
> sync...2010/0316 20:31:12 arenas01: indexing 430 clumps...
> announce tcp!127.1!17034...serving.
> fossil(#S/sdC0/fossil)...fsys: dialing venti at tcp!127.1!17034
> 
> 
> This image formerly ran in 256M, now requires 512M, because venti
> footprint is 140+211+211 ... wait, how does it ever fit in 512 anyway.
> 
> I think qemu changed too. An old kernel has the same problems. Not
> sure what is up here. but for now qemu on unongo 9.10 is useless for
> plan 9 from what I can see.
> 


You could configure venti to be less aggressive with its use of memory, but that would likely hurt performance.

Running venti inside qemu is silly.  If you really want venti for your vm, run venti on the host and target your in-vm fossil at it. 

       -eric

Re: [9fans] interesting qemu problem

[ron minnich]

On Tue, Mar 16, 2010 at 3:47 PM, Eric Van Hensbergen <ericvh@gmail.com> wrote:

> You could configure venti to be less aggressive with its use of memory, but that would likely hurt performance.
>
> Running venti inside qemu is silly.  If you really want venti for your vm, run venti on the host and target your in-vm fossil at it.


I tend to disagree. If I'm running qemu it is because I want to
simulate a whole-machine environment. If I don't need that simulation,
I'll go back to 9vx.

but YMMV.

ron

Re: [9fans] interesting qemu problem

[erik quanstrom]

> I tend to disagree. If I'm running qemu it is because I want to
> simulate a whole-machine environment. If I don't need that simulation,
> I'll go back to 9vx.

seems that keeping up with qemu is at least as hard
as keeping up with real hardware.

- erik

Re: [9fans] interesting qemu problem

[Eric Van Hensbergen]

On Mar 16, 2010, at 6:52 PM, ron minnich wrote:

> On Tue, Mar 16, 2010 at 3:47 PM, Eric Van Hensbergen <ericvh@gmail.com> wrote:
> 
>> You could configure venti to be less aggressive with its use of memory, but that would likely hurt performance.
>> 
>> Running venti inside qemu is silly.  If you really want venti for your vm, run venti on the host and target your in-vm fossil at it.
> 
> 
> I tend to disagree. If I'm running qemu it is because I want to
> simulate a whole-machine environment. If I don't need that simulation,
> I'll go back to 9vx.
> 
> but YMMV.
> 

Yeah, but Plan 9's a cluster environment, nothing wrong with the venti server being elsewhere (in fact, thats kind of expected) -- unless of course you are debugging the venti server.

      -eric

Re: [9fans] interesting qemu problem

[ron minnich]

On Tue, Mar 16, 2010 at 4:04 PM, Eric Van Hensbergen <ericvh@gmail.com> wrote:

> Yeah, but Plan 9's a cluster environment, nothing wrong with the venti server being elsewhere (in fact, thats kind of expected) -- unless of course you are debugging the venti server.

I'm using qemu to debug a problem I'm having on a real machine. So,
yeah, I need to do this.

I don't see why venti has gotten so memory hungry, this seems new
behavior. I realize I can twist the knobs myself but geez, this is a 4
GB disk -- why does it think it needs nearly 400 MB RSS to deal with
it?

ron

Re: [9fans] interesting qemu problem

[blstuart]

> I should mention that another person here tried qemu recently and
> commented that it was dog slow as well.
>
> Something changed in qemu I think and it's affecting plan 9. That was
> a very old qemu image and it was peppy in the old days.

I wonder if it's a 0.11 thing or maybe a Linux thing.  I've been
running 0.12.3 on the Mac lately and haven't noticed it being
particularly slow.  It's got the clock weirdness it's always had,
but things seem to run okay.  Though I haven't been running
venti, so I suppose there could be some issue with heavy memory
usage.

BLS

Re: [9fans] interesting qemu problem

[Georg Lehner]

ron minnich wrote:
> [..]
> I don't see why venti has gotten so memory hungry, this seems new
> behavior. I realize I can twist the knobs myself but geez, this is a 4
> GB disk -- why does it think it needs nearly 400 MB RSS to deal with
> it?
>
> ron
Please note that quite a lot of installation problems mentioned recently
on the
mailing list have been related to venti beeing very memory hungry.

Without beeing able to contribute technically to a solution i propose to
mitigate
it by adding support for setting up venti parameters in the installation
scripts,
or by setting safe values by default and leave performance tweeking to the
experienced.

Regards,

    Jorge-Le�n