Re: [PATCH] trace-cmd: Do not use instance from trace context

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sat, Jul 30, 2022 at 4:00 AM Alexander Aring <aahringo@xxxxxxxxxx> wrote:
>
> Hi,
>
> On Fri, Jul 29, 2022 at 8:55 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >
> > On Fri, 29 Jul 2022 17:48:08 -0400
> > Alexander Aring <aahringo@xxxxxxxxxx> wrote:
> >
> > > >
> > > > Bah, I kept getting confused by when to use instance vs ctx->instance,
> > > > and I guess I messed this one up.
> > >
> > > I tested it and it seems to fix the problem..., so if it's not to late:
> > >
> > > Tested-by: Alexander Aring <aahringo@xxxxxxxxxx>
> >
> > Not too late. I haven't downloaded the patch from patchwork yet (nor my
> > other patches).
> >
> > >
> > > I am not sure what I should expect from the PTP time synchronization
> > > over IP capable interfaces (it never worked for me) but I need to say
> > > it is significantly slower than kvm time synchronization with vsock
> > > and I am using only virtual interfaces. On the agents I get a couple
> > > of:
> >
> > Well, kvm time synchronization doesn't do much between the host and
> > guest. And I'm looking at making it do even less. That's because the
> > kvm synchronization is just "read the offset and shift of the guest
> > from the host and do the calculations via the reader (trace-cmd
> > report or kernelshark)". But the P2P is sending packets back and forth
> > between the host and guest and trying to figure out the round trip
> > latency.
> >
>
> okay. To be more specific it took about ~20 minutes until I saw the
> "Press Ctrl-C to stop recording..." message for 3 agents. Is that
> normal?

It doesn't seem normal, usually takes less than a minute when I tested
with 2 agents (VMs running on the same host). PTP works by exchanging
a lot of packets over the network between the host and the agent, so
it depends on the network and on the guest. Currently the PTP logic
works sequentially - one agent at a time. That could be optimized, the
agents can be processed in parallel. But definitely 20 minutes for 3
agents (~7min per agent) is not normal.

>
> - Alex
>


-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center



[Index of Archives]     [Linux USB Development]     [Linux USB Development]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux