Re: [PATCH v3 7/7] migration/dirtyrate: implement dirty-ring dirtyrate calculation

[Peter Xu <peterx@redhat.com>]

On Mon, Jun 07, 2021 at 09:15:20AM +0800, huangy81@chinatelecom.cn wrote:
> +static void calculate_dirtyrate_vcpu(struct DirtyRateConfig config)
> +{
> +    CPUState *cpu;
> +    int64_t msec = 0;
> +    int64_t start_time;
> +    uint64_t dirtyrate = 0;
> +    uint64_t dirtyrate_sum = 0;
> +    int nvcpu = 0;
> +    int i = 0;
> +
> +    CPU_FOREACH(cpu) {
> +        nvcpu++;
> +    }
> +
> +    dirty_pages = g_malloc0(sizeof(*dirty_pages) * nvcpu);
> +
> +    dirtyrate_global_dirty_log_start();
> +
> +    CPU_FOREACH(cpu) {
> +        record_dirtypages(cpu, true);
> +    }
> +
> +    DirtyStat.method.vcpu.nvcpu = nvcpu;
> +    if (last_method != CALC_DIRTY_RING) {
> +        DirtyStat.method.vcpu.rates =
> +            g_malloc0(sizeof(DirtyRateVcpu) * nvcpu);
> +    }
> +
> +    start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
> +    DirtyStat.start_time = start_time / 1000;
> +
> +    msec = config.sample_period_seconds * 1000;
> +    msec = set_sample_page_period(msec, start_time);
> +    DirtyStat.calc_time = msec / 1000;
> +
> +    CPU_FOREACH(cpu) {
> +        record_dirtypages(cpu, false);
> +    }
> +
> +    dirtyrate_global_dirty_log_stop();
> +
> +    for (i = 0; i < DirtyStat.method.vcpu.nvcpu; i++) {
> +        dirtyrate = do_calculate_dirtyrate_vcpu(i);
> +        DirtyStat.method.vcpu.rates[i].id = i;
> +        DirtyStat.method.vcpu.rates[i].dirty_rate = dirtyrate;
> +        dirtyrate_sum += dirtyrate;
> +    }
> +
> +    DirtyStat.dirty_rate = dirtyrate_sum / DirtyStat.method.vcpu.nvcpu;

Why you'd like to divide with nvcpu?  Isn't dirtyrate_sum exactly what we want?
As I don't think we care about average per-vcpu dirty rate, but total here.

> +    g_free(dirty_pages);
> +}

I did a run with 4G mem VM, alloc 1G and dirty it with 500MB/s, then

  - With old way: I got 95MB/s
  - With new way: I got 128MB/s

The new way has the output with:

Dirty rate: 128 (MB/s)
vcpu[0], Dirty rate: 0
vcpu[1], Dirty rate: 1
vcpu[2], Dirty rate: 0
vcpu[3], Dirty rate: 511

I think if without the division, it'll be 512MB/s, which is matching the dirty
workload I initiated.

-- 
Peter Xu