Let me share my performance report after applying the patches for the
information:

1. live mlx VF migration

outgoing migration:
+------------------+---------------+---------------+----------------+
| VF(s) number     | 1             | 2             | 4              |
+------------------+---------------+---------------+----------------+
| Memory bandwidth | 733.693 MiB/s | 556.565 MiB/s | 475.310 MiB/s  |
| Total downtime   | 227ms         | 358ms         | 460ms          |
+------------------+---------------+---------------+----------------+

incoming migration:
+------------------+---------------+---------------+----------------+
| VF(s) number     | 1             | 2             | 4              |
+------------------+---------------+---------------+----------------+
| Memory bandwidth | 738.758 MiB/s | 566.175 MiB/s | 458.936 MiB/s  |
| Total downtime   | 220ms         | 342ms         | 459ms          |
+------------------+---------------+---------------+----------------+


2. live mlx VF multifd migration

outgoing migration:
+------------------+---------------+----------------+
| VF(s) number     | 1             | 1              |
+------------------+---------------+----------------+
| Channel          | 4             | 6              |
| Memory bandwidth | 786.942 MiB/s | 848.362 MiB/s  |
| Total downtime   | 142ms         | 188ms          |
+------------------+---------------+----------------+

+------------------+----------------+---------------+----------------+
| VF(s) number     | 2              | 2             | 2              |
+------------------+----------------+---------------+----------------+
| Channel          | 4              | 6             | 8              |
| Memory bandwidth |  774.315 MiB/s | 831.938 MiB/s | 769.799 MiB/s  |
| Total downtime   | 160ms          | 178ms         | 156ms          |
+------------------+----------------+---------------+----------------+

+------------------+----------------+---------------+----------------+
| VF(s) number     | 4              | 4             | 4              |
+------------------+----------------+---------------+----------------+
| Channel          | 6              | 8             | 16             |
| Memory bandwidth |  715.210 MiB/s | 742.962 MiB/s | 747.188 MiB/s  |
| Total downtime   | 180ms          | 219ms         | 190ms          |
+------------------+----------------+---------------+----------------+

incoming migration:
+------------------+---------------+----------------+
| VF(s) number     | 1             | 1              |
+------------------+---------------+----------------+
| Channel          | 4             | 6              |
| Memory bandwidth | 807.958 MiB/s | 859.525 MiB/s  |
| Total downtime   | 150ms         | 177ms          |
+------------------+---------------+----------------+

+------------------+---------------+---------------+----------------+
| VF(s) number     | 2             | 2             | 2              |
+------------------+---------------+---------------+----------------+
| Channel          | 4             | 6             | 8              |
| Memory bandwidth | 768.104 MiB/s | 825.462 MiB/s | 791.582 MiB/s  |
| Total downtime   | 170ms         | 185ms         | 175ms          |
+------------------+---------------+---------------+----------------+

+------------------+---------------+---------------+----------------+
| VF(s) number     | 4             | 4             | 4              |
+------------------+---------------+---------------+----------------+
| Channel          | 6             | 8             | 16             |
| Memory bandwidth | 706.921 MiB/s | 750.706 MiB/s | 746.295 MiB/s  |
| Total downtime   | 174ms         | 193ms         | 191ms          |
+------------------+---------------+---------------+----------------+

Best Regards,
Yanghang Liu


On Mon, Feb 3, 2025 at 10:20β€―PM CΓ©dric Le Goater <c...@redhat.com> wrote:
>
> Hello Maciej,
>
> > This patch set is targeting QEMU 10.0.
> >
> > What's not yet present is documentation update under docs/devel/migration
> > but I didn't want to delay posting the code any longer.
> > Such doc can still be merged later when the design is 100% finalized.
> The changes are quite complex, the design is not trivial, the benefits are
> not huge as far as we know. I'd rather have the doc update first please.
>
> Thanks,
>
> C.
>
>
>


Reply via email to