Re: [ippm] Proposed changes to default settings in the Capacity metric load adjustment algorithm

[Martin Duke <[e-mail]>]

Hello IPPM,

If you have any comments about this change, please send them to this list
no later than 24 August.

On Tue, Aug 3, 2021 at 3:01 PM MORTON JR., AL <[e-mail]> wrote:

> IPPM,
>
> I want to point out that some recent results conducted by others
> encouraged us to test some additional values for seq error threshold and a
> wider value range of values for consecutive errored status (during feedback
> intervals).  We discovered that we would like to revise the default values
> to make the load adjustment a little less-sensitive to
> non-congestion-related loss and protect the fast ramp-up from ending
> prematurely. For example, a very short burst of lost packets spanning the
> boundary between two feedback intervals is currently sufficient to end fast
> ramp-up.
>
> The current table in 8.1 contains:
> +--------------+-------------+--------------+-----------------------+
> | Parameter    | Default     | Tested Range | Expected Safe Range   |
> |              |             | or values    | (not entirely tested, |
> |              |             |              | other values NOT      |
> |              |             |              | RECOMMENDED)          |
> +--------------+-------------+--------------+-----------------------+
> ...
> +--------------+-------------+--------------+-----------------------+
> | sequence     | 0           | 0, 100       | same as tested        |
> | error        |             |              |                       |
> | threshold    |             |              |                       |
> +--------------+-------------+--------------+-----------------------+
> | consecutive  | 2           | 2            | Use values >1 to      |
> | errored      |             |              | avoid misinterpreting |
> | status       |             |              | transient loss        |
> +--------------+-------------+--------------+-----------------------+
>
>
> The revised table rows would look like this:
>
> +--------------+-------------+--------------+-----------------------+
> | Parameter    | Default     | Tested Range | Expected Safe Range   |
> |              |             | or values    | (not entirely tested, |
> |              |             |              | other values NOT      |
> |              |             |              | RECOMMENDED)          |
> +--------------+-------------+--------------+-----------------------+
> ...
> +--------------+-------------+--------------+-----------------------+
> | sequence     | 10          | 0,1,5,10,100 | same as tested        |
> | error        |             |              |                       |
> | threshold    |             |              |                       |
> +--------------+-------------+--------------+-----------------------+
> | consecutive  | 3           | 2,3,4,5      | Use values >1 to      |
> | errored      |             |              | avoid misinterpreting |
> | status       |             |              | transient loss        |
> +--------------+-------------+--------------+-----------------------+
>
> Somehow, persistent non-congestion-related loss had mostly eluded our
> previous tests. But our goal has been to provide default values that will
> work in automated testing run by anyone in any implementation or
> circumstances - not just ~expert users and those who run the udpst utility.
>
> So we'd like to change default values for:
>
>   seq error threshold         from 0 to 10
>   consecutive errored status  from 2 to  3
>
> and update our tested range/values column as well.
>
> Al
>
>