-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 12 Apr 2006, at 14:27, Matthew Toseland wrote:

> The question posed in "Subject: [freenet-dev] Is a failed insert worth
> more than a failed request?" still stands.
>
> I suggest that:
> - Not all failed requests are equal.
> - *The* distinguishing factor is the number of RejectedOverload's
>   forwarded to the originator.
> - For example, an insert is more likely to produce several
>   RejectedOverload's than a request, but they are both counted equally
>   as a failure.
> - We should therefore count a single RejectedOverload as a failure -
>   even if the request is automatically retried by a later node and
>   succeeds. We can then count that success as a success. This is
>   analogous to packet retransmission; if the retransmit is lost as  
> well
>   as the original packet, it's reasonable to reduce the send rate  
> again.
>   However we differ in that the eventual successful resend is  
> counted as
>   a success... maybe we shouldn't?
> - We should still take the round trip time as the overall time for the
>   request (assuming it eventually succeeds rather than timing out).
>
> Is this reasonable?

Well, I think you are focussing on a sub-problem, when it is the meta- 
issue that is currently being debated: Is our TCP-inspired approach  
appropriate for request-level load limiting?

Ian.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2.2 (Darwin)

iD8DBQFEPX9TQtgxRWSmsqwRAneoAJ9pWOncYyXldyPrAHkmbFB7Zni+CQCfTyyr
3iI0YSRTjo8G5ESZwEzmMGI=
=be1l
-----END PGP SIGNATURE-----

Reply via email to