[cap-talk] Why is EQ so dang fascinating?

[Dean Tribble]

> >In typical cases, the
> >operation completes in bounded time for cooperating participants, and
> >only takes unbounded time for malicious participants. With
> >non-blocking send, that need pose no burden on the evaluator
>
> It imposes the burden of choosing the right timeout, which is
> notoriously difficult.

Indeed, it's sufficiently difficult, that I was not suggesting a
timeout. Instead, in previous scenarios, only success mattered, and
success requires forward progress. A malicious participant not
responding is equivalent to them not having joined the game in the
first place (e.g., not having submitted a grant for the grant
matcher). (A side note: Any issue of outstanding resources is subsumed
in the resources the target is spending to have open connections to
the malicious participants.)  Designing protocols to be robust against
high latency often adequately addresses the issue of malicious
participants being deliberately slow.