On Nov 16, 2012, at 9:38 AM, Randy Bush wrote:

> first, thanks for getting a larger model started.
> 
> but i think many of the numbers are off.  i have had no time this last
> week, and will not for another week.  so i will pick on just three, as
> examples.

Great, thanks.

> 
> for bgpsec, i would assume a million routers each having the current
> signing cert and the next one, i.e. two million total.  scattered over
> O(100) repositories.

Really?  I didn't realize such a large estimate was the case.  Do others agree 
with the 1m.  Also, the standby cert is a good point, I hadn't factored that in.

> the best handle we have on rcynic time today is the ripe repository,
> 4,400 objects in 47 seconds, i.e. about .01 secs/obj.  and note that,
> thank you rsync, time is seriously sub-linear in object count.

I just used the numbers that I think you measured and presented at NANOG, etc.  
You had some sync performance graphs with average #objects and average sync 
time in the lower right/left corners.  I just used those, but that made things 
seem a little more bleak.

> and a roa has it's ee cert built in for free.  so there is a factor of
> two there.

I just put some text to the rationale behind calling this out separately into 
an email to Tim (on list).  Rather than re-type, maybe we can converse over 
that?

> i have not had time to look further, but i think, when numbers are a
> factor of two to a factor of six off, and you are multiplying, a bit
> more rigor is appropriate.


I'm totally happy to address the formulation.  It could, very well, be off.  
However, I think having such a formulation is critical, so can we talk about 
evolving what's there to something that lets us measure?

Thanks,

Eric
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to