What is the limit? (was RE: multi-homing fixes)

Mailman List Mirror (Read Only)
1

> I have proposed to various router vendors the possibility of giving
> them a chilled water feed instead of lots of cool air. At the
> moment they seem to not need it, but I would not be surprized to
> find something like this needed at some point.

Err. Water and electricvity make a dangerous mix.

That sounds a bit FUDish--lots of electricity is generated using water
(most of it is, actually, given that steam is part of the cycle even in
most nukes).

And we have used water-cooled CPU's for a lot of years.

(And as a point of interest, distilled water at neutral ph is a
reasonably good insulator, if I recall correctly.)

2

(And as a point of interest, distilled water at neutral ph is a
reasonably good insulator, if I recall correctly.)

It has to be deionized, and yes, it is a very good insulator
if the purity of the deionization is high enough. But there are
better fluids that transfer heat better and will not pick up
free ions (making them conductive) from the metal tubing around them.
Alcohol cooling used to be a favorite for electronics. EthylGlycols are
good, but messy.

Maybe it's time to make some small point cooling systems for routers.
When I was young electronics nut, we'd cool stereo amps this way
to up the power output. As the last rack mount computer I installed
had 6 cooling fans.. maybe it's time...

3

APNICs approach (FYI):

Hi James,

Thanks for your mail and questions. Answers in-line below.

>http://www.apnic.net/meetings/12/docs/proposal-multihome-assign.html
>
>Are companies with existing historically assigned address space eligible for
>additional space via the small multi-homing assignment policy ?

Anyone is eligible.

>Should there be a limit of 1 of these assignments made per 'entity', with
>renumbering occurring if further address space is required ?

For clarification, the current proposal did not specify any limitation on
the number of such assignments.

What would be the motivation for this?

>If so should we consider reserving the next larger block for a period of
>time, to account for possible growth ?

No reservations will be made. It is generally assumed that applicants will
be paying a one-off service fee, rather than an ongoing membership based
fee. This is in fact identical to the way that PI assignments can currently
be obtained. The proposal restricts the assignments to (about to be)
multihomed orgs.

>If entity has a 'small multi-homing' assignment and later joins as a member
>and qualifies for a /20 PA allocation, are they required to renumber from
>the multi-homing assignment ?

Yes. The assumption would be that the organisation whose network has grown,
would then take the shorter portable prefix would then announce that prefix
instead of the longer one.

Maybe this should be made more explicit in the policy?

>Should 'small multi-homing' assignments be made from a specific (defined)
>netblock ?

I think there is a choice here. We can use "swamp" space, found mostly in
202/8 which by definition does not contain large ranges of address space,
or we can take a range from the less "swampy" space ie. 218/8 and use
that. My feeling is that it would be better to use the 202/8 range.

>There is significant risk in assigning greater than /20 space from existing
>APNIC allocated blocks given that operators often use RIR minium allocation
>guides for filtering, if the minimum assignment/allocation for all current
>APNIC blocks becomes /23 or /24 this may encourage the deaggregation of
>existing PA blocks.

I'm not sure I follow the logic. As Geoff pointed out, there is already
significant de-aggregation. Despite this, in de-aggregating you still run
the increased risk of exposure to more severe route flap dampening
parameters as well as filters. The objective in filtering is to limit the
size of the routing table, so I dont see this policy encouraging ISPs to
adjusting their filters to allow longer prefixes, if anything, rather the
other way round.

regards

Anne
APNIC

4

> > I have proposed to various router vendors the possibility of giving
> > them a chilled water feed instead of lots of cool air. At the
> > moment they seem to not need it, but I would not be surprized to
> > find something like this needed at some point.
>
> Err. Water and electricvity make a dangerous mix.

That sounds a bit FUDish--lots of electricity is generated using water
(most of it is, actually, given that steam is part of the cycle even in
most nukes).

yeah, although in steam or hydro turbines or expansion engines the
generator is connected via a drive-staft not immersed... water isn't as
decent an insulator as an air gap which has to be taken into account when
generating a few hundred thousand volts.

This is completely removed from the issue of immersing your computer in
it. any cooling system in a router or computer is going to be a balance
between complexity, thermal efficiency, and cost. we've hardly exhausted
to possibilities of air cooling at this point, that will occur when the
energy density of the individual componts is to high to be cooled by
airflow and reasonable sized heatsinks, not when the box collectivly draws
more power...

5

>Should there be a limit of 1 of these assignments made per 'entity', with
>renumbering occurring if further address space is required ?

For clarification, the current proposal did not specify any limitation
on the number of such assignments.

What would be the motivation for this?

I would say, to keep the tables from growing. Either have them renumber
into another 'larger' space or allow them to grow into a larger space.

micro allocations are a good thing (tm) if they are done right.

>If so should we consider reserving the next larger block for a period of
>time, to account for possible growth ?

No reservations will be made. It is generally assumed that applicants
will be paying a one-off service fee, rather than an ongoing
membership based fee. This is in fact identical to the way that PI
assignments can currently be obtained. The proposal restricts the
assignments to (about to be) multihomed orgs.

ok, if this is the case, have them renumber to keep the number of
announcements down.

>Should 'small multi-homing' assignments be made from a specific (defined)
>netblock ?

I think there is a choice here. We can use "swamp" space, found mostly
in 202/8 which by definition does not contain large ranges of address
space, or we can take a range from the less "swampy" space ie. 218/8
and use that. My feeling is that it would be better to use the 202/8
range.

right. but what ever space is used, publish it so those of us that have
filters can adjust for it. :slight_smile: