North American Network Operators Group|
Date Prev | Date Next |
Date Index |
Thread Index |
Author Index |
Re: Your class 'B' address space
- From: Jeremy Porter
- Date: Tue Sep 29 12:53:05 1998
In message <firstname.lastname@example.org>, Austin Schutz writes:
>On Mon, Sep 28, 1998 at 05:11:42PM -0700, David R. Conrad wrote:
>>> If the reason for filtering small block of space is to keep small
>>>companies not generating a lot of traffic from polluting routing space I
>>>don't think this applies.
>>Traffic generation has absolutely no relation to prefix length filtering.
> Yes, but the question is should it? If a content provider generates
>a monstrous amount of traffic should they be forced to buy transit just
>because the traffic is generated by a small number of hosts?
No one is forcing you to buy transit. You are free to cut deals with
anyone you want. I'm sure if you paid the right people the right
amount of money, they would be happy to listen to a /32.
You can't expect everyone else to give you for free, resources in
their routers. If you pay a providerthey will be happy to anounce
a /24 for you. If you pay the other people whose resources you are
using, maybe they will listen? (Hint, Sprint will listen to a /24
from a customer, most other providers will also.)
>>The reason the filters are in place is because there are very few tools
>>available to limit the growth of routing tables and filtering based on
>>prefix length is a simple to implement unilateral mechanism that
>>(theoretically) encourages people to think about their actions prior to
>>flooding the routing system with long prefixes.
> Ok, but in this situation 'think about' = 'forbid'. In the vast
>majority of conditions it may make more sense for the organization to be
>dual homed with respect to a single NSP. On the other hand perhaps the current
>system encourages waste of address space by forcing organizations to claim at
>least a /21 of space if they wish to be given routable space by ARIN.
So its good enough for everyone else but not for you? This seems
a bit thin. People that multi-home often do it for having
more paths, single homed to a single NSP does not always increase
reliablity to do some of the failure modes being the same.
ARIN has already relaxed the /19 require to a /21, this is pretty
significant, I think. There are proposals to relax it more.
It has been stated by people here, and elsewhere, that as long as
route table growth is maintained at a reasonable rate, prefix length
filters might someday go away. I suspect of ARIN recommended
to ISPs to relax the filter length, they would look seriously
at doing so.
> What _should_ make space routable? Just network size? What about
>situations where a medium size dialup ISP wants to be multihomed? Perhaps
>they could theoretically use NAT to handle their dialup rather than
>have a pool of ip addresses but are encouraged not to because they don't
>want to jeapordize their eligibility for routable space. The same example
>could be used for web hosting companies who give each virtual server an ip
Just because they are using NAT doesn't mean they can't justify usage.
If they are at T-1 level or dual T-1 level, there are plenty of solutions
that work. The reocmmendation by the ARIN Advidsory Council that WWW
hosting companies not use multiple IPs when it becomes technologicly
reasonable to do so, has already been made.
> How about some of the many universities with a gargantuan amount of
>space used mostly for dorm rooms? Couldn't they use NAT and free up space?
>Would they want to given a smaller amount of space may not be routable?
I'm not sure ARIN should or could force universities to renumber.
A good number of those use DS3s which makes the NAT solution a non-starter
> It just seems like it is more reasonable to give a company like
>realaudio with a legitimate traffic generating resource the ability to
>multihome than penalize them for not wasting space.
You can get space from your upstream provider and announce it. If
your provider is especically nice they can even give you de-aggregated
space so you can inbound load balance, (if you are just generating
traffic, this is moot, because outbound load balancing is not
dependent on the prefixes announced out.) I don't see why
you can do this.
> I'm not advocating a return to the bad old days where everyone and
>their brother gets a class C, rather suggesting that the current system
>of preventing a route explosion may have unwanted side effects.
We are quite aware of the side effects, but who should carry the
cost of the increased resources used? What obivous metric
should be used to choose who lives and who dies? (Hint: if the
solution was easy, we would have already done it.)
>> I'd strongly recommend reviewing the old
>>CIDRD archives (wherever they might be).
> I'll look for them.
We've been round and round before on this, in Nanog also.
Jeremy Porter, Freeside Communications, Inc. email@example.com
PO BOX 80315 Austin, Tx 78708 | 512-458-9810