At 07:31 PM 2/26/97 -0600, Karl Denninger wrote:
As long as a provider can get their own /19 I have no problem with prefix filtering at the /19 level.
The problem comes about when big ISPs filter at /19s *AND* the allocators of space refuse to give ISPs /19s.
These two goals seem to be at odds in the current system for address allocation. How would you change the system to allow people to aquire address space that they need and get it routed? The address allocation scheme is geared towards trying to promote utilization of IP space, thus the sorta "take just what you need" methodology. The filters that you talk of seem to me to be crude proxies for controlling routing space on a particular providers network, this seems to me to be a reasonable thing (i.e. they have to make their network work). If different providers were to sell routing "slots" on their network such that an ISP could guarantee that their announcements would be accepted (regardless of address length) this would seem to solve the problems of both those that can't "justify" a big block and those of the providers that want to control the use of their resources on their network as well. It appears that you're primary argument is one of fairness and level playing field for all comers regardless of size, and I think this is a worthy goal if it can be done technically. -scott
Shouldn't the big boys (the ones who started all of this filtering) and the InterNIC be forced to come up with a fairer solution? At least if they don't do so voluntarily?
At 07:31 PM 2/26/97 -0600, Karl Denninger wrote:
As long as a provider can get their own /19 I have no problem with prefix filtering at the /19 level.
The problem comes about when big ISPs filter at /19s *AND* the allocators of space refuse to give ISPs /19s.
These two goals seem to be at odds in the current system for address allocation. How would you change the system to allow people to aquire address space that they need and get it routed?
The address allocation scheme is geared towards trying to promote utilization of IP space, thus the sorta "take just what you need" methodology.
The filters that you talk of seem to me to be crude proxies for controlling routing space on a particular providers network, this seems to me to be a reasonable thing (i.e. they have to make their network work).
If different providers were to sell routing "slots" on their network such that an ISP could guarantee that their announcements would be accepted (regardless of address length) this would seem to solve the problems of both those that can't "justify" a big block and those of the providers that want to control the use of their resources on their network as well.
It appears that you're primary argument is one of fairness and level playing field for all comers regardless of size, and I think this is a worthy goal if it can be done technically.
-scott
-- Craig Nordin -- cnordin@vni.net Virtual Networks http://www.vni.net
On Sat, 1 Mar 1997, Craig Nordin wrote:
Shouldn't the big boys (the ones who started all of this filtering) and the InterNIC be forced to come up with a fairer solution? At least if they don't do so voluntarily?
It is a fair solution, I did not like it. I started a few years ago with a Sprint T1 and IP space, we then renumbered into some MCI space. After that we got some space from the NIC. We then had to renumber that into a larger block from the nic. When I started I wanted my small amounts of address space to come from internic, but now that we have a nationwide network and are connected to 8 NAPs I know why I had to wait. The internic is not out to get the small guys, and if you get larger they will give you space. You will just need to get your space from your transit provider and then if you get big, renumber. I know there are a few provider out to get the small guys, but most just want to make the net better. Things like only peering with providers who are connected to all NSF NAPs and filtering are things they need to do. Nathan Stratton President, NetRail,Inc. ------------------------------------------------------------------------ Phone (888)NetRail NetRail, Inc. Fax (404)522-1939 230 Peachtree Suite 500 WWW http://www.netrail.net/ Atlanta, GA 30303 ------------------------------------------------------------------------ "Therefore do not worry about tomorrow, for tomorrow will worry about itself. Each day has enough trouble of its own." Matthew 6:34
Nathan Stratton wrote:
The internic is not out to get the small guys, and if you get larger they will give you space. You will just need to get your space from your transit provider and then if you get big, renumber.
It's the renumbering part that I think gives people the most heartburn... By the time you get "big enough" to warrent your own block, you've got at least 32 ClassCs of which, I'm betting, at least 28 are "given" to LAN-connected customers. This is a _major_ headache not only for the ISP to go thru but also a major headache to force your customers to go thru. That is, what I think, is what really is most painful; that by the time you are big enough to have your own block, you're too big to want to renumber: Catch 22 -- ==================================================================== Jim Jagielski | jaguNET Access Services jim@jaguNET.com | http://www.jaguNET.com/ "Not the Craw... the CRAW!"
On Sun, 2 Mar 1997, Jim Jagielski wrote:
It's the renumbering part that I think gives people the most heartburn... By the time you get "big enough" to warrent your own block, you've got at least 32 ClassCs of which, I'm betting, at least 28 are "given" to LAN-connected customers. This is a _major_ headache not only for the ISP to go thru but also a major headache to force your customers to go thru. That is, what I think, is what really is most painful; that by the time you are big enough to have your own block, you're too big to want to renumber: Catch 22
Yes, but as a smaller ISP you can offer much better service, and help you customers renumber. Yes I of all people know it is a _major_ headache, but it can be done, and there are ways to do it. Just because it is a "_major_ headache", is not a good reason to add a route to the global table, or have the nic give you a bigger block then you need at that time. Nathan Stratton President, NetRail,Inc. ------------------------------------------------------------------------ Phone (888)NetRail NetRail, Inc. Fax (404)522-1939 230 Peachtree Suite 500 WWW http://www.netrail.net/ Atlanta, GA 30303 ------------------------------------------------------------------------ "Therefore do not worry about tomorrow, for tomorrow will worry about itself. Each day has enough trouble of its own." Matthew 6:34
On Sun, 2 Mar 1997, Jim Jagielski wrote:
It's the renumbering part that I think gives people the most heartburn... By the time you get "big enough" to warrent your own block, you've got at least 32 ClassCs of which, I'm betting, at least 28 are "given" to LAN-connected customers. This is a _major_ headache not only for the ISP to go thru but also a major headache to force your customers to go thru. That is, what I think, is what really is most painful; that by the time you are big enough to have your own block, you're too big to want to renumber: Catch 22
Yes, but as a smaller ISP you can offer much better service, and help you customers renumber. Yes I of all people know it is a _major_ headache, but it can be done, and there are ways to do it.
I think there are some technologies available now that would drastically reduce this headache, as well as strech out a block of assigned addresses. For example, what about offering DHCP/BOOTP service for your customers? You provide a common DHCP/BOOTP server for your customers, configure their routers to forward DHCP/BOOTP packets. It makes configuration for them a whole lot easier and more standard (with almost every platform), plus you can assign them a block based on what they actually need *at the moment*. Should they need a larger block in the future, or should you renumber, just reconfigure the DHCP/BOOTP server and their router, and you're done. NAT is also a cool technology for this type of thing. Only assign real IP addresses to machines that provide IP services. Put everything else on 10.0.0.0/8 (or something else that won't conflict with real addresses). There are few reasons (if any) why the client side of the Internet would not work with NAT. Then, you can dynamically adjust your NAT pool of real addresses based on how many are actually needed for real usage. Plus, a renumber of every single client is a matter of adjusting your NAT tables, and having them renumber whatever Web/FTP/IP-service machines they have. There are probably other cool ways to reduce the headache of IP management, but these are a few I thought the group might be interested in. Pete Kruckenberg inQuo Internet Services pete@inquo.net
Nathan Stratton wrote:
On Sun, 2 Mar 1997, Jim Jagielski wrote:
It's the renumbering part that I think gives people the most heartburn... By the time you get "big enough" to warrent your own block, you've got at least 32 ClassCs of which, I'm betting, at least 28 are "given" to LAN-connected customers. This is a _major_ headache not only for the ISP to go thru but also a major headache to force your customers to go thru. That is, what I think, is what really is most painful; that by the time you are big enough to have your own block, you're too big to want to renumber: Catch 22
Yes, but as a smaller ISP you can offer much better service, and help you customers renumber. Yes I of all people know it is a _major_ headache, but it can be done, and there are ways to do it.
Just because it is a "_major_ headache", is not a good reason to add a route to the global table, or have the nic give you a bigger block then you need at that time.
Oh I agree... It's just that I know of more than a few ISPs who have done things like keep their current NSP, but with something like a 56k line (so they don't have to renumber) and then get a bigger pipe from somebody else and just use BGP to make everything work... -- ==================================================================== Jim Jagielski | jaguNET Access Services jim@jaguNET.com | http://www.jaguNET.com/ "Not the Craw... the CRAW!"
Yes, but as a smaller ISP you can offer much better service, and help you customers renumber. Yes I of all people know it is a _major_ headache, but it can be done, and there are ways to do it.
Just because it is a "_major_ headache", is not a good reason to add a route to the global table, or have the nic give you a bigger block then you need at that time.
Look. I think Kim's point is true. They *do* allocate more space *than* you actually need so that when you need it, you can actually get it *then*. If you're growing fast enough, you'll have to renumber *once*. If you choose poorly your upstream providers, you'll have to renumber more than once.
Nathan Stratton President, NetRail,Inc.
Avi
On Sat, 1 Mar 1997, Scott Huddle wrote:
If different providers were to sell routing "slots" on their network such that an ISP could guarantee that their announcements would be accepted (regardless of address length) this would seem to solve the problems of both those that can't "justify" a big block and those of the providers that want to control the use of their resources on their network as well.
It appears that you're primary argument is one of fairness and level playing field for all comers regardless of size, and I think this is a worthy goal if it can be done technically.
I think this is more than a technical problem. It also impacts relationships (i.e. peering) and it becomes a business issue since money is changing hands. For this to work a core network provider would have to do several things. 1. set a fee schedule for routing slots and determine what the conditions of sale will be so that every Tom, Dick and Jane doesn't try to buy a /24 slot for their PowerMAC webserver with ISDN TA attached. 2. negotiate the peering relationships with at least the other core network providers such that they can provide a reasonably certain guarantee that the announcements will be accepted by their peers. Note that this does *NOT* neccessarily require settlements. 3. set up a feedback loop so that routing table growth does not go crazy. IMHO this would need to involve some sort of a quota system whereby the group of core network providers who have agreed to listen to purchased announcements will also agree how many such slots per month can be sold based solely on technical considerations. The sales force would then be given an inventory of routing table slots that they can sell and when they are gone, they are gone. 4. deal with antitrust issues. Because of the close coordination needed by the core network providers to make this work, as soon as prices for routing slots stabilize there will be charges of price-fixing. This needs to be dealt with up-front, and IMHO, it is the single most important issue because a) failure to do it properly will cause severe financial penalties to hit the providers and b) doing it properly will cost significant dollars in lawyers fees. Also, this becomes an international trade issue. North America covers more than one country and, as you are all well aware, most major European and Asian and Australian providers do peer at North American IXP's or are planning to do so in the near future. I wouldn't expect to see any quick solution to this problem but it is probably a good idea to start looking at the technical and other issues right now. It looks like the next generation of routers will be upon us by the middle of the year and the limits on routing table size will be significantly increased. The question is, what happens next? Simply loosening up the filters to allow /20's or /21's will not create as many problems as it solves. People whose equipment cannot handle routing tables with 80,000 - 90,000 routes will not be happy and their could be some serious antitrust implications as a result. The bottom line is that we need to have a consensus on how to take the next step and this mailing list is probably not the best place to work it out. The business and legal issues really belong on PIARA. Send subscribe to the address piara-request@apnic.net Note that in the past, PIARA has been focussed on the idea of selling IP allocations but that idea really never caught on and is basically dead for now. But the idea of selling routing table slots was never discussed much on PIARA so there is really no point in reading the list archives. Just join the list and start posting. Michael Dillon - Internet & ISP Consulting Memra Software Inc. - Fax: +1-250-546-3049 http://www.memra.com - E-mail: michael@memra.com
1. set a fee schedule for routing slots and determine what the conditions of sale will be so that every Tom, Dick and Jane doesn't try to buy a /24 slot for their PowerMAC webserver with ISDN TA attached.
Is there really a problem with that, as long as Tom, Dick, and Jane are willing to pay the $x-thousand annual cost of the routing slot for their /24? (Or /28?) -- Shields, CrossLink.
participants (8)
-
Avi Freedman -
Craig Nordin -
Jim Jagielski -
Michael Dillon -
Nathan Stratton -
Pete Kruckenberg -
Scott Huddle -
shields＠crosslink.net