I was just looking through the graphs on Tony's Web site and noticed something somewhat disturbing. For the first 2000 days of Tony's plots (starting from 1/1/1994) the number of routes has grown roughly by 10,000 every 500 days. so 20k --> 30k, 30k --> 40k, 40k --> 50k, & 50k --> 60k. But if you look at the last 250 days or so you see that the table has grown by more than 16k routes. So we are seeing growth at 300% of what we saw for the last 5 plus years. It also looks annoyingly geometric or perhaps exponential, instead of the nice linear growth since CIDR was introduced. Take a look at: http://www.employees.org/~tbates/cidr.hist.plot.html Just eyeballing it since I am too tired to actually plot the number right now it looks like we could hit 100k routes by the end of the year or early next year. To the $64,000^h^h^h^h^h^h $10,000,000 (you gotta change with the times :-) questions are why? anyone care to guess when it starts to become a problem and with what widely deployed boxes first? is there anything we can do about it? Have a good weekend. ---> Phil
On Sat, 13 May 2000 pjnesser@Nesser.COM wrote:
But if you look at the last 250 days or so you see that the table has grown by more than 16k routes. So we are seeing growth at 300% of what we saw for the last 5 plus years. It also looks annoyingly geometric or perhaps exponential, instead of the nice linear growth since CIDR was introduced.
If you just check from 01/01/99 to date then it looks linear or at least close to linear. I guess it *could* be that growing amount of new companies getting internet access is increasing. Is there any data that show "CIDR GAIN" from the cidr report, so we can see if the increase corresponds to an increase in (perhaps unneccessary) smaller announcements in larger blocks, or if it is actually just a lot more blocks allocated that needs to be routed. Any stats on arin/ripe/apnic new allocations of blocks in the same timeframe? Both in terms of IP adresses and in number of blocks of IP adresses. This would also give us some kind of hint as to when IPv4 space will be exhausted (or are there already projections about this?) -- Mikael Abrahamsson email: swmike@swm.pp.se
I've mentioned this before, so I'll just note it lightly. There are a growing number of companies (dot-coms are only one of them) that have small head-count (<4000), but are spread out from Sydney to New York, with many "lone eagles" in the MST zone. They could probably do everything on a portable /24. However, with everyone filtering out announcements less than /20, such companies are encouraged to drop NAT, and use other methods to justify a /19, just so they can participate in peering (I won't say whom, one is a CTI development company). The VPN solution is cute, but the entire VP then becomes single-homed, at the VPN gateway (The alternative is that each location gets their own /24, linked by a VPN, to the other /24s, there are serious performance issues with this approach and hte /24 may only represent a single actual user). All of this burns IP addresses. The point: Filtering BGP announcements costs in IP space allocations. There is a mathmatical relationship between IP address allocations, table sizes, and routing policies. Also, part of the relationship is determined by client business requirements. Organizations are becomeing more geophysically diffused, with many end-nodes actually participating in multiple organizations. This is only starting now (I still see over 100K nodes actually doing this), it will get much worse.
-----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of Mikael Abrahamsson Sent: Saturday, May 13, 2000 2:10 AM To: nanog@nanog.org Subject: Re: CIDR Report
On Sat, 13 May 2000 pjnesser@Nesser.COM wrote:
But if you look at the last 250 days or so you see that the table has grown by more than 16k routes. So we are seeing growth at 300% of what we saw for the last 5 plus years. It also looks annoyingly geometric or perhaps exponential, instead of the nice linear growth since CIDR was introduced.
If you just check from 01/01/99 to date then it looks linear or at least close to linear.
I guess it *could* be that growing amount of new companies getting internet access is increasing. Is there any data that show "CIDR GAIN" from the cidr report, so we can see if the increase corresponds to an increase in (perhaps unneccessary) smaller announcements in larger blocks, or if it is actually just a lot more blocks allocated that needs to be routed. Any stats on arin/ripe/apnic new allocations of blocks in the same timeframe? Both in terms of IP adresses and in number of blocks of IP adresses. This would also give us some kind of hint as to when IPv4 space will be exhausted (or are there already projections about this?)
-- Mikael Abrahamsson email: swmike@swm.pp.se
Of couse, folks have to decide what's important: 1) Routing Table size Is this as important of an issue as it was 3 years ago? Nope, not with M40s and GSRs as standard backbone routes. 100K routes is no sweat with these boxes. Of course, some of the big carriers would have to finally replace some of those older 7505s floating around their networks. 2) IP Address Space Scarcity There is no real address space scarcity. We still have most of 64/2 to go through. And there is always an easy answer to this "problem" - turn IP space into a commodity and have an exchange system where folks purchased address space from each other. This would result in the hoarded "class A" space being thrown back into circulation, as the companies that had them would have an incentive to unload them - money. As IP space is really not that scarce (and would be less so with the Class A's in circulation), the prices would be reasonable. When they became unreasonable, there would be a market-driven move to IPv6. Ah, capitalism. The answer to most of these issues is to allow much smaller allocations in a specific block - say 65/8. Allow /24-/21 allocations in this block for smaller multihomed enterprises. Then, start cracking down on the other blocks by filtering smaller prefixes. This would enforce necessary aggregation where possible, but still allow multihomed enterprises to obtain an appropriate amount of IP space on their own. As far as routing table growth - with current practices, we will continue to see this level of growth due to an increase in the number of multihomed enterpises. How mane non-ISPs ran BGP with upstream providers a few years ago? How many do now? The growth is astonishing... Daniel Golding Director, Network Evaluation and Design NetRail, Inc. On Sat, 13 May 2000, Roeland Meyer (E-mail) wrote:
I've mentioned this before, so I'll just note it lightly. There are a growing number of companies (dot-coms are only one of them) that have small head-count (<4000), but are spread out from Sydney to New York, with many "lone eagles" in the MST zone. They could probably do everything on a portable /24. However, with everyone filtering out announcements less than /20, such companies are encouraged to drop NAT, and use other methods to justify a /19, just so they can participate in peering (I won't say whom, one is a CTI development company). The VPN solution is cute, but the entire VP then becomes single-homed, at the VPN gateway (The alternative is that each location gets their own /24, linked by a VPN, to the other /24s, there are serious performance issues with this approach and hte /24 may only represent a single actual user). All of this burns IP addresses.
The point: Filtering BGP announcements costs in IP space allocations. There is a mathmatical relationship between IP address allocations, table sizes, and routing policies. Also, part of the relationship is determined by client business requirements.
Organizations are becomeing more geophysically diffused, with many end-nodes actually participating in multiple organizations. This is only starting now (I still see over 100K nodes actually doing this), it will get much worse.
-----Original Message----- From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of Mikael Abrahamsson Sent: Saturday, May 13, 2000 2:10 AM To: nanog@nanog.org Subject: Re: CIDR Report
On Sat, 13 May 2000 pjnesser@Nesser.COM wrote:
But if you look at the last 250 days or so you see that the table has grown by more than 16k routes. So we are seeing growth at 300% of what we saw for the last 5 plus years. It also looks annoyingly geometric or perhaps exponential, instead of the nice linear growth since CIDR was introduced.
If you just check from 01/01/99 to date then it looks linear or at least close to linear.
I guess it *could* be that growing amount of new companies getting internet access is increasing. Is there any data that show "CIDR GAIN" from the cidr report, so we can see if the increase corresponds to an increase in (perhaps unneccessary) smaller announcements in larger blocks, or if it is actually just a lot more blocks allocated that needs to be routed. Any stats on arin/ripe/apnic new allocations of blocks in the same timeframe? Both in terms of IP adresses and in number of blocks of IP adresses. This would also give us some kind of hint as to when IPv4 space will be exhausted (or are there already projections about this?)
-- Mikael Abrahamsson email: swmike@swm.pp.se
If I understand the thread correctly, part of the growth problem in the routing tables has been attributed to small clients with sub/24 address allocations implementing multi-homed solutions. As an Internet Access Provider (IAP), I thought I'd share a few opinions... + ip allocation + Address allocation is not very pretty when you start to move down from the tier 1's. Upstream providers tend to grab the "next available" /22, /23, or /24, which is usually located about a thousand miles away from your current allocation. (Well, you should have planned for growth...) Agreed, but ARIN and most upstreams do not accept "I think I'm gonna quadruple my customer base next year" as a viable excuse. And for good reason, given the allocation abuses that have occured in the past. So now we have these IAP's with multiple /24's scattered haphazardly across the upstream provider's delegated block. Now what happens when this IAP multi-homes? (There are very few valid reasons to multi-home, please consult with RFC x, BCP y, and an experienced network engineer.) Roger that. + continued in Multi-home II + -brad (Rural CNE)
Try www.telstra.net/ops/bgp.html for another view of the same accelerated curve, together with a number of other views of the BGP space. The interesting aspect to note is the declining amount of address space per routing advertisement on average. Geoff At 05:50 PM 5/13/00, pjnesser@Nesser.COM wrote:
I was just looking through the graphs on Tony's Web site and noticed something somewhat disturbing. For the first 2000 days of Tony's plots (starting from 1/1/1994) the number of routes has grown roughly by 10,000 every 500 days. so 20k --> 30k, 30k --> 40k, 40k --> 50k, & 50k --> 60k.
This is not really supprising if you have been following changes in RIR address allocation policy, which seems really to me to be following a trend, more than creating one, but it is worth noting that the trend exists. In message <4.3.1.2.20000513213942.00a9dd90@mako1.telstra.net>, Geoff Huston write s:
Try www.telstra.net/ops/bgp.html for another view of the same accelerated curve, together with a number of other views of the BGP space.
The interesting aspect to note is the declining amount of address space per routing advertisement on average.
Geoff
At 05:50 PM 5/13/00, pjnesser@Nesser.COM wrote:
I was just looking through the graphs on Tony's Web site and noticed something somewhat disturbing. For the first 2000 days of Tony's plots (starting from 1/1/1994) the number of routes has grown roughly by 10,000 every 500 days. so 20k --> 30k, 30k --> 40k, 40k --> 50k, & 50k --> 60k.
--- jerry@fc.net Director Network Operations/Network Engineering, Wayport, Inc. 512-519-6193 www.wayport.net 8303 Mopac Expressway Suite A300, Austin Tx.
On Sat, 13 May 2000 pjnesser@Nesser.COM wrote:
But if you look at the last 250 days or so you see that the table has grown by more than 16k routes. So we are seeing growth at 300% of what we saw for the last 5 plus years. It also looks annoyingly geometric or perhaps exponential, instead of the nice linear growth since CIDR was introduced.
This might be due to the fact that more companies are buying backup connections and announcing maybe only a /24 or /23. Of course, Verio will filter them saying it is good for the net. Tho they forget that they send out what they claim to filter. Seems they might have another motive for filtering... * i12.11.162.0/24 100 0 2914 4592 i * i12.16.244.0/22 100 0 2914 6201 i * i63.64.181.0/24 100 0 2914 11826 i * i63.73.199.0/24 100 0 2914 13626 i
To the $64,000^h^h^h^h^h^h $10,000,000 (you gotta change with the times :-) questions are why? anyone care to guess when it starts to become a problem and with what widely deployed boxes first? is there anything we can do about it?
Our routers carry 1.1 million paths + in their tables today. I have seen higher numbers (that was the first router I logged into). Current GSRs and Juunipers in the core should have no problems with 100,000 routes. Christian
On Sat, 13 May 2000, Christian Nielsen wrote: [ # of routing table entries growing]
connections and announcing maybe only a /24 or /23. Of course, Verio will filter them saying it is good for the net. Tho they forget that they send out what they claim to filter. Seems they might have another motive for filtering...
Their customers are paying them for the service and buy the extra cpu and memory and routing table slots. You could filter on the same boundaries as Verio.... do unto others before others do unto you?
Our routers carry 1.1 million paths + in their tables today. I have seen
That is impressive. /vijay
Take a look at: http://www.employees.org/~tbates/cidr.hist.plot.html
I think the other parts of the data also point to shoddy aggregation (not that this is new). Look at some of the /8 nets that have been broken down and you'll see some sequences crying out for order. I see (approximately) this right now: 600 pieces of net 12 850 pieces of net 24 1600 pieces of net 63 900 pieces of net 64 Just eyeballing the announcements shows that this could be a *lot* smaller. But even if all we saw were /19s in there we could expect to see 2048 /19s per /8, which is larger than what we see now. Is there data that shows the possible "NetGain" over the past 6 years? And if so, is it getting better or worse? -mark
participants (10)
-
Christian Nielsen -
Daniel L. Golding -
Geoff Huston -
Jeremy Porter -
Mark Kent -
Mikael Abrahamsson -
pjnesser＠Nesser.COM -
Roeland Meyer (E-mail) -
Rural CNE -
Vijay Gill