How big a network is routed these days?
Hello, my organization is considering PI addresses as a way to multihost. Having read the archives regarding disadvantages and alternatives, my question is how big a network must one have to be reasonably sure the BGP routers will accept the route? regards, JS
my organization is considering PI addresses as a way to multihost. Having read the archives regarding disadvantages and alternatives, my question is how big a network must one have to be reasonably sure the BGP routers will accept the route?
/24 Steinar Haug, Nethelp consulting, sthaug@nethelp.no
On Wed, 17 Jan 2007, John Smith wrote:
my organization is considering PI addresses as a way to multihost. Having read the archives regarding disadvantages and alternatives, my question is how big a network must one have to be reasonably sure the BGP routers will accept the route?
A /24 is the smallest block of IPv4 addresses that you can reasonably expect to be globally reachable. Depending on where you're located, the different address registries (ARIN, RIPE, APNIC, etc...) have different policies regarding the smallest PI block they'll allocate to end users. jms
I'm interested as to why RIRs dont set the minimum PI allocatable to /24 in order to fit with the current trend. I mean, I can see the reason for smaller allocations where an LIR routes and aggregates both but these are rare and probably legacy examples. Changing the allocation policy such that a /24 minimum exists for PI would be a good thing IMHO, forcing people to either apply for portions of PA space from LIRs (and having LIRs do what they should be doing) or lie through their teeth to get a /24 (but then not try and wonder why anything smaller is not routed correctly) Of course I'm probably opening the proverbial can of worms about who should or shouldn't apply and how they do, but I do find the possibility of obtaining a sub /24 PI allocation a little odd in this day and age. Dave. Justin M. Streiner wrote:
On Wed, 17 Jan 2007, John Smith wrote:
my organization is considering PI addresses as a way to multihost. Having read the archives regarding disadvantages and alternatives, my question is how big a network must one have to be reasonably sure the BGP routers will accept the route?
A /24 is the smallest block of IPv4 addresses that you can reasonably expect to be globally reachable. Depending on where you're located, the different address registries (ARIN, RIPE, APNIC, etc...) have different policies regarding the smallest PI block they'll allocate to end users.
jms
On Jan 17, 2007, at 12:19 PM, David Freedman wrote:
I'm interested as to why RIRs dont set the minimum PI allocatable to /24 in order to fit with the current trend.
In the 2002-3 micro-assignment policy, the RIR's assign a minimum of a /22. As far as I know, all of the PI /24's are thus "legacy" in nature. From my experience, /24's and longer assigned by RIRs likely to be routed, as well as ones from the old class C space, and people have mostly had problems with / 24 PA space in the old Class A and B space. Your milage, of course, definitely may vary here. Regards Marshall
I mean, I can see the reason for smaller allocations where an LIR routes and aggregates both but these are rare and probably legacy examples.
Changing the allocation policy such that a /24 minimum exists for PI would be a good thing IMHO, forcing people to either apply for portions of PA space from LIRs (and having LIRs do what they should be doing) or lie through their teeth to get a /24 (but then not try and wonder why anything smaller is not routed correctly)
Of course I'm probably opening the proverbial can of worms about who should or shouldn't apply and how they do, but I do find the possibility of obtaining a sub /24 PI allocation a little odd in this day and age.
Dave.
Justin M. Streiner wrote:
On Wed, 17 Jan 2007, John Smith wrote:
my organization is considering PI addresses as a way to multihost. Having read the archives regarding disadvantages and alternatives, my question is how big a network must one have to be reasonably sure the BGP routers will accept the route? A /24 is the smallest block of IPv4 addresses that you can reasonably expect to be globally reachable. Depending on where you're located, the different address registries (ARIN, RIPE, APNIC, etc...) have different policies regarding the smallest PI block they'll allocate to end users. jms
On 17-Jan-2007, at 12:43, Marshall Eubanks wrote:
On Jan 17, 2007, at 12:19 PM, David Freedman wrote:
I'm interested as to why RIRs dont set the minimum PI allocatable to /24 in order to fit with the current trend.
In the 2002-3 micro-assignment policy, the RIR's assign a minimum of a /22.
The RIRs all have different policies, and anybody interested in finding out the rules for any individual RIR are best advised to consult the policy documents published by the RIR in question directly. The original question in this thread (that of reachability of addresses covered by long-prefix announcements) is profoundly on- topic here, however, and explicitly disclaimed by all the RIRs last time I checked. For the ARIN region, the Number Resource Policy Manual currently published at <http://www.arin.net/> mentions the following minima:
4.3.2.1 Single Connection
The minimum block of IP address space assigned by ARIN to end- users is a /20. [...]
4.3.2.2 Multihomed Connection
For end-users who demonstrate an intent to announce the requested space in a multihomed fashion, the minimum block of IP address space assigned is a /22. [...]
4.4 Micro-allocation
ARIN will make micro-allocations to critical infrastructure providers of the Internet, including public exchange points, core DNS service providers (e.g. ICANN-sanctioned root, gTLD and ccTLD operators) as well as the RIRs and IANA. These allocations will be no longer than a /24 using IPv4 or a /48 using IPv6. [...]
As far as I know, all of the PI /24's are thus "legacy" in nature.
As the above snippet from the policy manual suggests (and as my experience confirms) there are recent assignments made to end users by ARIN under the micro-allocation policy which were made with the expectation that individual /24s would be advertised globally. Clearly these are not the most usual case, as the description of those who qualify for such assignments above indicates, but it would be a mistake to assume that *all* /24 assignments are legacy.
From my experience, /24's and longer assigned by RIRs likely to be routed, as well as ones from the old class C space, and people have mostly had problems with /24 PA space in the old Class A and B space.
I'm not aware of widespread filtering of /24s based on assignment boundaries in recent years. Can anybody confirm whether this is still a real problem? The real problem today, I thought, was that of allocations or assignments being made from fresh /8s that still feature in peoples' bogon filters. Joe
4.3.2.1 Single Connection
The minimum block of IP address space assigned by ARIN to end- users is a /20. [...]
4.3.2.2 Multihomed Connection
For end-users who demonstrate an intent to announce the requested space in a multihomed fashion, the minimum block of IP address space assigned is a /22. [...]
4.4 Micro-allocation
ARIN will make micro-allocations to critical infrastructure providers of the Internet, including public exchange points, core DNS service providers (e.g. ICANN-sanctioned root, gTLD and ccTLD operators) as well as the RIRs and IANA. These allocations will be no longer than a /24 using IPv4 or a /48 using IPv6. [...]
As far as I know, all of the PI /24's are thus "legacy" in nature.
As the above snippet from the policy manual suggests (and as my experience confirms) there are recent assignments made to end users by ARIN under the micro-allocation policy which were made with the expectation that individual /24s would be advertised globally. Clearly these are not the most usual case, as the description of those who qualify for such assignments above indicates, but it would be a mistake to assume that *all* /24 assignments are legacy.
Actually, generally, the expectation under 4.4 is that the addresses will not be advertised at all for the most part, since, generally, there's no need to advertise the route to the exchange point, itself, into the global routing table. 4.4 is intended to support internet exchanges, ala MAEs, etc. In terms of 4.3.2.1 and 4.3.2.2, I believe ARIN has worked very hard to express no expectation or intent about how assignments relate to route advertisements and routing policy. Owen
On 17-Jan-2007, at 18:36, Owen DeLong wrote:
Actually, generally, the expectation under 4.4 is that the addresses will not be advertised at all for the most part, since, generally, there's no need to advertise the route to the exchange point, itself, into the global routing table. 4.4 is intended to support internet exchanges, ala MAEs, etc.
... and operators of critical DNS infrastructure, as the text I quoted indicated. Not much point in numbering a TLD server out of a block that isn't going to be advertised.
In terms of 4.3.2.1 and 4.3.2.2, I believe ARIN has worked very hard to express no expectation or intent about how assignments relate to route advertisements and routing policy.
Indeed, as I believe I mentioned. Joe
participants (7)
-
David Freedman -
Joe Abley -
John Smith -
Justin M. Streiner -
Marshall Eubanks -
Owen DeLong -
sthaug＠nethelp.no