What is PI and PA ? Provider Independent and Provider Assigned ? In this post, I will explain the important considerations on PI (Provider Independent) and PA (Provider Assigned).
In this post, I will explain below points :
- What is PI and PA ?
- Why PI and PA addresses are used ?
- Where you can get PI address and where you can get PA address ?
- If multihoming is required, is PI or PA better ?
- Is there any deployment consideration for PI and PA ?
What is PI and PA?
PI stands for Provider Independent , PA stands for Provider Assigned. These are Globally Unique IPv4 addresses. So, both are public IP addresses which are used on the Internet to connect with the other networks.
PA addresses is known as Provider Aggregatable address as well. At the end of this post, I will give you an example, how aggregation might be an issue.
On the Internet, there are three business relationship. A company, can be a customer of another one, a company can be a provider to the others , and lastly a company can be a peer with the others. (This is called, settlement free peering)
Providers receive a public address (It is also called as real IP) to use over the Internet to connect with the other networks. Providers receive it from the RIRs (Regional Internet Registries)
If a company receives it’s IP address space from it’s provider, then this assignment is called PA (Provider Assigned), because, customer IP address block is assigned by the provider.
If a customer receives the IP address block from the RIRs (Regional Internet Registries) then this assignment is called as PI (Provider Independent). Because, IP address space is not assigned by the providers, but a higher address allocation place which is RIRs (RIPE is an RIR for example)
So, RIPE, ARIN, APNIC, AFRINIC etc. these are RIRs. (Regional Internet Registries) and Internet Service Providers is called as LIR (Local Internet Registries)
If a customer leaves the Service Provider who assigned the PA address space, it can be assigned to another customer.
Why PI and PA addresses are used ?
What is the selection criteria ? Should you receive PI or PA address block then ?
Smaller companies generally work with a single provider. In fact, there are some researches which prove this claim. For these companies, receiving PA (Provider Assigned) space is fine. Unless, they are totally okay to work with the current providers or if they don’t have any other provider option at their territory.
Because, PI address space is paid service. Company needs to pay for PI address space.
Another consideration, when the customer receives PI address space, even it is single homed currently, when it wants to leave the current Service Provider, it doesn’t have to readdress all of it’s hosts, servers, routers etc. Because IP addresses belong to customer itself.
Where you can get PI and PA addresses ?
Companies can get PI address space either directly from RIR, then they need to deal with the paperwork, negotiation etc. or their provider can do these paperworks and discussion with the RIRs on behalf of their customers. In any case, customer pays the money. In the former case, customer pays to the RIR, in latter case, customer pays to it’s Service Provider.
If multihoming is required, is PI or PA better ?
If a company has two different upstream Service Providers, then PI is generally a better choice. Imagine there are two providers. Provider A and Provider B.
If the customer receives PA (Provider Assigned) address space from Provider A, Provider B may not accept the prefixes from that customer. This is not always the case, but I heard it happened to some networks.
Also, as a IP address block owner by getting PI address space from RIRs, companies can do better traffic engineering. I mean, better ingress and egress BGP path selection. I will show you a topology for this later in a separate post.
Is there any deployment consideration for PI and PA address space ?
There might be more than one but I will explain you the common one.
This consideration, especially with PA (Provider Assigned) address space.
Same company in the above example, receives a PA block from Provider A.
When Provider A, assigns an IP address block to it’s customers, basically longer IP prefix is assigned. Let me explain.
Let’s say customer receives, 10.20.30.40/24 IP address block from Provider A. Provider A assigns this block from its larger block, for example, 10.20.0.0/16.
So far no problem.
But if this customer, advertises 10.20.30.40/24 block to both, Provider A and Provider B.
What happens ?
Provider A advertises 10.20.0.0/16 towards Internet (It sends an aggregate) Provider B advertises 10.20.30.40/24 towards Internet (Because the owner of this smaller IP address space is Provider A, Provider B has to advertise 10.20.30.40/24)
All the traffic comes to Provider B.
So customer cannot receive the traffic from Provider A link. In this case, to provide a load balancing/multihoming capability for the customer, Provider A needs to advertise 10.20.30.40/24, in addition to 10.20.0.0/16.
It solved the issue. Right ?
For the customer multihoming , yes.
For the overall DFZ ? Worse.. And in fact, this was a big debate, PI would create an issue for the routers resources, networks couldn’t handle the amount of prefixes due to lack of aggregation with PI space. PA, as a Provider Assigned/Aggregatable address space, comes with multihoming problem but provides smaller Internet routing table.
See you in the next post.