IP-in-SCION Tunneling

Overview

IP-in-SCION tunneling is a mechanism that allows to tunnel IP packets between two IP-in-SCION tunneling endpoints, i.e., Anapaya EDGE appliances. This mechanism enables any type of application and communication to take advantage of SCION-based networking without the need to update any application, client, or server.

../../../_images/scion_tunneling_overview1.png

On a high-level, IP-in-SCION tunneling involves the following steps:

  1. A sender sends an IP packet towards an IP destination.

  2. The IP packet reaches an IP-in-SCION tunneling endpoint, such as an Anapaya EDGE appliance, in the sender’s network via standard local IP routing.

  3. Based on the destination IP address, the Anapaya EDGE appliance determines the destination SCION AS and possible IP-in-SCION tunneling endpoints within the destination SCION AS. To achieve this, the Anapaya EDGE appliance uses the SCION Gateway Routing Protocol (SGRP).

  4. Next, the Anapaya EDGE appliance determines the optimal SCION path to reach the destination SCION AS. The choice of the path depends on the Routing Domain Configuration and the real-time network performance characteristics, such as latency, jitter, drop rate, etc.

  5. Then, the original IP packet is encapsulated within a SCION packet by the Anapaya EDGE appliance and sent to the remote IP-in-SCION tunneling endpoint in the destination SCION AS. The SCION path used is the one chosen in the previous step.

  6. The remote IP-in-SCION tunneling endpoint receives the SCION packet and decapsulates the original IP packet. It then forwards the packet to the final IP destination using standard local IP routing.

Routing Domains

Routing Domains are the primary mechanism to configure the traffic policies used when communicating with remote IP destinations. Conceptually, a routing domain is defined by a set of IP prefixes to which a common set of traffic policies should be applied as shown in the following figure.

../../../_images/domains_ip_space1.svg

These IP prefixes could be announced by one or multiple SCION ASes. The Anapaya EDGE appliance can optimize path selection across all possible destination SCION ASes (while respecting the configured traffic policy). This flexibility enables operators to capture a wide range of routing use-cases - from simple site-to-site tunnels, to complex setups involving remotes that are part of multiple SCION isolation domains, or destinations that are reachable via many SCION ASes.

A routing domain can be further specified by a set of remote SCION ASes. This is useful for example when a domain should only include remote ASes of a specific ISD or a specific set of remote ASes, e.g., the sites of the company’s wide area network. Furthermore, it is also possible to limit the routing domain to a set of source ASes. This is used by Anapaya EDGE appliances that are part of multiple ISDs and need to separate the traffic and use different policies per ISD.

Finally, a domain also includes the ability to limit the IP prefixes that are announced by a local Anapaya EDGE appliance to remote IP-in-SCION tunneling endpoints.

Traffic Policies

Traffic policies allow users to influence and control the path selection process of the Anapaya EDGE appliance. For example, an operator can define policies to keep traffic within Swiss jurisdiction or avoid certain network providers. A traffic policy uses path filters to specify the set of SCION paths that should be considered for the path selection. Each path filter defines a subset of all available SCION paths.

Path filters are combined into a failover sequence to define the order in which the path filters are considered. The path filters in a failover sequence are applied separately to the set of possible paths until one of them returns a non-empty path set.

Finally, the traffic policy includes a traffic matcher that defines the set of IP packets for which the failover sequence should be applied.

SCION Gateway Routing Protocol (SGRP)

The SCION Gateway Routing Protocol (SGRP) is a routing protocol that enables IP-in-SCION tunneling endpoints, i.e., Anapaya EDGE appliances, to map IP prefixes to SCION ASes.

At a high level, a tunneling endpoint participating in SGRP between two SCION ASes does the following steps:

  1. It discovers the tunneling endpoints in the remote SCION AS. To that end, it periodically sends a discovery message to the control plane of the remote AS which replies with a list of local IP-in-SCION tunneling endpoints.

  2. It periodically queries each discovered endpoint in the remote AS to learn the IP prefixes it announces. From that, the local endpoint builds a mapping from IP prefix to remote tunneling endpoints.

  3. When queried by a remote tunneling endpoint about its prefixes, the local endpoint replies with the set of IP prefixes it wants to announce to the remote.

For the Anapaya EDGE appliance, the set of announced IP prefixes can either be statically configured or is dynamically learned via BGP. Please refer to the BGP section for information on how to configure a BGP session to a peering BGP router.

Domain Configuration

As explained in Routing Domains, domains define a common set of traffic policies towards a distinct range of IP destinations. Thus, there are two main aspects to configuring a domain:

  1. What entities are part of the domain?

  2. How is traffic routed within the domain?

The what includes the IP prefixes that are accepted and announced by the local endpoint, the remote SCION ASes, and possibly the local SCION AS identities (in case the Anapaya EDGE appliance is part of multiple SCION ISDs).

The how includes the traffic policies that are applied to outgoing traffic towards any remote destination that is part of the domain.

Domain Entities

This sections explains how to define what entities are part of a domain.

IP Prefixes

The prefixes section of the domain configuration defines the set of IP prefixes that are accepted and announced by the local endpoint.

The accept_filter subsection defines the set of IP prefixes that are accepted from remote tunneling endpoints. IP packets towards an address in these prefixes are tunneled to a remote tunneling endpoint that matches the domain’s remote matchers.

There cannot be more than one routing domain that accepts a specific IP prefix. Thus, the accept_filter matchers of different domains must not overlap.

The prefixes matched by the announce_filter subsection of the domain define the maximum set of IP prefixes that will be announced to remote tunneling endpoints.

Each filter in either of these sections is a list of IP prefix matchers combined with an action. The action is either ACCEPT or REJECT to implement allow- and block lists.

Note

The accept and announce filters are what the name implies - filters. This means that not everything they contain is actually accepted or announced. The set of announced IP prefixes are all the IP prefixes configured in Static Announcements or received from BGP peers configured in Border Gateway Protocol (BGP) and filtered by the announce filter. Similarly, the set of accepted IP prefixes is the set of all IP prefixes received from remote IP-in-SCION tunneling endpoints filtered by the accept filter.

Remote ASes

The remote_isd_ases section contains a list of SCION AS matchers that - together with an action (ACCEPT or REJECT) - define the set of SCION ASes that are part of the domain. This has the effect that IP prefix announcements will be accepted from these remote ASes and all IP traffic will be tunneled over paths that end in one of these remote ASes. A SCION AS matcher has the form <isd>-<as> where <isd> is the ISD ID and <as> is the AS ID. Both <isd> and <as> can be 0 to match all ISDs and all ASes respectively. E.g., the matcher 1-ff00:0:1 matches exactly this single AS, 1-0 matches all ASes that are part of ISD 1, and 0-ff00:0:1 matches the specific AS in all ISDs. A single 0 (or 0-0) matches all ASes in all ISDs.

Local AS Identities

The local_isd_ases section defines a list of local SCION AS identities that are part of the domain. Setting this field is only necessary if the Anapaya EDGE appliance is part of multiple SCION ISDs and an operator wants to restrict the domain to a subset of the local SCION ISDs. This has the effect that traffic towards remote tunneling endpoints in this domain only uses paths that start in one of the listed local ASes. Note that if the Anapaya EDGE appliance is only part of a single ISD or if all available local AS identities are part of the domain, then local_isd_ases does not need to be set.

Default Domain

There can be a single default domain. The default domain is used when no other domain matches a packet’s destination, i.e., the default domain implicitly accepts the entire IP space that is not covered by other domains. For that reason, the default domain can not include any accept_filter. Specifying a default domain is optional.

Examples

Allow-all domain

This example shows a domain that accepts any IP prefix from any remote AS and announces any learned or configured IP prefixes to any remote AS.

{
   "remote_isd_ases": [
      {
         "isd_as": "0-0",
         "action": "ACCEPT",
         "description": "Include all remote ASes",
         "sequence_id": 0
      },
   ],
   "prefixes": {
      "accept_filter": [
         {
            "prefixes": [ "0.0.0.0/0, ::/0" ],
            "action": "ACCEPT",
            "sequence_id": 0
         },
      ],
      "announce_filter": [
         {
            "prefixes": [ "0.0.0.0/0, ::/0" ],
            "action": "ACCEPT",
            "sequence_id": 0
         },
      ],
   },
}

ISD-specific domain

This example shows a domain that only accepts 1.0.0.0/8 and 2.0.0.0/8 from ASes in ISD 1 and announces only prefixes in 1.100.0.0/16. Furthermore, the domain only includes the local AS identity that is part of ISD 1.

{
   "remote_isd_ases": [
      {
         "isd_as": "1-0"
         "action": "ACCEPT"
         "description": "Only ASes in ISD 1 are part of the domain"
         "sequence_id": 0
      },
   ],
   "local_isd_ases": [ "1-ff00:0:1" ],
   "prefixes": {
      "accept_filter": [
         {
            "prefixes": [ "1.0.0.0/8", "2.0.0.0/8" ],
            "action": "ACCEPT"
            "sequence_id": 0
         }
      ],
      "announce_filter": [
         {
            "prefixes": [ "1.100.0.0/16" ],
            "action": "ACCEPT",
            "sequence_id": 0
         }
      ]
   }
}

ISD-specific default Domain

This example shows the definition of a default domain. The default domain can be thought of as the default route in classical IP routing and is entirely optional. While the default domain cannot limit the set of accepted IP prefixes, it is perfectly possible to limit the set of announced IP prefixes, as well as the set of remote ASes.

{
   "default": true,
   "remote_isd_ases": [
      {
         "isd_as": "1-0"
         "action": "ACCEPT"
         "description": "Only ASes in ISD 1 are part of the domain"
         "sequence_id": 0
      },
   ],
   "prefixes": {
      "announce_filter": [
         {
            "prefixes": [ "1.100.0.0/16" ],
            "action": "ACCEPT",
            "sequence_id": 0
         }
      ]
   }
}

Traffic Policies

Traffic policies let an operator influence how traffic is routed within a domain. Each domain can have one or multiple traffic policies. A traffic policy contains a traffic matcher that defines the set of packets to which the policy applies and a set of path filters (grouped into a failover sequence) that defines what paths are allowed to be used by the matched packets.

A path filter defines a set of rules that a path must satisfy to be accepted by the filter. To select a path for the tunnel, the following steps are taken:

  1. The path filters are traversed in order of their sequence ID within the failover sequence.

  2. Each path filter is applied to the candidate set of paths.

  3. The best path out of the acceptable set is chosen according to path health and performance metrics. If no such path exists, i.e., no candidate path is allowed by the filter or no path is healthy, selection continues to the next path filter in the failover sequence.

  4. If all filters have been considered and no healthy path was selected, traffic will be dropped.

Failover Sequence

To illustrate the failover sequence consider the example below. First, the path filter provider_1_outgoing is applied to the set of all available paths leading to an acceptable set of paths. If one or more paths of the acceptable set are healthy, the best path is chosen according to its performance metrics. If no path of the acceptable set is healthy, the path filter allow_all is applied to the set of all available paths. This leads to a (potentially) different acceptable set. Again, the best healthy path is chosen and if no such path exists, the traffic is dropped.

{
   "failover_sequence": [
      {
         "sequence_id": 0,
         "path_filter": "provider_1_outgoing"
      },
      {
         "sequence_id": 1,
         "path_filter": "allow_all"
      },
   ]
}

Traffic policies are applied in the order of their sequence ID (ascending). For each IP packet, only a single traffic policy will be applied even if it matches multiple policies. A traffic policy matches if the traffic matcher defined in the policy matches the incoming IP packet.

While only a single traffic policy applies to an incoming IP packet, within the traffic policy the failover sequence can be used to configure “failover” behavior. If a path filter in the failover sequence has no healthy path to the remote AS, the next filter will be tried in the order defined by the failover sequence.

Traffic matchers and path filters are referenced by name. This allows to easily reuse these objects in different traffic policies. The traffic matchers are defined in Traffic Matchers and the path filters are defined in Path Filters.

Examples

Traffic policy limited to a single ISD

This example shows a configuration with a single traffic policy that configures all traffic to only use paths within the Swiss isolation domain. Refer to the path filters and traffic matchers examples to see how the corresponding filters and matchers are defined.

{
   "traffic_policies": [
      {
         "sequence_id": 0,
         "name": "CH ISD only",
         "description": "No traffic leaves the Swiss ISD.",
         "traffic_matcher": "all",
         "failover_sequence": [
            {
               "sequence_id": 0,
               "path_filter": "ch_isd_only"
            }
         ]
      }
   ]
}

Traffic engineering for priority traffic

This example shows a configuration with two traffic policies. The first policy chooses paths via provider 1 for priority traffic (defined as traffic tagged with the EF DSCP flag) and falls back to paths via provider 2 if no paths via provider 1 are available. The second policy chooses paths via provider 2 for all non-priority traffic. If no paths are available via provider 2 no fallback happens and traffic is dropped. Refer to the path filters and traffic matchers examples to see how the corresponding filters and matchers are defined.

{
   "traffic_policies": [
      {
         "sequence_id": 0,
         "name": "Priority traffic"
         "description": "Priority traffic is sent preferred via provider 1",
         "traffic_matcher": "ef_tagged",
         "failover_sequence": [
            {
               "sequence_id": 0,
               "path_filter": "provider_1_outgoing"
            },
            {
               "sequence_id": 1,
               "path_filter": "allow_all"
            },
         ]
      },
      {
         "sequence_id": 1,
         "name": "Non-priority traffic"
         "description": "Non-priority traffic is sent preferred via provider 2",
         "traffic_matcher": "all",
         "failover_sequence": [
            {
               "sequence_id": 0,
               "path_filter": "provider_2_outgoing"
            },
         ]
      },
   ]
}

Path Filters

A path filter defines a set of paths by applying the filter to all available paths. The filter is identified by a unique name and can contain an Access Control List (ACL) and a hop pattern both of which operate on hop predicates.

A hop predicate defines a hop of a SCION path and has the form <isd>-<as>#<interface>, where <isd> is the SCION ISD identifier, <as> is the SCION AS identifier, and <interface> is the interface identifier of a SCION path hop, i.e., the SCION link in the given SCION AS. The hop predicate can be fully or partially qualified, i.e., all entries of the hop predicate are optional or can include wildcards (0). A single 0 matches every SCION path hop.

An ACL consists of a list of ACL entries. Each ACL entry has the form <action> <hop-predicate>, where the action can either be accept (+) or deny (-). The hop predicate is optional. If no hop predicate is specified, the action matches every hop, i.e., a single + is the default accept action and a single - is the default deny action. The ACL is applied by sequentially applying all ACL entries to paths. If the ACL is empty, it defaults to accepting all paths.

A hop pattern defines a sequence of hop predicates that a path has to match to be accepted. It can be thought of as a regular expression on SCION paths. Each hop predicate can optionally be extended with a modifier * or +. The * modifier means 0 or more occurrences. The + means one or more occurrences.

Path Filters

{
   "path_filters": [
      {
         "name": "allow_all",
         "description": "Allow all paths",
         "acl": [
            "+ 0"
         ]
      },
      {
         "name": "ch_isd_only",
         "description": "Only allow paths within the Swiss ISD (64)",
         "acl": [
            "+ 64-0",
            "-"
         ]
      },
      {
         "name": "provider_1_outgoing",
         "description": "Paths that leave the local AS through SCION interface 1",
         "hop_pattern": "1-ff00:1:111#1 0*"
      },
      {
         "name": "provider_1_incoming",
         "description": "Paths that enter the local AS through SCION interface 1",
         "hop_pattern": "0* 1-ff00:1:111#1"
      },
      {
         "name": "provider_2_outgoing",
         "description": "Paths that leave the local AS through SCION interface 2",
         "hop_pattern": "1-ff00:1:111#2 0*"
      }
      {
         "name": "provider_2_incoming",
         "description": "Paths that enter the local AS through SCION interface 2",
         "hop_pattern": "0* 1-ff00:1:111#2"
      }
   ]
}

Traffic Matchers

A traffic matcher is used to classify traffic in traffic policies. Each packet is classified based on configured traffic matchers and put in a traffic class. A traffic class is used in a traffic policy to map a failover sequence to a traffic class.

A traffic matcher consists of a unique name (name) and a condition (condition). The condition defines the matching criteria for the traffic matcher. It is expressed as a boolean expression that then evaluates to either true - the IP packet matches the traffic matcher - or false - the IP packet does not match the traffic matcher. The expression consists of atoms and combinators. The following atoms are supported:

BOOL=<true|false>

Evaluates to the specified boolean value. This can be used to define a traffic matcher that always matches: BOOL=true matches all packets. On the other hand, BOOL=false matches no packets.

SRC=<IP-prefix>

Evaluates to true if IP-prefix contains the source IP address of the packet, otherwise false.

DST=<IP-prefix>

Evaluates to true if IP-prefix contains the destination IP address of the packet, otherwise false.

SRCPORT=<port-range>

Evaluates to true if the source port of the packet is in the range port-range, otherwise false. This applies to TCP and UDP packets.

DSTPORT=<port-range>

Evaluates to true if the destination port of the packet is in the range port-range, otherwise false. This applies to TCP and UDP packets.

PROTOCOL=<protocol>

Evaluates to true if the L4 protocol of the packet is protocol, otherwise false. Valid strings for protocol are tcp, udp, and icmp.

DSCP=<dscp-value>

Evaluates to true if the DSCP value of the packet is dscp-value, otherwise false. The DSCP value must be specified as a hexadecimal number of the form 0xYY.

The following combinators are supported:

ANY(cond1, cond2, ...)

Evaluates to true if any of the cond conditions evaluates to true, otherwise false.

ALL(cond1, cond2, ...)

Evaluates to true if all of the cond conditions evaluate to true, otherwise false.

NOT(cond)

Evaluates to true if cond evaluates to false and vice versa.

Traffic Matchers

{
   "traffic_matchers": [
      {
         "name": "all",
         "description": "All traffic",
         "condition": "BOOL=true"
      },
      {
         "name": "ef_tagged",
         "description": "Traffic tagged as Expedited Forwarding",
         "condition": "DSCP=0x2e"
      },
      {
         "name": "site1_to_site2",
         "description": "Traffic from Site 1 to Site 2.",
         "condition": "ALL(SRC=10.1.0.0/24, DST=10.2.0.0/24)"
      }
   ]
}

Local Endpoint

The local endpoint configuration defines the local IP-in-SCION tunneling endpoint. The local endpoint configuration contains the local network endpoint on which IP packets are received and encapsulated. In addition to the local ip (host) address, a set of ports for data, control, and probing purposes can be configured. However, these ports do not need to be set explicitly as the Anapaya EDGE appliance will automatically assign them if left empty. Both, IPv4 and IPv6 addresses can be configured. Furthermore, SCION RSS can be activated by setting enable_scion_rss if the local routers support it. With SCION RSS activated, the gateway utilizes UDP source port entropy on the underlay to enable RSS for SCION traffic, which can drastically increase forwarding performance.

Additionally, the local endpoint configuration contains a set of allowed_interfaces for each ISD-AS the appliance is part of, to be used by the IP-in-SCION tunnels involving this appliance. This can be used to control incoming traffic. For example, if an Anapaya EDGE appliance should only be reachable via links that enter the local site for ISD-AS 1-ff00:1:1 and those links correspond to SCION interfaces 1 and 2, allowed_interfaces should list them explicitly. Remote tunnel endpoints will then only choose paths entering the local AS via SCION interface 1 or 2. An empty list means all interfaces can be used by remote IP-in-SCION tunneling endpoints.

Note

Please note that if you specify a set of allowed interfaces for an appliance, you need to ensure that the allowed interfaces are also included in the Cluster section of the peer appliances’ configuration.

Local Endpoint Configuration

Local endpoint configuration including the host IP address to bind to and limiting the set of interfaces that can be used for the incoming IP-in-SCION tunnels.

{
   "endpoint": {
      "enabled": true,
      "ip": "10.8.0.1",
      "allowed_interfaces": [
         {
            "interfaces": [
                  1,
                  2
            ],
            "isd_as": "1-ff00:0:1"
         }
      ],
      "enable_scion_rss": true
   }
}

Remote Endpoints

Remote IP-in-SCION tunneling endpoints are automatically discovered within their respective SCION AS. However, the set of remote ASes which should be considered for remote endpoint discovery needs to be statically provided. The remotes section contains a list of all remote SCION ASes towards which the local endpoint should establish IP-in-SCION tunnels.

Note

Routing Domains also include remote_isd_ases, however, that section is used to limit the set of remote ASes to which a domain applies and can include wildcards as well as allow and block lists. If a remote SCION AS is not listed as part of the remotes section, it will not be considered during remote endpoint discovery.

Remotes

Consider ASes 1-ff00:0:1 and 2-ff00:0:2 during remote endpoint discovery.

{
   "remotes": [
      {
         "isd_as": "1-ff00:0:1",
         "description": "Branch office in Zurich"
      },
      {
         "isd_as": "2-ff00:0:2",
         "description": "Branch office in London"
      }
   ]
}

Static Announcements

The IP prefixes that will be statically announced are defined in the static_announcements section. Each entry defines a list of prefixes to announce.

Note

The set of prefixes that are announced to any given remote tunnel endpoint is not necessarily the statically defined prefixes list. Configured announce-filters in each domain can limit the set of announced prefixes in a given domain.

Optionally, next-hop tracking can be enabled for a set of prefixes. If next-hop tracking is enabled, the prefixes are only distributed if the specified target responds to ICMP ECHO requests. This can be used to implement dynamically retractable routes without having to resort to a dynamic routing protocol.

Static IP Prefixes

Statically announce 10.8.0.2/32 and 10.8.0.5/32 if 10.8.0.2 is reachable and unconditionally announce 10.8.0.1/32.

{
   "static_announcements": [
      {
         "next_hop_tracking": {
            "enabled": true,
            "target": "10.8.0.2"
         },
         "prefixes": [
            "10.8.0.2/32",
            "10.8.0.5/32"
         ],
         "sequence_id": 0
      },
      {
         "prefixes": [
            "10.8.0.1/32"
         ],
         "sequence_id": 1
      },
   ]
}

Use Cases

This section contains common IP-in-SCION tunneling use cases and how they can be implemented using the Anapaya EDGE appliance’s domain-based routing configuration. There are of course many more use cases and implementation options. The examples below are intended to illustrate common use cases and should serve the reader as a basis for implementation and adaptation.

Company WAN and Public Cloud SaaS Access

../../../_images/scion_tunneling_usecase11.svg

In this use case, the operator has two Anapaya EDGE appliances (edge1 and edge2) in two local data center sites, each with one SCION access link to an upstream SCION ISP. This is a typical deployment for a regional or global headquarter site (HQ Zurich). The operator has to consider the company’s remote sites (Branch Paris and Branch London) that are part of the company’s wide area network (WAN). Furthermore, the operator wants to configure access to two Software-as-a-Service (SaaS) applications that run in a public cloud. As the public cloud is not directly SCION-enabled, the company makes use of two service providers that provide SCION-enabled access to the public cloud (Cloud Access Provider). Please refer to the figure when working through the examples.

This use case can be naturally implemented with two domains: one domain for the company’s WAN and a separate domain to configure public cloud access.

Domains Configuration Skeleton

"domains": [
    {
        "name": "company_wan",
        "description": "The company's wide area network.",
    },
    {
        "name": "public_cloud",
        "description": "Access to SaaS applications in a public cloud.",
    }
]

Domain Entities

The first step is to define what entities are part of the domains, i.e., announced and accepted IP prefixes and remote SCION ASes.

Let us first focus on the company_wan domain and assume the following requirements:

  • Possible IP ranges for the company’s WAN: 10.0.0.0/8 and 192.168.0.0/16.

  • The local sites should only announce prefixes in 10.1.0.0/16 and 10.2.0.0/16

  • Only the remote sites 2-ff00:0:2 (Branch Paris) and 2-ff00:0:3 (Branch London) should be part of the company_wan domain.

  • IP-in-SCION tunnels should be established to both these branches from HQ Zurich.

Putting all this together, we get the following configuration for the company_wan domain:

Company WAN Domain with Configured Remotes

"scion_tunneling": {
    "domains": [
        {
            "name": "company_wan",
            "description": "The company's wide area network.",
            "prefixes": {
                "accept_filter": [
                    {
                        "prefixes": ["10.0.0.0/8", "192.168.0.0/16"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ],
                "announce_filter": [
                    {
                        "prefixes": ["10.1.0.0/16", "10.2.0.0/16"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ]
            },
            "remote_isd_ases": [
                {
                    "isd_as": "2-ff00:0:2",
                    "action": "ACCEPT",
                    "sequence_id": 0
                },
                {
                    "isd_as": "2-ff00:0:3",
                    "action": "ACCEPT",
                    "sequence_id": 1
                }
            ]
        }
    ],
    "remotes": [
        {
            "isd_as": "2-ff00:0:2",
            "description": "Branch office in Paris"
        },
        {
            "isd_as": "2-ff00:0:3",
            "description": "Branch office in London"
        }
    ]
}

Next, the operator needs to define the what entities that are part of the public_cloud domain. The following requirements are assumed:

  • The public cloud is accessible via the Cloud Access Providers (CAP) 2-ff00:0:100, 1-ff00:0:200, and 2-ff00:0:200. Note that the latter two ASes are distinct on the SCION level, but obviously they belong to the same CAP.

  • The IP prefix assigned app1 is 1.0.1.0/24 and the one assigned to app2 is 1.0.2.0/24, therefore these prefixes must be accepted.

  • All internal hosts are NATed to an IP in 2.0.0.0/24, i.e., this is the IP prefix that needs to be announced in the public_cloud domain, such that the return traffic can be correctly routed from the public cloud to the local site.

Putting all this together, we get the following configuration for the public_cloud domain:

Public Cloud Domain Entities

"scion_tunneling": {
    "domains": [
        {
            "name": "public_cloud",
            "description": "Access to SaaS applications in a public cloud.",
            "prefixes": {
                "accept_filter": [
                    {
                        "prefixes": ["1.0.1.0/24", "1.0.2.0/24"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ],
                "announce_filter": [
                    {
                        "prefixes": ["2.0.0.0/24"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ]
            },
            "remote_isd_ases": [
                {
                    "isd_as": "1-ff00:0:100",
                    "action": "ACCEPT",
                    "sequence_id": 0
                },
                {
                    "isd_as": "0-ff00:0:200",
                    "action": "ACCEPT",
                    "sequence_id": 1
                }
            ]
        }
    ],
    "remotes": [
        {
            "isd_as": "1-ff00:0:100",
            "description": "Access provider 1 to public cloud"
        },
        {
            "isd_as": "1-ff00:0:200",
            "description": "Access provider 2 to public cloud (in ISD 1)"
        },
        {
            "isd_as": "2-ff00:0:200",
            "description": "Access provider 2 to public cloud (in ISD 2)"
        }
    ]
}

Note the use of the ISD wildcard 0-ff00:0:200 in the remote_isd_ases section. This matches the AS identifier ff00:0:200 in every ISD, however, since the operator only explicitly adds the ASes 1-ff00:0:200 and 2-ff00:0:200 in the remotes section, only those will be used to establish tunnels with.

Announced Prefixes

The prefixes defined in the announce_filter of the domains are not the prefixes that are necessarily announced. Instead, these is a filter applied on statically configured and dynamically learned prefixes (via BGP). In this example, both Anapaya EDGE appliances are connected internally to a BGP peer that announces prefixes to and accepts prefixes from the Anapaya EDGE appliances. To learn how to configure BGP peers on an Anapaya EDGE appliance, please refer to the Border Gateway Protocol (BGP) section.

Traffic Policies

Besides defining the entities that are part of the domains, the operator also needs to configure at least one traffic policy per domain.

Note

Often, a permissive traffic policy, i.e., a traffic policy that does not limit the set of acceptable SCION paths too much, is advantageous. The Anapaya EDGE appliance has a powerful path optimization engine that can ensure optimal path selection amongst the set of acceptable paths.

The requirements are as follows:

  • Generally, reliability and performance should be maximized.

  • app1 must only be accessible via paths within ISD 1 as it deals with highly sensitive and regulated data.

  • There are no specific requirements for app2.

Note

The operator has little influence over the traffic policies used by the remote endpoints. To ensure that the traffic for app1 also only uses paths within ISD 1 on the return path, the operator needs to coordinate that with the remote endpoint. If the operator also controls the remote endpoint then this is not an issue.

Given these requirements, the operator first needs to define three traffic matchers: one that matches traffic belonging to app1, one that matches traffic belonging to app2, and a third that matches any traffic.

Traffic Matchers

"traffic_matchers": [
    {
        "name": "app1",
        "description": "Traffic belonging to app1",
        "condition": "DST=1.0.1.0/24"
    },
    {
        "name": "app2",
        "description": "Traffic belonging to app2",
        "condition": "DST=1.0.2.0/24"
    },
    {
        "name": "all",
        "description": "All traffic",
        "condition": "BOOL=true"
    }
]

Furthermore, two path filters need to be defined - one that only allows paths within ISD 1 and one that allows all paths.

Path Filters

"path_filters": [
    {
        "name": "isd_1_only",
        "description": "Only allow paths within ISD 1",
        "acl": ["+ 1-0", "-"]
    },
    {
        "name": "allow_all",
        "description": "Allow all paths",
        "acl": ["+ 0"]
    }
]

Having defined the traffic matchers and path filters, the operator can now assemble the traffic policies for the domains.

Traffic Policies

"domains": [
    {
        "name": "company_wan",
        "traffic_policies": [
            {
                "sequence_id": 0,
                "description": "Allow all traffic to use all available paths.",
                "traffic_matcher": "all",
                "failover_sequence": [
                    {
                        "sequence_id": 0,
                        "path_filter": "allow_all"
                    }
                ]
            }
        ]
    },
    {
        "name": "public_cloud",
        "traffic_policies": [
            {
                "sequence_id": 0,
                "description": "Traffic belonging to app1 can only use paths within ISD 1.",
                "traffic_matcher": "app1",
                "failover_sequence": [
                    {
                        "sequence_id": 0,
                        "path_filter": "isd_1_only"
                    }
                ]
            },
            {
                "sequence_id": 1,
                "description": "Traffic belonging to app2 can use all paths",
                "traffic_matcher": "app2",
                "failover_sequence": [
                    {
                        "sequence_id": 0,
                        "path_filter": "allow_all"
                    }
                ]
            }
        ]
    }
]

Local Endpoint

To finalize the IP-in-SCION tunneling configuration, the operator needs to configure the local tunneling endpoint for each Anapaya EDGE appliance. This is the part of the configuration that will be different for each appliance. Besides the host address to which the tunneling endpoint is bound, the operator also wants to control incoming traffic such that traffic coming via SCION interface 1 is always handled by edge1, and traffic coming via SCION interface 2 is always handled by edge2. To that end, the operator needs to specify allowed_interfaces accordingly.

Local Endpoint Configuration

edge1:

"endpoint": {
    "ip": "10.1.0.1",
    "allowed_interfaces": [
        {
           "interfaces": [
                 1
           ],
           "isd_as": "1-ff00:1:1"
        }
    ],
    "enabled": true
},

edge2:

"endpoint": {
   "ip": "10.1.0.2",
   "allowed_interfaces": [
      {
         "interfaces": [
               2
         ],
         "isd_as": "1-ff00:1:1"
      }
   ],
   "enabled": true
}

Complete Configuration

For completeness the entire scion_tunneling configurations for edge1 is shown below. The one for edge2 only differs in the endpoint section.

Multi-ISD Anapaya EDGE Setup

../../../_images/scion_tunneling_usecase21.svg

This example shows a single Anapaya EDGE appliance (edge) that is part of two ISDs. One use case for this setup is having an appliance that is part of an ISD that is completely disconnected from the rest of the public SCION Internet, but needs to be remotely managed. In this case, a second ISD that is connected to the public SCION Internet needs to be configured on the appliance, such that it can be remotely managed. Another use case is if the Anapaya EDGE appliance has two upstream SCION ISPs that are in different ISDs themselves. This is only possible if the Anapaya EDGE appliance is also part of both those ISDs.

The following example shows an Anapaya EDGE appliance that is part of two separate and isolated ISDs (ISD 1 and ISD 2). The requirements are as follows:

  • Only IP prefixes in 1.0.0.0/8 are announced and accepted to and from ISD 1.

  • Only IP prefixes in 2.0.0.0/8 are announced and accepted to and from ISD 2.

  • Only remote ASes that are part of the corresponding ISD can announce these prefixes to the local Anapaya EDGE appliance.

  • The Anapaya EDGE appliance should use its local AS identity that corresponds the ISD when communicating with other IP-in-SCION tunneling endpoints in the same ISD.

  • IP prefix 1.0.1.0/24 should be announced to ISD 1 and 2.0.1.0/24 to ISD 2 if (and only if) internally the firewall at 192.168.2.100 is reachable.

Given these requirements, the operator defines two routing domains: one for ISD 1 and one for ISD 2.

Domains Configuration Skeleton

"domains": [
    {
        "name": "isd_1",
        "description": "Domain for communication in ISD 1.",
    },
    {
        "name": "isd_2",
        "description": "Domain for communication in ISD 2.",
    }
]

Domain Entities

The first step is to define what entities are part of the domains, i.e., announced and accepted IP prefixes and remote SCION ASes.

Let us first focus on the isd_1 domain. As described above, the operator wants to ensure that only IP prefixes in 1.0.0.0/8 are announced to and accepted from remote IP-in-SCION tunneling endpoints in ISD 1. Furthermore, the operator wants to limit the set of remote ASes to those that are part of ISD 1 and use the local AS identity in ISD 1 for communication in this domain. Finally, the remote site Remote 1 (1-ff00:0:2) needs to be explicitly added to the remotes section such that the Anapaya EDGE appliance will establish IP-in-SCION tunnels to endpoints in Remote 1.

ISD1 Domain with Configured Remotes

"scion_tunneling": {
    "domains": [
        {
            "name": "isd_1",
            "description": "Domain for communication in ISD 1.",
            "prefixes": {
                "accept_filter": [
                    {
                        "prefixes": ["1.0.0.0/8"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ],
                "announce_filter": [
                    {
                        "prefixes": ["1.0.0.0/8"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ]
            },
            "local_isd_ases": ["1-ff00:0:1"],
            "remote_isd_ases": [
                {
                    "isd_as": "1-0",
                    "action": "ACCEPT",
                    "sequence_id": 0
                }
            ]
        }
    ],
    "remotes": [
        {
            "isd_as": "1-ff00:0:2",
            "description": "Remote AS in ISD 1"
        }
    ]
}

The isd_2 domain is configured similarly, with the straightforward adaption of the configuration to the requirements of ISD 2.

ISD2 Domain with Configured Remotes

"scion_tunneling": {
    "domains": [
        {
            "name": "isd_2",
            "description": "Domain for communication in ISD 2.",
            "prefixes": {
                "accept_filter": [
                    {
                        "prefixes": ["2.0.0.0/8"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ],
                "announce_filter": [
                    {
                        "prefixes": ["2.0.0.0/8"],
                        "action": "ACCEPT",
                        "sequence_id": 0
                    }
                ]
            },
            "local_isd_ases": ["2-ff00:0:1"],
            "remote_isd_ases": [
                {
                    "isd_as": "2-0",
                    "action": "ACCEPT",
                    "sequence_id": 0
                }
            ]
        }
    ],
    "remotes": [
        {
            "isd_as": "2-ff00:0:3",
            "description": "Remote AS in ISD 2"
        }
    ]
}

Announced Prefixes

As described in the introduction, the Anapaya EDGE appliance should announce the local prefixes 1.0.1.0/24 in ISD 1 and 2.0.1.0/24 in ISD 2 if internally the firewall at 192.168.2.100 is reachable. This can be accomplished by configuring the static_announcements section with next_hop_tracking activated. Note that the static_anouncements section contains both IP prefixes. The announce_filter in the corresponding domain configurations ensure that 1.0.1.0/24 is announce in ISD 1 and 2.0.1.0/24 in ISD 2.

Static Announcements

"static_announcements": [
    {
        "next_hop_tracking": {
            "enabled": true,
            "target": "192.168.2.100"
        },
        "prefixes": ["1.0.1.0/24", "2.0.1.0/24"],
        "sequence_id": 0
    }
]

Traffic Policies

Besides defining the entities that are part of the domains, the operator also needs to configure at least one traffic policy per domain.

Note

Often, a permissive traffic policy, i.e., a traffic policy that does not limit the set of acceptable SCION paths too much, is advantageous. The Anapaya EDGE appliance has a powerful path optimization engine that can ensure optimal path selection amongst the set of acceptable paths.

In this example, the operator wants to ensure that traffic for destinations in ISD 1 only uses paths in ISD 1 and similarly for destinations in ISD 2. Besides that, the operator does not want to further limit the set of acceptable paths to give the Anapaya EDGE appliance the biggest possible path selection scope.

Note

If the two ISDs are truly isolated from the rest of the SCION Internet, limiting the set of acceptable paths to those in the same ISD is not required as no other paths exist anyway. Nevertheless, it is good practice to be specific about the set of acceptable paths.

To implement these requirements, the operator first needs to define one traffic matcher to match all traffic and two path filters to match paths in ISD 1 and ISD 2 respectively.

Traffic Matchers and Path Filters

"traffic_matchers": [
    {
        "name": "all",
        "description": "All traffic",
        "condition": "BOOL=true"
    }
],
"path_filters": [
    {
        "name": "isd_1_only",
        "description": "Only allow paths within ISD 1",
        "acl": ["+ 1-0", "-"]
    },
    {
        "name": "isd_2_only",
        "description": "Only allow paths within ISD 2",
        "acl": ["+ 2-0", "-"]
    }
]

Using the defined traffic matcher and path filters, the operator can define the traffic policies for each domain:

Traffic Policies

"domains": [
    {
        "name": "isd_1",
        "traffic_policies": [
            {
                "sequence_id": 0,
                "description": "All traffic must only use paths within ISD 1.",
                "traffic_matcher": "all",
                "failover_sequence": [
                    {
                        "sequence_id": 0,
                        "path_filter": "isd_1_only"
                    }
                ]
            }
        ]
    },
    {
        "name": "isd_2",
        "traffic_policies": [
            {
                "sequence_id": 0,
                "description": "All traffic must only use paths within ISD 2.",
                "traffic_matcher": "all",
                "failover_sequence": [
                    {
                        "sequence_id": 0,
                        "path_filter": "isd_2_only"
                    }
                ]
            }
        ]
    }
]

Local Endpoint

To finalize the IP-in-SCION tunneling configuration, the operator needs to configure the local tunneling endpoint for the Anapaya EDGE appliance. Since the operator does not need to control incoming traffic, the only required configuration is to specify the host address to bind to and to enable the IP-in-SCION tunneling endpoint.

Local Endpoint Configuration

"endpoint": {
    "ip": "192.168.2.1",
    "enabled": true
},

Complete Configuration

For completeness the entire scion_tunneling configurations for edge is shown below.

Configuring Egress Source NAT if only few public IP addresses are available

In some cases, the available IP addresses that an appliance can announce to remote IP-in-SCION tunneling endpoints are limited, e.g., if a customer only has a single public IP address assigned by its ISP. To allow multiple internal hosts to share a single IP address, the appliance can be configured to perform source NATing on the outgoing traffic. This way, the appliance can use a single IP address to represent multiple internal hosts.

Note

It is possible and also used in practice to announce private IP ranges to a remote IP-in-SCION tunneling endpoint. However, sometimes it is simpler to limit the set of announced prefixes to public IP ranges because then there is no coordination needed with the remote network to ensure that the announced IP ranges do not clash with the remote network’s internal IP ranges.

Let us assume that the operator has only a single public IP address available 203.0.113.50/32 but wants to connect multiple internal hosts to the remote while only announcing the single public IP address. Let us further assume that the internal hosts are in the range 192.168.1.0/24. The operator can use the following configuration to achieve this:

Egress NAT Configuration

"interfaces": {
   "ethernets": [
         {
            "name": "lan0",
            "addresses": ["192.168.1.1/24"],
            "driver": "VPP_DPDK"
         },
         {
            "name": "wan0",
            "addresses": ["169.254.0.2/31"],
            "driver": "VPP_DPDK"
         }
      }
   ],
   "loopbacks": [
      {
         "name": "loop0",
         "addresses": ["203.0.113.50/32"]
      }
   ]
},
"nat": {
   "snat": {
      "address_pool": ["203.0.113.50/32"],
      "exclude": [],
      "interfaces": ["scion-gateway"]
   }
}

The lan0 interface is configured with an address of the internal private IP range. Note that the public IP address is added on a loopback interface and NOT on the wan0 interface. This is because on the wan0 interface we have the IP underlay for the SCION access link (usually provided by the upstream SCION ISP). The public IP address is only used for IP-in-SCION tunneling.

The NAT configuration specifies the public IP address that should be used for source NATing and the interface on which the NAT should be applied to. In this case the address pool to use is the single public IP address 203.0.113.50/32 and the NAT should be applied to the special scion-gateway interface to apply it to all IP-in-SCION tunneling traffic.

Warning

When using egress NAT it is not possible by default to create IP connections from a host in a remote AS to a host in the local AS that would go through the IP-in-SCION tunnel. Using NAT means that only outgoing connections are possible. If incoming connections are needed, the appliance needs to be configured to exclude the IP addresses that should not be NATed.

Configuring Ingress NAT to collect Users from the Anapaya GATE

../../../_images/scion_tunneling_usecase31.svg

In this use case, the operator wants to ensure that traffic streams coming from an IP-in-SCION tunnel from a remote Anapaya GATE are returned via the local Anapaya EDGE appliance, even though the remote IP prefixes are also reachable via the legacy Internet.

To address this use case, the EDGE appliance can do NATing on the incoming traffic. Source addresses in packets arriving from the tunnel are rewritten by one of the addresses from the NAT’s address pool.

The operator is responsible for setting up the static routes so that replies to the addresses in the NAT’s address pool are routed back to the appliance. This way the operator can ensure that replies to the packets arriving via the tunnel are routed back to the tunnel, even if the remote IP prefixes are also reachable via the legacy Internet.

If, on the other hand, a packet arrives through the Internet it keeps its original source address and the reply is routed back through the legacy Internet. Again, it is up to the operator to set up the routes in the local network accordingly.

Ingress NAT Configuration

"nat": {
   "snat": {
      "address_pool": ["192.168.0.1/32"],
      "exclude": [],
      "interfaces": ["eth0"]
   }
}

Note

Ingress NAT does not work with BGP enabled in the appliance.

Warning

When using ingress NAT it is not possible by default to create connections from the local AS to the remote AS that would go through the IP-in-SCION tunnel. Using NAT means that there are no fixed IP addresses to connect to. If this is needed, a specific IP address, or a range of addresses, can be excluded from the NAT. The packet will then arrive from the tunnel with its original source address unmodified. In this case it’s up to the operator to route the packets destined to that address to the SCION tunnel rather than routing them through the legacy Internet.