RFC 7971
Application-Layer Traffic Optimization (ALTO) Deployment Considerations
Pages: 77
Informational
Informational
Part 2 of 4 – Pages 15 to 36
3. Deployment Considerations by ISPs
3.1. Objectives for the Guidance to Applications
3.1.1. General Objectives for Traffic Optimization
The Internet consists of many networks. The networks are owned and managed by different network operators, such as commercial ISPs, enterprise IT departments, universities, and other organizations. These network operators provide network connectivity, e.g., by access networks, such as cable networks, xDSL networks, 3G/4G mobile networks, etc. Network operators need to manage, control, and audit the traffic. Therefore, it is important to understand how to deploy an ALTO service and what its expected impact might be. The general objective of ALTO is to give guidance to applications on what endpoints (e.g., IP addresses or IP prefixes) are to be preferred according to the operator of the ALTO server. The ALTO protocol gives means to let the ALTO server operator express its preference, whatever this preference is. ALTO enables network operators to support application-level traffic engineering by influencing application resource provider selection. This traffic engineering can have different objectives: 1. Inter-network traffic localization: ALTO can help to reduce inter-domain traffic. The networks of different network operators are interconnected through peering points. From a business view, the inter-network settlement is needed for exchanging traffic between these networks. These peering agreements can be costly. To reduce these costs, a simple objective is to decrease the traffic exchange across the peering points and thus keep the traffic in the own network or Autonomous System (AS) as far as possible. 2. Intra-network traffic localization: In case of large network operators, the network may be grouped into several networks, domains, or ASes. The core network includes one or several backbone networks, which are connected to multiple aggregation, metro, and access networks. If traffic can be limited to certain areas such as access networks, this decreases the usage of backbone and thus helps to save resources and costs. 3. Network offloading: Compared to fixed networks, mobile networks have some special characteristics, including lower link bandwidth, high cost, limited radio frequency resource, and limited terminal battery. In mobile networks, wireless links should be used efficiently. For example, in the case of a P2P
service, it is likely that hosts should prefer retrieving data
from hosts in fixed networks, and avoid retrieving data from
mobile hosts.
4. Application tuning: ALTO is also a tool to optimize the
performance of applications that depend on the network and
perform resource provider selection decisions among network
endpoints; an example is the network-aware selection of CDN
caches.
In the following, these objectives are explained in more detail with
examples.
3.1.2. Inter-Network Traffic Localization
ALTO guidance can be used to keep traffic local in a network, for
instance, in order to reduce peering costs. An ALTO server can let
applications prefer other hosts within the same network operator's
network instead of randomly connecting to other hosts that are
located in another operator's network. Here, a network operator
would always express its preference for hosts in its own network,
while hosts located outside its own network are to be avoided (i.e.,
they are undesired to be considered by the applications). Figure 5
shows such a scenario where hosts prefer hosts in the same network
(e.g., Host 1 and Host 2 in ISP1 and Host 3 and Host 4 in ISP2).
,-------. +-----------+ ,---. ,-' `-. | Host 1 | ,-' `-. / ISP 1 ########|ALTO Client| / \ / # \ +-----------+ / ISP X \ | # | +-----------+ / \ \ ########| Host 2 | ; +----------------------------|ALTO Client| | | | `-. ,-' +-----------+ | | | `-------' | Inter- | | ,-------. +-----------+ : network | ; ,-' `########| Host 3 | \ traffic | / / ISP 2 # \ |ALTO Client| \ | / / # \ +-----------+ \ |/ | # | +-----------+ `-. ,-| \ ########| Host 4 | `---' +----------------------------|ALTO Client| `-. ,-' +-----------+ `-------' Legend: ### preferred "connections" --- non-preferred "connections" Figure 5: Inter-Network Traffic Localization Examples for corresponding ALTO maps can be found in Section 3.5. Depending on the application characteristics, it may not be possible or even desirable to completely localize all traffic.3.1.3. Intra-Network Traffic Localization
The previous section describes the results of the ALTO guidance on an inter-network level. In the same way, ALTO can also be used for intra-network localization. In this case, ALTO provides guidance on which internal hosts are to be preferred inside a single network (e.g., one AS). This application-level traffic engineering can reduce the capacity requirements in the core network of an ISP. Figure 6 shows such a scenario where Host 1 and Host 2 are located in an access net 1 of ISP 1 and connect via a low capacity link to the core of the same ISP 1. If Host 1 and Host 2 exchange their data with remote hosts, they would probably congest the bottleneck link.
Bottleneck ,-------. +-----------+ ,---. | ,-' `-. | Host 1 | ,-' `-. | / ISP 1 ########|ALTO Client| / \ | / (Access # \ +-----------+ / ISP 1 \| | net 1) # | +-----------+ / (Core V \ ########| Host 2 | ; network) +--X~~~X---------------------|ALTO Client| | | | `-. ,-' +-----------+ | | | `-------' | | | ,-------. +-----------+ : | ; ,-' `########| Host 3 | \ | / / ISP 1 # \ |ALTO Client| \ | / / (Access # \ +-----------+ \ |/ | net 2) # | +-----------+ `-. ,-X \ ########| Host 4 | `---' ~~~~~~~X---------------------|ALTO Client| ^ `-. ,-' +-----------+ | `-------' Bottleneck Legend: ### preferred "connections" --- non-preferred "connections" Figure 6: Intra-Network Traffic Localization In such a situation, the operator can guide the hosts to try local hosts in the same network islands first, avoiding or at least lowering the effect on the bottleneck link, as shown in Figure 6. The objective is to avoid bottlenecks by optimized endpoint selection at the application level. That said, it must be understood that ALTO is not a general-purpose method to deal with the congestion at the bottleneck.3.1.4. Network Offloading
Another scenario is offloading traffic from networks. This use of ALTO can be beneficial in particular in mobile networks. A network operator may have the desire to guide hosts in its mobile network to use hosts outside this mobile network. One reason could be that the wireless network or the mobile hosts were not designed for direct peer-to-peer communications between mobile hosts, and therefore, it makes sense for peers to fetch content from remote peers in other parts of the Internet.
,-------. +-----------+ ,---. ,-' `-. | Host 1 | ,-' `-. / ISP 1 +-------|ALTO Client| / \ / (Mobile | \ +-----------+ / ISP X \ | network) | | +-----------+ / \ \ +-------| Host 2 | ; #############################|ALTO Client| | # | `-. ,-' +-----------+ | # | `-------' | # | ,-------. : # ; ,-' `-. \ # / / ISP 2 \ \ # / / (Fixed \ \ #/ | network) | +-----------+ `-. ,-# \ / | Host 3 | `---' #############################|ALTO Client| `-. ,-' +-----------+ `-------' Legend: ### preferred "connections" --- non-preferred "connections" Figure 7: ALTO Traffic Network De-localization Figure 7 shows the result of such a guidance process where Host 2 prefers a connection with Host 3 instead of Host 1, as shown in Figure 5. A realization of this scenario may have certain limitations and may not be possible in all cases. For instance, it may require the ALTO server to distinguish mobile and non-mobile hosts based on their IP address. This may depend on mobility solutions and may not be possible or accurate. In general, ALTO is not intended as a fine- grained traffic engineering solution for individual hosts. Instead, it typically works on aggregates (e.g., if it is known that certain IP prefixes are often assigned to mobile users).3.1.5. Application Tuning
ALTO can also provide guidance to optimize the application-level topology of networked applications, e.g., by exposing network performance information. Applications can often run their own measurements to determine network performance, e.g., by active delay measurements or bandwidth probing, but such measurements result in overhead and complexity. Accessing an ALTO server can be a simpler
alternative. In addition, an ALTO server may also expose network information that applications cannot easily measure or reverse- engineer.3.2. Provisioning of ALTO Topology Data
3.2.1. High-Level Process and Requirements
A process to generate ALTO topology information typically comprises several steps. The first step is to gather information, which is described in the following section. The subsequent sections describe how the gathered data can be processed and which methods can be applied to generate the information exposed by ALTO, such as network and cost maps. Providing ALTO guidance can result in a win-win situation for network providers and users of the ALTO information. Applications possibly get a better performance, while the network provider has means to optimize the traffic engineering and thus its costs. Yet, there can be security concerns with exposing topology data. Corresponding limitations are discussed in Section 7.2. ISPs may have important privacy requirements when deploying ALTO, which have to be taken into account when processing ALTO topology data. In particular, an ISP may not be willing to expose sensitive operational details of its network. The topology abstraction of ALTO enables an ISP to expose the network topology at a desired granularity only, determined by security policies. With the ECS, the ALTO client does not have to implement any specific algorithm or mechanism in order to retrieve, maintain and process network topology information (of any kind). The complexity of the network topology (computation, maintenance and distribution) is kept in the ALTO server and ECS is delivered on demand. This allows the ALTO server to enhance and modify the way the topology information sources are used and combined. This simplifies the enforcement of privacy policies of the ISP. The ALTO Network and Cost Map Service expose an abstract view on the ISP network topology. Therefore, care is needed when constructing those maps in order to take privacy policies into account, as further discussed in Section 3.2.3. The ALTO protocol also supports further features such as endpoint properties, which could also be used to expose topology guidance. The privacy considerations for ALTO maps also apply to such ALTO extensions.
3.2.2. Data Collection from Data Sources
The first step in the process of generating ALTO information is to gather the required information from the network. An ALTO server can collect topological information from a variety of sources in the network and provides a cohesive, abstract view of the network topology to applications using an ALTO client. Topology data sources may include routing protocols, network policies, state and performance information, geolocation, etc. An ALTO server requires at least some topology and/or routing information, i.e., information about existing endpoints and their interconnection. With this information, it is in principle possible to compute paths between all known endpoints. Based on such basic data, the ALTO server builds an ALTO-specific network topology that represents the network as it should be understood and utilized by applications (resource consumers) at endpoints using ALTO services (e.g., Network and Cost Map Service or ECS). A basic dataset can be extended by many other information obtainable from the network. The ALTO protocol does not assume a specific network technology or topology. In principle, ALTO can be used with various types of addresses (Endpoint Addresses). [RFC7285] defines the use of IPv4/ IPv6 addresses or prefixes in ALTO, but further address types could be added by extensions. In this document, only the use of IPv4/IPv6 addresses is considered. The exposure of network topology information is controlled and managed by the ALTO server. ALTO abstract network topologies can be automatically generated from the physical or logical topology of the network, e.g., using "live" network data. The generation would typically be based on policies and rules set by the network operator. The maps and the guidance can significantly differ depending on the use case, the network architecture, and the trust relationship between ALTO server and ALTO client, etc. Besides the security requirements that consist of not delivering any confidential or critical information about the infrastructure, there are efficiency requirements in terms of what aspects of the network are visible and required by the given use case and/or application. The ALTO server operator has to ensure that the ALTO topology does not reveal any details that would endanger the network integrity and security. For instance, ALTO is not intended to leak raw Interior Gateway Protocol (IGP) or Border Gateway Protocol (BGP) databases to ALTO clients.
+--------+ +--------+ | ALTO | | ALTO | | Client | | Client | +--------+ +--------+ /\ /\ || || ALTO protocol || || \/ \/ +---------+ | ALTO | | Server | +---------+ : : : : : : : : +..........+ : : +..........+ Provisioning : : : : protocol : : : : +---------+ +---------+ +---------+ +---------+ | BGP | | I2RS | | PCE | | NMS | Potential | Speaker | | Client | | | | OSS | data sources +---------+ +---------+ +---------+ +---------+ ^ ^ ^ ^ | | | | Link-State I2RS TED Topology and traffic-related NLRI for data data data from SNMP, NETCONF, IGP/BGP RESTCONF, REST, IPFIX, etc. Figure 8: Potential Data Sources for ALTO As illustrated in Figure 8, the topology data used by an ALTO server can originate from different data sources: o Relevant information sources are IGPs or BGP. An ALTO server could get network routing information by listening to IGPs and/or peering with BGP speakers. For data collection, link-state protocols are more suitable since every router propagates its information throughout the whole network. Hence, it is possible to obtain information about all routers and their neighbors from one single router in the network. In contrast, distance-vector protocols are less suitable since routing information is only shared among neighbors. To obtain the whole topology with distance-vector routing protocols it is necessary to retrieve routing information from every router in the network. o [RFC7752] describes a mechanism by which link-state and Traffic Engineering (TE) information can be collected from networks and shared with external components using the BGP routing protocol. This is achieved using a new BGP Network Layer Reachability
Information (NLRI) encoding format. The mechanism is applicable
to physical and virtual IGP links and can also include TE data.
For instance, prefix data can be carried and originated in BGP,
while TE data is originated and carried in an IGP. The mechanism
described is subject to policy control.
o The Interface to the Routing System (I2RS) is a solution for state
transfer in and out of the Internet's routing system [RFC7921].
An ALTO server could use an I2RS client to observe routing-related
information. With the rise of Software-Defined Networking (SDN)
and a decoupling of network data and control plane, topology
information could also be fetched from an SDN controller. If I2RS
is used, [RFC7922] provides traceability for these interactions.
This scenario is not further discussed in the remainder of this
document.
o Another potential source of topology information could be a Path
Computation Element (PCE) [RFC4655]. Topology and traffic-related
information can be retrieved from the Traffic Engineering Database
(TED) and Label Switched Path Database (LSP-DB). This scenario is
not further discussed in the remainder of this document.
o An ALTO server can also leverage a Network Management System (NMS)
or an Operations Support System (OSS) as data sources. NMS or OSS
solutions are used to control, operate, and manage a network,
e.g., using the Simple Network Management Protocol (SNMP) or
Network Configuration Protocol (NETCONF). As explained for
instance in [RFC7491], the NMS and OSS can be consumers of network
events reported and can act on these reports as well as displaying
them to users and raising alarms. In addition, NMS and OSS
systems may have access to routing information and network
inventory data (e.g., links, nodes, or link properties not visible
to routing protocols, such as Shared Risk Link Groups).
Furthermore, Operations, Administration, and Maintenance (OAM)
information can be leveraged, including traffic utilization
obtained from IP Flow Information Export (IPFIX), event
notifications (e.g., via syslog), liveness detection (e.g.,
bidirectional forwarding detection, BFD). NMS or OSS systems also
may have functions to correlate and orchestrate information
originating from other data sources. For instance, it could be
required to correlate IP prefixes with routers (Provider, Provider
Edge, Customer Edge, etc.), IGP areas, VLAN IDs, or policies.
In the context of the provisioning protocol, topology information
could be modeled in a YANG data model [NETWORK-TOPO].
The data sources mentioned so far are only a subset of potential topology sources and protocols. Depending on the network type, (e.g., mobile, satellite network) different hardware and protocols are in operation to form and maintain the network. In general, it is challenging to gather detailed information about the whole Internet, since the network consists of multiple domains and in many cases it is not possible to collect information across network borders. Hence, potential information sources may be limited to a certain domain.3.2.3. Partitioning and Grouping of IP Address Ranges
ALTO introduces provider-defined network location identifiers called Provider-defined Identifiers (PIDs) to aggregate network endpoints in the Map Services. Endpoints within one PID may be treated as single entity, assuming proximity based on network topology or other similarity. A key use case of PIDs is to specify network preferences (costs) between PIDs instead of individual endpoints. It is up to the operator of the ALTO server how to group endpoints and how to assign PIDs. For example, a PID may denote a subnet, a set of subnets, a metropolitan area, a POP, an autonomous system, or a set of autonomous systems. This document only considers deployment scenarios in which PIDs expand to a set of IP address ranges (CIDR). A PID is characterized by a string identifier and its associated set of endpoint addresses [RFC7285]. If an ALTO server offers the Map Service, corresponding identifiers have to be configured. An automated ALTO implementation may use dynamic algorithms to aggregate network topology. However, it is often desirable to have a mechanism through which the network operator can control the level and details of network aggregation based on a set of requirements and constraints. This will typically be governed by policies that enforce a certain level of abstraction and prevent leakage of sensitive operational data. For instance, an ALTO server may leverage BGP information that is available in a network's service provider network layer and compute the group of prefix. An example being BGP communities, which are used in MPLS/IP networks as a common mechanism to aggregate and group prefixes. A BGP community is an attribute used to tag a prefix to group prefixes based on mostly any criteria (as an example, most ISP networks originate BGP prefixes with communities identifying the Point of Presence (PoP) where the prefix has been originated). These BGP communities could be used to map IP address ranges to PIDs. By an additional policy, the ALTO server operator may decide an
arbitrary cost defined between groups. Alternatively, there are algorithms that allow the dynamic computation of costs between groups. The ALTO protocol itself is independent of such algorithms and policies.3.2.4. Rating Criteria and/or Cost Calculation
An ALTO server indicates preferences amongst network locations in the form of abstract costs. These costs are generic costs and can be internally computed by the operator of the ALTO server according to its own policy. For a given ALTO network map, an ALTO cost map defines directional costs pairwise amongst the set of source and destination network locations defined by the PIDs. The ALTO protocol permits the use of different cost types. An ALTO cost type is defined by the combination of a cost metric and a cost mode. The cost metric identifies what the costs represent. The cost mode identifies how the costs should be interpreted, i.e., whether returned costs should be interpreted as numerical values or ordinal rankings. The ALTO protocol also allows the definition of additional constraints defining which elements of a cost map shall be returned. The ALTO protocol specification [RFC7285] defines the "routingcost" cost metric as the basic set of rating criteria, which has to be supported by all implementations. This cost metric conveys a generic measure for the cost of routing traffic from a source to a destination. A lower value indicates a higher preference for traffic to be sent from a source to a destination. How that metric is calculated is up to the ALTO server. It is possible to calculate the "routingcost" cost metric based on actual routing protocol information. Typically, IGPs provide details about endpoints and links within a given network, while the BGP is used to provide details about links to endpoints in other networks. Besides topology and routing information, networks have a multitude of other attributes about their state, condition, and operation that comprises but is not limited to attributes like link utilization, bandwidth and delay, ingress/egress points of data flows from/towards endpoints outside of the network up to the location of nodes and endpoints. In order to enable use of extended information, there is a protocol extension procedure to add new ALTO cost types. The following list gives an overview on further rating criteria that have been proposed or that are in use by ALTO-related prototype implementations. This list is not intended as normative text. Instead, its only purpose is to document and discuss rating criteria that have been proposed so far. Whether such rating criteria are useful and whether the
corresponding information would actually be made available by ISPs can also depend on the use case of ALTO. A list of rating criteria for which normative specifications exist and which have successfully passed the IETF review process can be found at IANA's "ALTO Cost Metric Registry", available from [ALTO-REG]. Distance-related rating criteria: o Relative topological distance: The term relative means that a larger numerical value means greater distance, but it is up to the ALTO service how to compute the values, and the ALTO client will not be informed about the nature of the computation. One way to determine relative topological distance may be counting AS hops, but when querying this parameter, the ALTO client must not assume that the numbers actually are AS hops. In addition to the AS path, a relative cost value could also be calculated taking into account other routing protocol parameters, such as BGP local preference or Multi-Exit Discriminator (MED) attributes. o Absolute topological distance, expressed in the number of traversed autonomous systems. o Absolute topological distance, expressed in the number of router hops (i.e., how much the TTL value of an IP packet will be decreased during transit). o Absolute physical distance, based on knowledge of the approximate geolocation (e.g., continent, country) of an IP address. Performance-related rating criteria: o The minimum achievable throughput between the resource consumer and the candidate resource provider, which is considered useful by the application (only in ALTO queries). o An arbitrary upper bound for the throughput from/to the candidate resource provider (only in ALTO responses). This may be, but is not necessarily, the provisioned access bandwidth of the candidate resource provider. o The maximum Round-Trip Time (RTT) between resource consumer and the candidate resource provider, which is acceptable for the application for useful communication with the candidate resource provider (only in ALTO queries).
o An arbitrary lower bound for the RTT between resource consumer and
the candidate resource provider (only in ALTO responses). This
may be, for example, based on measurements of the propagation
delay in a completely unloaded network.
Charging-related rating criteria:
o Metrics representing an abstract cost, e.g., determined by
policies that distinguish "cheap" from "expensive" IP subnet
ranges without detailing the cost function. According to
[RFC7285], the abstract metric "routingcost" is an example for a
metric for which the cost function does not have to be disclosed.
o Traffic volume caps, in case the Internet access of the resource
consumer is not charged with a "flat rate". For each candidate
resource location, the ALTO service could indicate the amount of
data or the bitrate that may be transferred from/to this resource
location until a given point in time, and how much of this amount
has already been consumed. Furthermore, an ALTO server may have
to indicate how excess traffic would be handled (e.g., blocked,
throttled, or charged separately at an indicated price), e.g., by
a new endpoint property. This is outside the scope of this
document. Also, it is left for further study how several
applications would interact if only some of them use this
criterion. Also left for further study is the use of such a
criterion in resource directories that issue ALTO queries on
behalf of other endpoints.
All the above-listed rating criteria are subject to the remarks
below:
The ALTO client must be aware that with high probability the actual
performance values will differ from whatever an ALTO server exposes.
In particular, an ALTO client must not consider a throughput
parameter as a permission to send data at the indicated rate without
using congestion control mechanisms.
The discrepancies are due to various reasons, including, but not
limited to the following facts:
o The ALTO service is not an admission control system.
o The ALTO service may not know the instantaneous congestion status
of the network.
o The ALTO service may not know all link bandwidths, i.e., where the
bottleneck really is, and there may be shared bottlenecks.
o The ALTO service may not have all information about the actual
routing.
o The ALTO service may not know whether the candidate endpoint
itself is overloaded.
o The ALTO service may not know whether the candidate endpoint
throttles the bandwidth it devotes for the considered application.
o The ALTO service may not know whether the candidate endpoint will
throttle the data it sends to the client (e.g., because of some
fairness algorithm, such as tit for tat).
Because of these inaccuracies and the lack of complete, instantaneous
state information, which are inherent to the ALTO service, the
application must use other mechanisms (such as passive measurements
on actual data transmissions) to assess the currently achievable
throughput, and it must use appropriate congestion control mechanisms
in order to avoid a congestion collapse. Nevertheless, the rating
criteria may provide a useful shortcut for quickly excluding
candidate resource providers from such probing, if it is known in
advance that connectivity is in any case worse than what is
considered the minimum useful value by the respective application.
Rating criteria that should not be defined for and used by the ALTO
service include:
o Performance metrics that are closely related to the instantaneous
congestion status. The definition of alternate approaches for
congestion control is explicitly out of the scope of ALTO.
Instead, other appropriate means, such as using TCP-based
transport, have to be used to avoid congestion. In other words,
ALTO is a service to provide network and policy information, with
update intervals that are possibly several orders of magnitude
slower than congestion-control loops (e.g., in TCP) can react on
changes in network congestion state. This clear separation of
responsibilities avoids traffic oscillations and can help for
network stability and cost optimization.
o Performance metrics that raise privacy concerns. For instance, it
has been questioned whether an ALTO service should publicly expose
the provisioned access bandwidth of cable/DSL customers, as this
could enable identification of "premium customers" of an ISP.
3.3. ALTO Focus and Scope
The purpose of this section is ensure that administrators and users of ALTO services are aware of the objectives of the ALTO protocol design. Using ALTO beyond this scope may limit its efficiency. Likewise, Map-based and Endpoint-based ALTO Services may face certain issues during deployment. This section explains these limitations and also outlines potential solutions.3.3.1. Limitations of Using ALTO beyond Design Assumptions
ALTO is designed as a protocol between clients integrated in applications and servers that provide network information and guidance (e.g., basic network location structure and preferences of network paths). The objective is to modify network resource consumption patterns at application level while maintaining or improving application performance. This design focus results in a number of characteristics of ALTO: o Endpoint focus: In typical ALTO use cases, neither the consumer of the topology information (i.e., the ALTO client) nor the considered resources (e.g., files at endpoints) are part of the network. The ALTO server presents an abstract network topology containing only information relevant to an application overlay for better-than-random resource provider selection among its endpoints. The ALTO protocol specification [RFC7285] is not designed to expose network internals such as routing tables or configuration data that are not relevant for application-level resource provider selection decisions in network endpoints. o Abstraction: The ALTO services such as the Network and Cost Map Service or the ECS provide an abstract view of the network only. The operator of the ALTO server has full control over the granularity (e.g., by defining policies how to aggregate subnets into PIDs) and the level of detail of the abstract network representation (e.g., by deciding what cost types to support). o Multiple administrative domains: The ALTO protocol is designed for use cases where the ALTO server and client can be located in different organizations or trust domains. ALTO assumes a loose coupling between server and client. In addition, ALTO does not assume that an ALTO client has any a priori knowledge about the ALTO server and its supported features. An ALTO server can be discovered automatically. o Read-only: ALTO is a query/response protocol to retrieve guidance information. Neither network/cost map queries nor queries to the ECS are designed to affect state in the network.
If ALTO shall be deployed for use cases beyond the scope defined by these assumptions, the protocol design may result in limitations. For instance, in an Application-Based Network Operations (ABNO) environment, the application could issue an explicit service request to the network [RFC7491]. In this case, the application would require detailed knowledge about the internal network topology and the actual state. A network configuration would also require a corresponding security solution for authentication and authorization. ALTO is not designed for operations to control, operate, and manage a network. Such deployments could be addressed by network management solutions, e.g., based on SNMP [RFC3411] or NETCONF [RFC6241] and YANG [RFC6020], that are typically designed to manipulate configuration state. [RFC7491] contains a more detailed discussion of interfaces between components such as Element Management System (EMS), Network Management System (NMS), Operational Support System (OSS), Traffic Engineering Database (TED), Label Switched Path Database (LSP-DB), Path Computation Element (PCE), and other Operations, Administration, and Maintenance (OAM) components.3.3.2. Limitations of Map-Based Services and Potential Solutions
The specification of the Map Service in the ALTO protocol [RFC7285] is based on the concept of network maps. A network map partitions the network into PIDs that group one or more endpoints (e.g., subnetworks) to a single aggregate. The "costs" between the various PIDs are stored in a cost map. Map-based approaches such as the ALTO Network and Cost Map Service lower the signaling load on the server as maps have to be retrieved only if they change. One main assumption for map-based approaches is that the information provided in these maps is static for a long period of time. This assumption is fine as long as the network operator does not change any parameter, e.g., routing within the network and to the upstream peers, and IP address assignment stays stable (and thus the mapping to the partitions). However, there are several cases where this assumption is not valid: 1. ISPs reallocate IP subnets from time to time. 2. ISPs reallocate IP subnets on short notice. 3. IP prefix blocks may be assigned to a router that serves a variety of access networks.
4. Network costs between IP prefixes may change depending on the
ISP's routing and traffic engineering.
These effects can be explained as follows:
Case 1: ISPs may reallocate IP subnets within their infrastructure
from time to time, partly to ensure the efficient usage of IPv4
addresses (a scarce resource), and partly to enable efficient route
tables within their network routers. The frequency of these
"renumbering events" depends on the growth in number of subscribers
and the availability of address space within the ISP. As a result, a
subscriber's household device could retain an IP address for as short
as a few minutes or for months at a time or even longer.
It has been suggested that ISPs providing ALTO services could
subdivide their subscribers' devices into different IP subnets (or
certain IP address ranges) based on the purchased service tier, as
well as based on the location in the network topology. The problem
is that this sub-allocation of IP subnets tends to decrease the
efficiency of IP address allocation, in particular for IPv4. A
growing ISP that needs to maintain high efficiency of IP address
utilization may be reluctant to jeopardize their future acquisition
of IP address space.
However, this is not an issue for map-based approaches if changes are
applied in the order of days.
Case 2: ISPs can use techniques that allow the reallocation of IP
prefixes on very short notice, i.e., within minutes. An IP prefix
that has no IP address assignment to a host anymore can be
reallocated to areas where there is currently a high demand for IP
addresses.
Case 3: In residential access networks (e.g., DSL, cable), IP
prefixes are assigned to broadband gateways, which are the first IP-
hop in the access-network between the Customer Premises Equipment
(CPE) and the Internet. The access-network between CPE and broadband
gateway (called aggregation network) can have varying characteristics
(and thus associated costs), but still using the same IP prefix. For
instance, one IP address IP1 out of a given CIDR prefix can be
assigned to a VDSL access line (e.g., 2 Mbit/s uplink) while another
IP address IP2 within the same given CIDR prefix is assigned to a
slow ADSL line (e.g., 128 kbit/s uplink). These IP addresses may be
assigned on a first come first served basis, i.e., a single IP
address out of the same CIDR prefix can change its associated costs
quite fast. This may not be an issue with respect to the used
upstream provider (thus the cross ISP traffic), but, depending on the
capacity of the aggregation network, this may raise to an issue.
Case 4: The routing and traffic engineering inside an ISP network, as well as the peering with other autonomous systems, can change dynamically and affect the information exposed by an ALTO server. As a result, cost maps and possibly also network maps can change. One solution to deal with map changes is to use incremental ALTO updates [UPDATE-SSE].3.3.3. Limitations of Non-Map-Based Services and Potential Solutions
The specification of the ALTO protocol [RFC7285] also includes the ECS mechanism. ALTO clients can ask the ALTO server for guidance for specific IP addresses, thereby avoiding the need of processing maps. This can mitigate some of the problems mentioned in the previous section. However, frequent requests, particularly with long lists of IP addresses, may overload the ALTO server. The server has to rank each received IP address, which causes load at the server. This may be amplified when a large number of ALTO clients are asking for guidance. The results of the ECS are also more difficult to cache than ALTO maps. Therefore, the ALTO client may have to await the server response before starting a communication, which results in an additional delay. Caching of IP addresses at the ALTO client or the use of the H12 approach [ALTO-H12] in conjunction with caching may lower the query load on the ALTO server. When an ALTO server receives an ECS request, it may not have the most appropriate topology information in order to accurately determine the ranking. [RFC7285] generally assumes that a server can always offer some guidance. In such a case, the ALTO server could adopt one of the following strategies: o Reply with available information (best effort). o Query another ALTO server presumed to have better topology information and return that response (cascaded servers). o Redirect the request to another ALTO server presumed to have better topology information (redirection). The protocol mechanisms and decision processes that would be used to determine if redirection is necessary and which mode to use is out of the scope of this document, since protocol extensions could be required.
3.4. Monitoring ALTO
3.4.1. Impact and Observation on Network Operation
ALTO presents a new opportunity for managing network traffic by providing additional information to clients. In particular, the deployment of an ALTO server may shift network traffic patterns, and the potential impact to network operation can be large. An ISP providing ALTO may want to assess the benefits of ALTO as part of the management and operations (cf. [RFC7285]). For instance, the ISP might be interested in understanding whether the provided ALTO maps are effective in order to decide whether an adjustment of the ALTO configuration would be useful. Such insight can be obtained from a monitoring infrastructure. An ISP offering ALTO could consider the impact on (or integration with) traffic engineering and the deployment of a monitoring service to observe the effects of ALTO operations. The measurement of impacts can be challenging because ALTO-enabled applications may not provide related information back to the ALTO service provider. To construct an effective monitoring infrastructure, the ALTO service provider should decide how to monitor the performance of ALTO and identify and deploy data sources to collect data to compute the performance metrics. In certain trusted deployment environments, it may be possible to collect information directly from ALTO clients. It may also be possible to vary or selectively disable ALTO guidance for a portion of ALTO clients either by time, geographical region, or some other criteria to compare the network traffic characteristics with and without ALTO. Monitoring an ALTO service could also be realized by third parties. In this case, insight into ALTO data may require a trust relationship between the monitoring system operator and the network service provider offering an ALTO service. The required monitoring depends on the network infrastructure and the use of ALTO, and an exhaustive description is outside the scope of this document.3.4.2. Measurement of the Impact
ALTO realizes an interface between the network and applications. This implies that an effective monitoring infrastructure may have to deal with both network and application performance metrics. This document does not comprehensively list all performance metrics that could be relevant, nor does it formally specify metrics.
The impact of ALTO can be classified regarding a number of different
criteria:
o Total amount and distribution of traffic: ALTO enables ISPs to
influence and localize traffic of applications that use the ALTO
service. Therefore, an ISP may be interested in analyzing the
impact on the traffic, i.e., whether network traffic patterns are
shifted. For instance, if ALTO shall be used to reduce the inter-
domain P2P traffic, it makes sense to evaluate the total amount of
inter-domain traffic of an ISP. Then, one possibility is to study
how the introduction of ALTO reduces the total inter-domain
traffic (inbound and/our outbound). If the ISP's intention is to
localize the traffic inside his network, the network-internal
traffic distribution will be of interest. Effectiveness of
localization can be quantified in different ways, e.g., by the
load on core routers and backbone links or by considering more-
advanced effects, such as the average number of hops that traffic
traverses inside a domain.
o Application performance: The objective of ALTO is to improve
application performance. ALTO can be used by very different types
of applications, with different communication characteristics and
requirements. For instance, if ALTO guidance achieves traffic
localization, one would expect that applications achieve a higher
throughput and/or smaller delays to retrieve data. If
application-specific performance characteristics (e.g., video or
audio quality) can be monitored, such metrics related to user
experience could also help to analyze the benefit of an ALTO
deployment. If available, selected statistics from the TCP/IP
stack in hosts could be leveraged, too.
Of potential interest can also be the share of applications or
customers that actually use an offered ALTO service, i.e., the
adoption of the service.
Monitoring statistics can be aggregated, averaged, and normalized in
different ways. This document does not mandate specific ways how to
calculate metrics.
3.4.3. System and Service Performance
A number of interesting parameters can be measured at the ALTO
server. [RFC7285] suggests certain ALTO-specific metrics to be
monitored:
o Requests and responses for each service listed in an Information
Directory (total counts and size in bytes).
o CPU and memory utilization o ALTO map updates o Number of PIDs o ALTO map sizes (in-memory size, encoded size, number of entries) This data characterizes the workload, the system performance as well as the map data. Obviously, such data will depend on the implementation and the actual deployment of the ALTO service. Logging is also recommended in [RFC7285].3.4.4. Monitoring Infrastructures
Understanding the impact of ALTO may require interaction between different systems operating at different layers. Some information discussed in the preceding sections is only visible to an ISP, while application-level performance can hardly be measured inside the network. It is possible that not all information of potential interest can directly be measured, either because no corresponding monitoring infrastructure or measurement method exists or because it is not easily accessible. One way to quantify the benefit of deploying ALTO is to measure before and after enabling the ALTO service. In addition to passive monitoring, some data could also be obtained by active measurements, but due to the resulting overhead, the latter should be used with care. Yet, in all monitoring activities, an ALTO service provider has to take into account that ALTO clients are not bound to ALTO server guidance as ALTO is only one source of information, and any measurement result may thus be biased. Potential sources for monitoring the use of ALTO include: o Network monitoring and performance management systems: Many ISPs deploy systems to monitor the network traffic, which may have insight into traffic volumes, network topology, bandwidth information inside the management area. Data can be obtained by SNMP, NETCONF, IP Flow Information Export (IPFIX), syslog, etc. On-demand OAM tests (such as Ping or BDF) could also be used. o Applications/clients: Relevant data could be obtained by instrumentation of applications. o ALTO server: If available, log files or other statistics data could be analyzed.
o Other application entities: In several use cases, there are other
application entities that could provide data as well. For
instance, there may be centralized log servers that collect data.
In many ALTO use cases, some data sources are located within an ISP
network while some other data is gathered at the application level.
Correlation of data could require a collaboration agreement between
the ISP and an application owner, including agreements of data
interchange formats, methods of delivery, etc. In practice, such a
collaboration may not be possible in all use cases of ALTO, because
the monitoring data can be sensitive and because the interacting
entities may have different priorities. Details of how to build an
overarching monitoring system for evaluating the benefits of ALTO are
outside the scope of this memo.
(page 36 continued on part 3)