A framework for large-scale measurement
platforms (LMAP)British TelecomAdastral Park, Martlesham HeathIpswichENGLANDphilip.eardley@bt.comAT&T Labs200 Laurel Avenue SouthMiddletown, NJUSAacmorton@att.comUniversidad Carlos III de
MadridAv. Universidad 30LeganesMadrid28911SPAIN34 91 6249500marcelo@it.uc3m.eshttp://www.it.uc3m.esBritish TelecomAdastral Park, Martlesham HeathIpswichENGLANDtrevor.burbridge@bt.comCisco Systems, Inc.96 Commercial StreetEdinburghScotlandEH6 6LXUKpaitken@cisco.comCisco Systems, Inc.7025 Kit Creek RoadRTPNC27709USAaakhter@cisco.comMeasuring broadband service on a large scale requires standardisation
of the logical architecture and a description of the key protocols that
coordinate interactions between the components. The document presents an
overall framework for large-scale measurements. It also defines
terminology for LMAP (large-scale measurement platforms). The document
is a contribution towards the LMAP working group's milestone.There is a desire to be able to coordinate the execution of broadband
measurements and the collection of measurement results across a large
scale set of diverse devices. These devices could be software based
agents on PCs, embedded agents in consumer devices (e.g. blu-ray
players), service provider controlled devices such as set-top players
and home gateways, or simply dedicated probes. It is expected that such
a system could easily comprise 100k devices. Such a scale presents
unique problems in coordination, execution and measurement result
collection. Several use cases have been proposed for large- scale
measurements including:Operators: to help plan their network and identify faultsRegulators: to benchmark several network operators and support
public policy developmentFurther details of the use cases can be found at . The LMAP framework should be
useful for these, as well as other use cases that the LMAP WG doesn't
concentrate on, such as to help end users run diagnostic checks like a
network speed test.The LMAP framework has four basic elements: Measurement Agents,
Measurement Peers, Controllers and Collectors.Measurement Agents (MAs) perform network measurements. They are
pieces of code that can be executed in specialized hardware (hardware
probe) or on a general-purpose device (like a PC or mobile phone). The
Measurement Agents may have multiple interfaces (WiFi, Ethernet, DSL,
fibre, etc.) and the measurements may specify any one of these.
Measurements may be active (the MA or Measurement Peer (MP) generates
test traffic), passive (the MA observes user traffic), or some hybrid
form of the two. For active measurement tasks, the MA (or MP) generates
test traffic and measures some metric associated with its transfer over
the path to (or from) a Measurement Peer. For example, one active
measurement task could be to measure the UDP latency between the MA and
a given MP. MAs may also conduct passive testing through the observation
of traffic. The measurements themselves may be on IPv4, IPv6, and on
various services (DNS, HTTP, XMPP, FTP, VoIP, etc.).The Controller manages one or more MAs by instructing it which
measurement tasks it should perform and when. For example it may
instruct a MA at a home gateway: “Measure the ‘UDP
latency’ with the Measurement Peer mp.example.org; repeat every
hour at xx.05”. The Controller also manages a MA by instructing it
how to report the measurement results, for example: “Report
results once a day in a batch at 4am”. We refer to these as the
Measurement Schedule and Report Schedule.The Collector accepts Reports from the MAs with the results from
their measurement tasks. Therefore the MA is a device that initiates the
measurement tasks, gets instructions from the Controller and reports to
the Collector.There are additional elements that are part of a measurement system,
but that are out of the scope for LMAP. We provide a detailed discussion
of all the elements in the rest of the document.Over the years various efforts inside and outside the IETF have
worked on independent components of such a system. There are also
existing systems that are deployed today. However, these are either
proprietary, closed, and/or not standardized. The IETF Large-Scale
Measurement of Broadband Performance (LMAP) Working Group is chartered
to specify the information model, associated data models, and
select/extend one or more protocols for secure measurement control and
measurement result collection.The goal is to have the measurements (made using the same metrics and
mechanisms) for a large number of points on the Internet, and to have
the results collected and stored in the same form.The desirable features for a large-scale measurement systems we are
designing for are:Standardised - in terms of the tests that they perform, the
components, the data models and protocols for transferring
information between the components. For example so that it is
meaningful to compare measurements made of the same metric at
different times and places. For example so that the operator of a
measurement system can buy the various components from different
vendors. Today's systems are proprietary in some or all of these
aspects.Large-scale -
envisages Measurement Agents in every home gateway and edge device
such as set-top-boxes and tablet computers. Existing systems have up
to a few thousand Measurement Agents (without judging how much
further they could scale).Diversity - a measurement system should handle different types of
Measurement Agent - for example Measurement Agents may come from
different vendors, be in wired and wireless networks and be on
devices with IPv4 or IPv6 addresses.This section defines terminology for LMAP. Please note that defined
terms are capitalized.Active Measurement Method (Task): A type of Measurement Method (Task)
that involves a Measurement Agent and a Measurement Peer (or possibly
Peers), where either the Measurement Agent or the Measurement Peer
injects test packet(s) into the network destined for the other, and
which involves one of them measuring some performance or reliability
parameter associated with the transfer of the packet(s).Bootstrap Protocol: A protocol that initialises a Measurement Agent
with the information necessary to be integrated into a measurement
system.Collector: A function that receives a Report from a Measurement
Agent. Colloquially, a Collector is a physical device that performs this
function.Controller: A function that provides a Measurement Agent with
Instruction(s). Colloquially, a Controller is a physical device that
performs this function.Control Protocol: The protocol delivering Instruction(s) from a
Controller to a Measurement Agent.Cycle-ID: A tag that is sent by the Controller in an Instruction and
echoed by the MA in its Report; Measurement Results with the same
Cycle-ID are expected to be comparable.Data Model: The implementation of an Information Model in a
particular data modelling language.Derived Metric: A Metric that is a combination of other Metrics,
and/or a combination of the same Metric measured over different parts of
the network, or at different times.Environmental Constraint: A parameter that is measured as part of the
Measurement Task, its value determining whether the rest of the
Measurement Task proceeds.Group-ID: An identifier of a group of MAs.Information Model: The protocol-neutral definition of the semantics
of the Instructions, the Report, the status of the different elements of
the measurement system as well of the events in the system.Instruction: The description of Measurement Tasks to perform and the
details of the Report to send. The Instruction is sent by a Controller
to a Measurement Agent.Measurement Agent (MA): The function that receives Instructions from
a Controller, performs Measurement Tasks (perhaps in concert with a
Measurement Peer) and reports Measurement Results to a Collector.
Colloquially, a Measurement Agent is a physical device that performs
this function.Measurement Method: The process for assessing the value of a Metric;
the process of measuring some performance or reliability parameter; the
generalisation of a Measurement Task.Measurement Parameter: A parameter whose value is left open by the
Measurement Method.Measurement Peer: The function that receives control messages and
test packets from a Measurement Agent and may reply to the Measurement
Agent as defined by the Measurement Method.Measurement Result: The output of a single Measurement Task (the
value obtained for the parameter of interest, or Metric).Measurement Schedule: the schedule for performing a series of
Measurement Tasks.Measurement Suppression: a type of Instruction that stops
(suppresses) Measurement Tasks.Measurement Task: The act that yields a single Measurement Result;
the act consisting of the (single) operation of the Measurement Method
at a particular time and with all its parameters set to specific
values.Metric: The quantity related to the performance and reliability of
the Internet that we'd like to know the value of, and that is carefully
specified.Passive Measurement Method (Task): A Measurement Method (Task) in
which a Measurement Agent observes existing traffic at a specific
measurement point, but does not inject test packet(s).Report: The Measurement Results and other associated information (as
defined by the Instruction); a specific instance of the Data Model. The
Report is sent by a Measurement Agent to a Collector.Report Channel: a specific Report Schedule and CollectorReport Protocol: The protocol delivering Report(s) from a Measurement
Agent to a Collector.Report Schedule: the schedule for sending a series of Reports to a
Collector.Subscriber: An entity (associated with one or more users) that is
engaged in a subscription with a service provider. The subscriber is
allowed to subscribe and un-subscribe services, to register a user or a
list of users authorized to enjoy these services, and also to set the
limits relative to the use that associated users make of these services.
(This definition is from .) Test Traffic: for Active Measurement Tasks, the traffic generated by
the Measurement Agent and/or the Measurement Peer to execute the
requested Measurement Task.Figure 1 shows the main components of a measurement system, and the
interactions of those components. Some of the components are outside the
scope of LMAP. In this section we provide an overview on the whole
measurement system, whilst the subsequent sections study the LMAP
components in more detail.The first component is a Measurement Task, which measures some
performance or reliability Metric of interest. An Active Measurement
Task involves either a Measurement Agent injecting Test Traffic into the
network destined for a Measurement Peer, and/or a MP sending Test
Traffic to a MA; one of them measures the some parameter associated with
the transfer of the packet(s). A Passive Measurement Task involves only
a MA, which simply observes existing traffic - for example, it could
simply count bytes or it might calculate the average loss for a
particular flow.It is very useful to standardise Measurement Methods (a Measurement
Method is a generalisation of a Measurement Task), so that it is
meaningful to compare measurements of the same Metric made at different
times and places. It is also useful to define a registry for
commonly-used Metrics so that a Measurement
Method can be referred to simply by its identifier in the registry. The
Measurement Methods and registry would hopefully also be referenced by
other standards organisations.In order for a Measurement Agent and a Measurement Peer to execute an
Active Measurement Task, they exchange Test Traffic. The protocols used
for the Test Traffic is out of the scope of the LMAP WG and falls within
the scope of the IETF WGs such as IPPM.For Measurement Results to be truly comparable, as might be required
by a regulator, not only do the same Measurement Methods need to be used
but also the set of Measurement Tasks should follow a similar
Measurement Schedule and be of similar number. The details of such a
characterisation plan are beyond the scope of work in IETF although
certainly facilitated by IETF's work.The next components we consider are the Measurement Agent (MA),
Controller and Collector. The main work of the LMAP working group is to
define the Control Protocol between the Controller and MA, and the
Report Protocol between the MA and Collector. Section 4 onwards
considers the LMAP compnents in more detail; here we introduce them.The Controller manages a MA by instructing it which tests it should
perform and when. For example it may instruct a MA at a home gateway:
“Run the ‘download speed test’ with the test server at
the end user's first IP point in the network; if the end user is active
then delay the test and re-try 1 minute later, with up to 3 re-tries;
repeat every hour at xx.05 + Unif[0,180] seconds”. The Controller
also manages a MA by instructing it how to report the test results, for
example: “Report results once a day in a batch at 4am +
Unif[0,180] seconds; if the end user is active then delay the report 5
minutes”. As well as regular tests, a Controller can initiate a
one-off test ("Do test now", "Report as soon as possible"). These are
called the Measurement and Report Schedule.The Collector accepts a Report from a MA with the results from its
tests. It may also do some processing on the results – for
instance to eliminate outliers, as they can severely impact the
aggregated results.Finally we introduce several components that are out of scope of the
LMAP WG and will be provided through existing protocols or applications.
They affect how the measurement system uses the Measurement Results and
how it decides what set of Measurement Tasks to perform.The MA needs to be bootstrapped with initial details about its
Controller, including authentication credentials. The LMAP WG considers
the boostrap process, since it affects the Information Model. However,
it does not define a bootstrap protocol, since it is likely to be
technology specific and could be defined by the Broadband Forum, DOCSIS
or IEEE. depending on the device. Possible protocols are SNMP, NETCONF
or (for Home Gateways) CPE WAN Management Protocol (CWMP) from the Auto
Configuration Server (ACS) (as specified in TR-069).A Subscriber Parameter Database contains information about the line,
for example the customer's broadband contract (perhaps 2, 40 or 80Mb/s),
the line technology (DSL or fibre), the time zone where the MA is
located, and the type of home gateway and MA. These are all factors
which may affect the choice of what Measurement Tasks to run and how to
interpret the Measurement Results. For example, a download test suitable
for a line with an 80Mb/s contract may overwhelm a 2Mb/s line. Another
example is if the Controller wants to run a one-off test to diagnose a
fault, then it should understand what problem the customer is
experiencing and what tests have already been run. The Subscribers'
service parameters are already gathered and stored by existing
operations systems.A Results Database records all measurements in an equivalent form,
for example an SQL database, so that they can be easily accessed by the
Data Analysis Tools. The Data Analysis Tools also need to understand the
Subscriber's service information, for example the broadband
contract.The Data Analysis Tools receive the results from the Collector or via
the Results Database. They might visualise the data or identify which
component or link is likely to be the cause of a fault or
degradation.The operator's OAM (Operations, Administration, and Maintenance) uses
the results from the tools.The LMAP framework makes some important assumptions, which constrain
the scope of the work to be done.In the LMAP framework (as defined in the WG's charter) the
measurement system is under the direction of a single organisation
that is responsible both for the data and the quality of experience
delivered to its users. Clear responsibility is critical given that a
misbehaving large-scale measurement system could potentially harm user
experience, user privacy and network security.However, the components of an LMAP measurement system can be
deployed in administrative domains that are not owned by the measuring
organisation. Thus, the system of functions deployed by a single
organisation constitutes a single LMAP domain which may span ownership
or other administrative boundaries.A MA is instructed by one Controller and is in one measurement
system. The constraint avoids different Controllers giving a MA
conflicting instructions and so means that the MA does not have to
manage contention between multiple Measurement (or Report) Schedules.
This simplifies the design of MAs (critical for a large-scale
infrastructure) and allows a Measurement Schedule to be tested on
specific types of MA before deployment to ensure that the end user
experience is not impacted (due to CPU, memory or broadband-product
constraints).An operator may have several Controllers, perhaps with a Controller
for different types of MA (home gateways, tablets) or location
(Ipswich, Edinburgh).A protocol model presents (RFC4101) "an architectural model for how
the protocol operates ... a short description of the system in overview
form, ... [which] needs to answer three basic questions:What problem is the protocol trying to achieve?What messages are being transmitted and what do they mean?What are the important, but unobvious, features of the
protocol?"An LMAP system goes through the following phases:a bootstrapping process before the MA can take part in the three
items belowa Control Protocol, which delivers an Instruction from a
Controller and a MA. The Instruction details what Measurement Tasks
the MA should perform and when, and how it should report the
Measurement Resultsthe actual Measurement Tasks are performed. An Active Measurement
Task involves sending test traffic between the Measurement Agent and
a Measurement Peer, whilst a Passive Measurement Task involves
(only) the Measurement Agent observing existing user traffic. The
LMAP WG does not define Measurement Methods, however the IPPM WG
does. a Report Protocol, which delivers a Report from the MA to a
Collector. The Report contains the Measurement Results.In the diagrams the following convention is used:(optional): indicated by round brackets[potentially repeated]: indicated by square bracketsThe Protocol Model is closely related to the Information Model, which
is the abstract definition of the information carried by the protocol
model. The purpose of both is to provide a protocol and device
independent view, which can be implemented via specific protocols. The
LMAP WG will define a specific Control Protocol and Report Protocol, but
other Protocols could be defined by other standards bodies or be
proprietary. However it is important that they all implement the same
Information and Protocol Model, in order to ease the definition,
operation and interoperability of large-scale measurement systems.The primary purpose of bootstrapping is to enable the MA and
Controller to be integrated into a measurement system. In order to do
that, the MA needs to retrieve information about itself (like its
identity in the measurement system), about the Controller and the
Collector(s) as well as security information (such as certificates and
credentials).Typically the MA is behind a NAT, so needs to initiate
communications, in order that the Controller can communicate with it.
The normal NAT interactions are not shown in the figure. The MA knows how to contact a Controller through some device
/access specific mechanism. For example, this could be in the
firmware, downloaded, manually configured or via a protocol like
TR-069. The Controller could either be the one that will send it
Instructions (see next sub-section) or else an initial Controller. The
role of an initial Controller is simply to inform the MA how to
contact its actual Controller; this could be useful, for example, for
load balancing or if the details of the initial Controller are
statically configured or if the measurement system has specific
Controllers for different devices types. When the MA registers with
the Controller it learns its MA identifier; it may also be told a
Group-ID and whether to include the MA-ID as well as the Group-ID in
its Reports. A Group-ID would be shared by several MAs and could be
useful for privacy reasons (for instance to hinder tracking of a
mobile MA device). The MA may also tell the Controller the list of
Measurement Methods that its capable of (see next sub-section).Whilst the LMAP WG considers the bootstrapping process, it is out
of scope to define a bootstrap mechanism, as it depends on the type of
device and access.Open issue: what happens if a Controller fails, how is the MA is
homed onto a new one?The primary purpose of the Control Protocol is to allow the
Controller to configure a Measurement Agent with Measurement
Instructions, which it then acts on autonomously.The Instruction contains:what measurements to do: the Measurement Methods could be
defined by reference to a registry entry, along with any
parameters that need to be set (such as the address of the
Measurement Peer) and any Environmental Constraint (such as,
'delay the test if the end user is active')when to do them: the Measurement Schedule details the timings
of regular tests, one-off testshow to report the Measurement Results: via Reporting
Channel(s), each of which defines a target Collector and Report
ScheduleAn Instruction could contain one or more of the above elements,
since the Controller may want the MA to perform several different
Measurement Tasks (measure UDP latency and download speed), at several
frequencies (a regular test every hour and a one-off test
immediately), and report to several Collectors. The different elements
can be updated independently at different times and regularities, for
example it is likely that the Measurement Schedule will be updated
more often than the other elements.In general we expect that the Controller knows what Measurement
Methods the MA supports, such that the Controller can correctly
instruct the MA. Note that the Control Protocol does not allow
negotiation (which would add complexity to the MA, Controller and
Control Protocol for little benefit).The MA can send to the Controller the complete list of Measurement
Methods that it is capable of. Note that it is not intended to
indicate dynamic capabilities like the MA's currently unused CPU,
memory or battery life. The list of Measurement Methods could be
useful in several circumstances: when the MA first communicates with a
Controller; when the MA becomes capable of a new Measurement Method;
when requested by the Controller (for example, if the Controller
forgets what the MA can do or otherwise wants to resynchronize what it
knows about the MA).The Controller has the ability to send a “suppress”
message to MAs. This could be useful if there is some unexpected
network issue and so the measurement system wants to eliminate
inessential traffic. As a result, temporarily the MA does not start
new Active Measurement Tasks, and it may also stop in-progress
Measurement Tasks, especially ones that are long-running &/or
creates a lot of traffic. See the next section for more information on
stopping Measuremet Tasks.The figure shows that the various messages are acknowledged, which
means that they have been delivered successfully. However, the
"suppress" message is not acknowledged, since it is likely to be
broadcast to several /many MAs at a time when the measurement system
wants to eliminate inessential traffic. Note also that the MA does not
inform the Controller about Measurement Tasks starting and
stopping.There is no need for the MA to confirm to the Controller that it
has understood and acted on the Instruction, since the Controller
knows the capabilities of the MA. However, the Control Protocol must
support robust error reporting by the MA, to provide the Controller
with sufficiently detailed reasons for any failures. There are two
broad categories of failure: the MA cannot action the Instruction (for
example, it doesn't include a parameter that is mandatory for the
requested Measurement Method); or the Measurement Task could not be
executed (for example, the MA unexpectedly has no spare CPU cycles).
Note that it is not considered a failure if a Measurement Task
(correctly) doesn't start - for example if the MA detects
cross-traffic; instead this is reported to the Collector in the normal
manner (see Section below).Comment: the detailed list of reasons below would be more
appropriate in the Information Model i-d.no value for a mandatory parametertime of test is in pasttype wrong, eg string given where expect integerSchedule refers to a Measurement configuration or Report
Channel that doesn't existMA has crashedMA doesn’t (any longer) understand requested MethodMA has run out of CPU, memory, battery powerCollector has disappearedMP has disappearedFinally, note that the MA doesn't do a 'safety check' with the
Controller (that it should still continue with the requested
Measurement Tasks) - it simply carries out the Measurement Tasks as
instructed, unless it gets an updated Instruction.The LMAP WG will define a Control Protocol and its associated Data
Model that implements the Protocol & Information Model. This may
be a simple instruction - response protocol, and LMAP will specify how
it operates over an existing protocol -to be selected, perhaps
REST-style HTTP(s) or NETCONF-YANG.The LMAP WG is neutral to what the actual Measurement Task is. The
WG does not define a generic start and stop process, since the correct
approach depend on the particular Measurement Task; the details are
defined as part of each Measurement Method, and hence potentially by
the IPPM WG.Once the MA gets its Measurement and Report Schedules from its
Controller then it acts autonomously, in terms of operation of the
Measurement Tasks and reporting of the result. One implication is that
the MA initiates Measurement Tasks. Therefore for the common case
where the MA is on a home gateway, the MA initiates a ‘download
speed test’ by asking a Measurement Peer to send the file.Many Active Measurement Tasks begin with a pre-check before the
test traffic is sent. Action could include:the MA checking that there is no cross-traffic (ie that the
user isn’t already sending traffic);the MA checking with the Measurement Peer that it can handle a
new Measurement Task (in case the MP is already handling many
Measurement Tasks with other MAs);the first part of the Measurement Task consisting of traffic
that probes the path to make sure it isn’t overloaded.It is possible that similar checks continue during the
Measurement Task, especially one that is long-running &/or creates
a lot of Test Traffic, which may be abandoned whilst in-progress. A
Measurement Task could also be abandoned in response to a "suppress"
message (see previous section). Action could include:For ‘upload’ tests, the MA not sending trafficFor ‘download’ tests, the MA closing the TCP
connection or sending a TWAMP Stop control message.Comment: presumably Passive Measurement Tasks don't do pre-checking
or stopping?The primary purpose of the Report Protocol is to allow a
Measurement Agent to report its Measurement Results to a Collector,
and the context in which they were obtained.The MA acts autonomously in terms of reporting; it simply sends
Reports as defined by the Controller's Instruction.The Report contains:the MA's identifier, or perhaps a Group-ID to anonymise
resultsthe actual Measurement Results, including the time they were
measuredthe details of the Measurement Task (to avoid the Collector
having to ask the Controller for this information later)Depending on the requirements of the measurement system, the MA
might label, or perhaps not include, Measurement Results impacted by
for instance cross-traffic or the MP being busy. If applicable the
Measurement Report includes the start and end of suppression.The MA may report the results to more than one Collector, if the
Instruction says so. It could report a different subset of Results to
different Collectors. The LMAP WG will define a Report Protocol and its associated Data
Model that implements the Protocol & Information Model. This may
be a simple instruction - response protocol, and LMAP will specify how
it operates over an existing protocol - to be selected, perhaps
REST-style HTTP(s) or IPFIX.There are several potential interactions between LMAP elements that
are out of scope of definition by the LMAP WG:It does not define a coordination process between MAs. Whilst a
measurement system may define coordinated Measurement Schedules
across its various MAs, there is no direct coordination between
MAs.It does not define interactions between the Collector and
Controller. It is quite likely that there will be such
interactions, probably intermediated by the data analysis tools.
For example if there is an "interesting" Measurement Result then
the measurement system may want to trigger extra Measurement Tasks
that explore the potential cause in more detail.It does not define coordination between different measurement
systems. For example, it does not define the interaction of a MA
in one measurement system with a Controller or Collector in a
different measurement system. Whilst it is likely that the Control
and Report protocols could be re-used or adapted for this
scenario, any form of coordination between different organisations
involves difficult commercial and technical issues and so, given
the novelty of large-scale measurement efforts, any form of
inter-organisation coordination is outside the scope of the LMAP
WG. Note that a single MA is instructed by a single Controller and
is only in one measurement system.An interesting scenario is where a home contains two
independent MAs, for example one controlled by a regulator and
one controlled by an ISP. Then the test traffic of one MA is
treated by the other MA just like any other user traffic.It does not specifically define a user-initiated measurement
system, see sub-section.The WG concentrates on the cases where an ISP or a regulator runs
the measurement system. However, we expect that LMAP functionality
will also be used in the context of an end user-controlled
measurement system. There are at least two ways this could happen
(they have various pros and cons):a user could somehow request the ISP- (or regulator-) run
measurement system to test his/her line. The ISP (or regulator)
Controller would then send an Instruction to the MA in the usual
LMAP way. Note that a user can’t directly initiate a
Measurement Task on an ISP- (or regulator-) controlled MA.a user could deploy their own measurement system, with their
own MA, Controller and Collector. For example, the user could
download all three functions onto the same user-owned end
device; then the LMAP Control and Report protocols do not need
to be used, but using LMAP's Information Model would still be
beneficial. The MP could be in the home gateway or outside the
home network; in the latter case the MP is highly likely to be
run by a different organisation, which raises extra privacy
considerations.In both cases there will be some way for the user to initiate the
Measurement Task(s). The mechanism is out-of-scope of the LMAP WG,
but could include the user clicking a button on a GUI or sending a
text message. Presumably the user will also be able to see the
Measurement Results, perhaps summarised on a webpage. It is
suggested that these interfaces conform to the LMAP guidance on the
privacy of the Measurement Results and Subscriber information.This section contains a more detailed discussion of the four
components of the LMAP framework.The Measurement Agent is the component that is responsible for
executing the Measurement Tasks. The Measurement Agent could take a
number of forms: a dedicated probe, software on a PC, embedded into an
appliance, or even embedded into a gateway. A single site (home,
branch office etc.) that is participating in a measurement could make
use of one or multiple Measurement Agents in a single measurement
e.g., if there are multiple output interfaces, there might be a
Measurement Agent per interface. The Measurement Agent's configuration
(specifically which Controller to initially connect to), is out of
scope within LMAP. However, depending on the type of probe, it could
be manually configured by the user, pre-configured before shipment to
the end user, or configured by the application (in the case of some PC
based Measurement Agents). For example, a Measurement Agent that is
included in the app for a content provider might be configured
automatically by the content provider to use the content provider's
LMAP Controller. That said, there should be an element of local
premises configuration that allows the Measurement Agent (especially
in the case of Active Measurements Tasks) to mimic performance of user
applications at the same site. For example, making use of the same DNS
server as the remainder of the site. The Measurement Agent could be
deployed in a variety of locations. Not all deployment locations are
available to every kind of Measurement Agent operator. There are also
a variety of limitations and trade-offs depending on the final
placement. The next sections outline some of the locations a
Measurement Agent may be deployed. This is not an exhaustive list and
combinations of the below may also apply.A Measurement Agent embedded with the site gateway (e.g. in the
case of a a branch office in a managed service environment) is one
of better places the Measurement Agent could be deployed. All site
to ISP traffic would traverse through the gateway and passive
measurements could easily be performed. Similarly, due to this user
traffic visibility, an Active Measurements Task could be rescheduled
so as not to compete with user traffic. Generally NAT and firewall
services are built into the gateway, allowing the Measurement Agent
the option to offer its Controller facing management interface
outside of the NAT/firewall. This placement of the management
interface allows the Controller to unilaterally contact the
Measurement Agent for instructions. However, if the site gateway is
owned and operated by the service provider, the Measurement Agent
will generally not be available for over the top providers, the
regulator, end users or enterprises.The Measurement Agent could also be embedded behind a NAT, a
firewall, or both. In this case the Controller may not be able to
unilaterally contact the Measurement Agent unless either static port
forwarding configuration or firewall pin holing is configured. This
would require user intervention, and ultimately might not be an
option available to the user (perhaps due to permissions). The
Measurement Agent may originate a session towards the Controller and
maintain the session for bidirectional communications. This would
alleviate the need to have user intervention on the gateway, but
would reduce the overall scalability of the Controller as it would
have to maintain a higher number of active sessions. That said,
sending keepalives to prop open the firewall could serve a dual
purpose in testing network reachability for the Measurement Agent.
An alternative would be to use a protocol such as UPnP or PCP
[RFC6887] to control the NAT/firewall if the gateway supports this
kind of control.As mentioned earlier, there are benefits in the Measurement
Agent's ability to observe the site's user traffic. It allows the
Measurement Agent to back off a potentially disruptive Active
Measurements Task to avoid impacting the user. A Passive
Measurements Task allows the Measurement Agent to gather data
without the overhead of Test Traffic (of interest to both the site
user and network operator) as well as potentially provide a greater
number of samples. A Measurement Agent behind the gateway would
generally not be privy to observation of the user traffic unless the
Measurement Agent was placed in-line with the site gateway or the
site gateway traffic was replicated to the Measurement Agent (a
capability generally not found in home broadband gateways).A broadband site may be multi-homed. For example, the site may be
connected to multiple broadband ISPs (perhaps for redundancy or
load- sharing), or have a broadband as well as mobile/WiFi
connectivity. It may also be helpful to think of dual stack IPv4 and
IPv6 broadband sites as multi-homed. In these cases, there needs to
be clarity on which network connectivity option is being measured.
Sometimes this is easily resolved by the location of the MA itself.
For example, if the MA is built into the gateway (and the gateway
only has a single WAN side interface), there is little confusion or
choice. However, for multi-homed gateways or devices behind the
gateway(s) of multi-homed sites it would be preferable to explicitly
select the network to measure (e.g. [RFC5533]) but the network
measured should be included in the Measurement Result. Section 3.2
of [I-D.ietf-homenet-arch] describes dual-stack and multi-homing
topologies that might be encountered in a home network (which is
generally a broadband connected site). The Multiple Interfaces (mif)
working group covers cases where hosts are either directly attached
to multiple networks (physical or virtual) or indirectly (multiple
default routers, etc.). xref target="RFC6419"/> provides the
current practices of multi-interfaces hosts today. As some of the
end goals of a MA is to replicate the end user's network experience,
it is important to understand the current practices.A Measurement Peer is the other side of an Active Measurements Task
- the target of Test Traffic from a Measurement Agent. The Measurement
Peer could also take many different forms: a web site, a service
(VoIP), a DNS server, an application specific server (e.g., webex), a
well known web site (e.g., youtube, google search), even another
Measurement Agent in another home could perform as a Measurement Peer
for a given Measurement Task. Particularly useful could be a MP that
is well placed bandwidth-wise and can handle thousand of sessions of
Test Traffic.A Controller is responsible for providing the Measurement Agent
with instructions which include the Measurement Schedule, parameters,
etc. It is basically the entity controlling the Measurement Agents in
a LMAP domain.For scaling purposes there may be several Controllers, perhaps
regionally located. A large scale test making use of multiple
Controllers would need a master Controller that is the ultimate source
of direction.A Collector is responsible for receiving the Measurement Results
from the Measurement Agent at the end of a Measurement Task. It may
have additional features such as aggregating the results across
multiple Measurement Agents, remove outliers, create additional
statistics, (depending on usage of data) anonymization of results for
privacy reasons (if not done already in the Measurement Agents) etc.
The work of anonymization of user identifiable data has been addressed
for IPFIX via RFC6235 [RFC6235]. For scaling purposes there may be
several Collectors, perhaps regionally located. A large scale test
making use of multiple Collectors would need to aggregate/consolidate
their results for the complete picture.The security of the LMAP framework should protect the interests of
the measurement operator(s), the network user(s) and other actors who
could be impacted by a compromised measurement deployment.We assume that each Measurement Agent will receive test
configuration, scheduling and reporting instructions from a single
organisation (operator of the Controller). These instructions must be
authenticated (to ensure that they come from the trusted Controller),
checked for integrity (to ensure no-one has tampered with them) and be
prevented from replay. If a malicious party can gain control of the
Measurement Agent they can use the MA capabilities to launch DoS attacks
at targets, reduce the network user experience and corrupt the
measurement results that are reported to the Collector. By altering the
tests that are operated and/or the Collector address they can also
compromise the confidentiality of the network user and the MA
environment (such as information about the location of devices or their
traffic).The reporting of the MA must also be secured to maintain
confidentiality. The results must be encrypted such that only the
authorised Collector can decrypt the results to prevent the leakage of
confidential or private information. In addition it must be
authenticated that the results have come from the expected MA and that
they have not been tampered with. It must not be possible to fool a MA
into injecting falsified data into the measurement platform or to
corrupt the results of a real MA.Availability should also be considered. While the loss of some MAs
may not be considered critical, the unavailability of the Collector
could mean that valuable business data or data critical to a regulatory
process is lost. Similarly, the unavailability of a Controller could
mean that the MAs do not operate a correct Measurement Schedule.A malicious party could "game the system". For example, where a
regulator is running a measurement system in order to benchmark
operators, an operator could try to identify the broadband lines that
the regulator was measuring and prioritise that traffic. This potential
issue is currently handled by a code of conduct. It is outside the scope
of the LMAP WG to consider the issue.Comment: It may be better to create a separate draft about 'LMAP
threats and considerations' containing this section and perhaps the
security section.The LMAP Working Group will consider privacy as a core requirement
and will ensure that by default measurement and collection mechanisms
and protocols operate in a privacy-sensitive manner, i.e. that privacy
features are well-defined.This section provides a set of privacy considerations for LMAP. This
section benefits greatly from the timely publication of . There are dependencies on the integrity of the LMAP
security mechanisms, described in the Security Considerations section
above.We begin with a set of assumptions related to protecting the
sensitive information of individuals and organizations participating in
LMAP-orchestrated measurement and data collection.LMAP protocols need to protect the sensitive information of the
following entities, including individuals and organizations who
participate in measurement and collection of results.Individual Internet Users: Persons who utilize Internet access
services for communications tasks, according to the terms of
service of a service agreement. Such persons may be a Service
Subscriber, or have been given permission by the subscriber to use
the service.Internet Service Providers: Organizations who offer Internet
access service subscriptions, and thus have access to sensitive
information of Individuals who choose to use the service. These
organizations desire to protect their subscribers and their own
sensitive information which may be stored in the process of
measurement and result collection.Other LMAP system Operators: Organizations who operate
measurement systems or participate in measurements in some
way.This section gives examples of sensitive information which may be
measured or stored in a measurement system, and which is to be kept
private by default in the LMAP core protocols.Examples of Subscriber or authorized Internet User Sensitive
Information:IP address in usePersonal Identification (Real Name)Location (street address, city)Subscribed Service ParametersContents of Traffic (Activity, DNS queries, Destinations,
Equipment types, Account info for other services, etc.)Status as a study volunteer and Schedule of (Active)
Measurement TasksExamples of Internet Service Provider Sensitive Information:Measurement Device Identification (Equipment ID and IP
address)Measurement Instructions (choice of measurements)Measurement Results (some may be shared, others may be
private)Measurement Schedule (exact times)Network Topology (Locations, Connectivity, Redundancy)Subscriber billing information, and any of the above Subscriber
Information known to the provider.Authentication credentials (e.g., certificates)Other organizations will have some combination of the lists
above.For the purposes of this memo, we define Passive and Active
Measurements Tasks as follows:Passive: measurements conducted on Internet User traffic, such that
sensitive information is present and stored in the measurement system
(however briefly this storage may be).Active: measurements conducted on traffic which serves only the
purpose of measurement. Even if a user host generates active
measurement traffic, there is significantly limited sensitive
information present and stored in the measurement system compared to
the passive case, as follows:IP address in useStatus as a study volunteer and schedule of active testsOn the other hand, the sensitive information for an Internet
Service Provider is the same whether active or passive measurements
are used.This section briefly presents a set of communication models for
LMAP. We assume that the Measurement Agent is located behind a
NAT/Firewall, so it performs the role of Initiator for all
communications.From a privacy perspective, all LMAP entities can be considered
"observers" according to the definition in .
Their stored information potentially poses a threat to privacy,
especially if one or more of these functional entities has been
compromised.Likewise, all devices on the paths used for control, reporting, and
measurement are also observers. We note this in the figures below by
identifying the possible presence of a NAT, which has additional
significance to the protocols and direction of initiation.The high-level communication model for interactions between the
LMAP Controller and Measurement Agent is illustrated below. The
primary purpose of this exchange is to authenticate and task a
Measurement Agent with Measurement Instructions, which the
Measurement Agent then acts on autonomously.Primarily IP addresses and pseudonyms are exchanged
first, then measurement-related information of interest such as the
metrics, schedule, and IP addresses of measurement devices.An organization operating the controller having no service
relationship with the user who hosts the measurement agent *could*
gain real-name mapping to public IP address through user
participation in an LMAP system.The high-level communication model for interactions between the
LMAP Measurement Agent and Collector is illustrated below. The
primary purpose of this exchange is to authenticate and collect
results from a Measurement Agent, which it has measured autonomously
and stored.Primarily IP addresses and pseudonyms are exchanged
first, then measurement-related information of interest such as the
metrics, schedule, results, and IP addresses of measurement
devices.An organization operating the collector having no service
relationship with the user who hosts the measurement agent *could*
gain real-name mapping to public IP address through user
participation in an LMAP system.Although the specification of the mechanisms for measurement is
beyond the LMAP scope, the high-level communications model below
illustrates measurement information and results flowing between
active measurement devices as a potential privacy issue. The primary
purpose of this exchange is to execute measurements and store the
results.This exchange primarily exposes the IP addresses of
measurement devices and the inference of measurement participation
from such traffic. There may be information on key points in a
service provider's network. There may also be access to
measurement-related information of interest such as the metrics,
schedule, and results.If the measurement traffic is unencrypted, as found in many
systems today, then both timing and limited results are open to
observers.Although the specification of the mechanisms for measurement is
beyond the LMAP scope, the high-level communications model below
illustrates passive monitoring and measurement of information
flowing between production network devices as a potential privacy
issue. The primary purpose of this model is to illustrate collection
of user information of interest with the Measurement Agent
performing the monitoring and storage of the results. This
particular exchange is for DNS Response Time, which most frequently
uses UDP transport.This exchange primarily exposes the IP addresses of
measurement devices and the intent to communicate with, or access
the services of "Domain Name". There may be information on key
points in a service provider's network, such as the address of one
of its DNS servers. The Measurement Agent may be embedded in the
User host, or it may be located in another device capable of
observing user traffic.In principle, any of the Internet User information of interest
(listed above) can be collected and stored in the passive monitoring
scenario.Although the mechanisms for communicating results (beyond the
initial Collector) are beyond the LMAP scope, there are potential
privacy issues related to a single organization's storage and
reporting of measurement results. Both storage and reporting
functions can help to preserve privacy by implementing the
mitigations described below.This section indicates how each of the threats described in apply to the LMAP entities and their communication
and storage of "information of interest".Section 5.1.1 of describes Surveillance
as the "observation or monitoring of and individual's communications
or activities."All of passive measurement is surveillance, with inherent
risks.Active measurement methods which avoid periods of user
transmission indirectly produce a record of times when a subscriber
or authorized user has utilized their Internet access service.Active measurements may also utilize and store a subscriber's
currently assigned IP address when conducting measurements that are
relevant to a specific subscriber. Since the measurements are
time-stamped, the measurement results could provide a record of IP
address assignments over time.Either of the above pieces of information could be useful in
correlation and identification, described below.Section 5.1.2 of describes Stored Data
Compromise as resulting from inadequate measures to secure stored
data from unauthorized or inappropriate access.The primary LMAP entity subject to compromise is the results
storage which serves the Collector function (also applicable to
temporary storage on the Collector itself). Extensive security and
privacy threat mitigations are warranted for the storage system.
Although the scope of its measurement and storage is smaller than
the collector's, an individual Measurement Agent stores sensitive
information temporarily, and also needs protections.The LMAP Controller may have direct access to storage of Service
Parameters, Subscriber information (location, billing, etc.), and
other information which the controlling organization considers
private, and needs protection in this case.The communications between the local storage of the Collector and
other storage facilities (possibly permanent mass storage), is
beyond the scope of the LMAP work at this time, though this
communications channel will certainly need protection as well as the
mass storage.Sections 5.2.1 and 5.2.2 of describes
Correlation as combining various pieces of information to obtain
desired characteristics of an individual, and Identification as
using this process to infer identity.The main risk is that the LMAP system could un-wittingly provide
a key piece of the correlation chain, starting with an unknown
Subscriber's IP address and another piece of information (e.g.,
Subscriber X utilized Internet access from 2000 to 2310 UTC, because
the active measurements were deferred, or sent a name resolution for
www.example.com at 2300 UTC).Sections 5.2.3 and 5.2.4 of describes
Secondary Use as unauthorized utilization of an individual's
information for a purpose the individual did not intend, and
Disclosure is when such information is revealed causing other's
notions of the individual to change, or confidentiality to be
violated.The collection and reporting of passive traffic measurements is a
form of secondary use, and subscribers' permission should be
obtained before measurement. Although user traffic is only
indirectly involved, active measurement results provide limited
information about the subscriber and may constitute secondary
use.This section examines the mitigations listed in section 6 of and their applicability to LMAP systems. Note that
each section in identifies the threat
categories that each technique mitigates.Section 6.1 of encourages collecting and
storing the minimal information needed to perform a task.There are two levels of information needed for LMAP results to be
useful for a specific task: Network Operator and User
troubleshooting, and General results reporting.The minimal supporting information for general results is
conducive to protection of sensitive information, as long as the
results can be aggregated into large categories (e.g., the month of
March, all subscribers West of the Mississippi River). In this case,
all individual identifications (including IP address of the MA) can
be excluded, and only the results applicable to the desired
measurement path are provided.. However, this implies a filtering
process to reduce the information fields allocated to this task,
because greater detail was needed to conduct the measurements in the
first place.For a Network Operator and User troubleshooting a performance
issue or failure, potentially all the network information (e.g., IP
addresses, equipment IDs, location), measurement schedule, service
configuration, measurement results and other information may assist
in the process. This includes the information needed to conduct the
measurements, and represents a need where the maximum relevant
information is desirable, therefore the greatest protections should
be applied.We note that a user may give temporary permission for passive
measurements to enable detailed troubleshooting, but withhold
permission for passive measurements in general. Here the greatest
breadth of sensitive information is potentially exposed, and the
maximum privacy protection must be provided.For MAs with access to the sensitive information of users (e.g.,
within a home or a personal host/handset), it is desirable for the
results collection to minimize the data reported, but also to
balance this desire with the needs of troubleshooting when a service
subscription exists between the user and organization operating the
measurements.For passive measurements where the MA reports flow information to
the Collector, the Collector may perform pre-storage minimization
and other mitigations (below) to help preserve privacy.Section 6.1.1 of describes a way in
which anonymity is achieved: "there must exist a set of individuals
that appear to have the same attributes as the individual", defined
as an "anonymity set".Experimental Methods for anonymization of user identifiable data
applicable to passive measurement have been identified in . However, the findings of several of the same
authors is that "there is increasing evidence that anonymization
applied to network trace or flow data on its own is insufficient for
many data protection applications as in ."Essentially, the details of passive flow measurements can only be
accessed by closed organizations, and unknown injection attacks are
always less expensive than the protections from them. However, some
forms of summarized passive measurement may protect the user's
sensitive information sufficiently well, and so each metric must be
evaluated in the light of privacy.The methods in could be applied more
successfully in active measurement, where there are protections from
injection attack. The successful attack would require breaking the
integrity protection of the LMAP reporting protocol and injecting
measurement results (known fingerprint, see section 3.2 of ) for inclusion with the shared and anonymized
results, then fingerprinting those records to ascertain the
anonymization process.Beside anonymization of measured results for a specific user or
provider, the value of sensitive information can be further diluted
by summarizing the results over many individuals or areas served by
the provider. There is an opportunity enabled by forming anonymity
sets based on the reference path
measurement points in [I-D.ietf-ippm-lmap-path]. For example, all
measurements from the Subscriber device can be identified as
"mp000", instead of using the IP address or other device
information. The same anonymization applies to the Internet Service
Provider, where their Internet gateway would be referred to as
"mp190".Section 6.1.2 of indicates that
pseudonyms, or nicknames, are a possible mitigation to revealing
one's true identity, since there is no requirement to use real names
in almost all protocols.A pseudonym for a measurement device's IP address could be an
LMAP-unique equipment ID. However, this would likely be a permanent
handle for the device, and long-term use weakens a pseudonym's power
to obscure identity.Sections 6.2 and 6.3 of describe User
Participation and Security, respectively.Where LMAP measurements involve devices on the Subscriber's
premises or Subscriber-owned equipment, it is essential to secure
the Subscriber's permission with regard to the specific information
that will be collected.LMAP protocols, devices, and the information they store clearly
need to be secure from unauthorized access. This is the hand-off
between privacy and security considerations, found elsewhere in this
memo.There are no IANA considerations in this memo.This document is a merger of three individual drafts:
draft-eardley-lmap-terminology-02, draft-akhter-lmap-framework-00, and
draft-eardley-lmap-framework-02. Thanks to numerous people for much discussion, directly and on the
LMAP list. This document tries to capture the current conclusions.
Thanks to Juergen Schoenwaelder for his detailed review of the
terminology.Philip Eardley, Trevor Burbridge and Marcelo Bagnulo work in part on
the Leone research project, which receives funding from the European
Union Seventh Framework Programme [FP7/2007-2013] under grant agreement
number 317647.A YANG Data Model for LMAP Measurement AgentsConsiderations on using NETCONF with LMAP Measurement
AgentsAn LMAP application for IPFIXA registry for commonly used metrics. Independent
registriesNetwork Configuration Protocol (NETCONF)YANG-API ProtocolLarge-Scale Measurement of Broadband Performance: Use Cases,
Architecture and Protocol RequirementsInformation Model for Large-Scale Measurement Platforms
(LMAP)The Role of Network Trace Anonymization Under AttackBurkhartACM Computer Communications Review, vol. 40, no. 1,
pp. 6-11IMT-2000 references to Release 9 of GSM-evolved UMTS core
networkBoeing Computer
Services