accounting data with newer data for the same session rather than
accumulating interim data in the buffer.
2.2.3. Non-volatile storage
Since accounting data stored in memory will typically be lost in the
event of a device reboot or a timeout, it may be desirable to provide
non-volatile storage for undelivered accounting data. With the costs
of non-volatile storage declining rapidly, network devices will be
increasingly capable of incorporating non-volatile storage support
over the next few years.
Non-volatile storage may be used to store interim or session records.
As with memory utilization, interim accounting overwrite is desirable
so as to prevent excessive storage consumption. Note that the use of
ASCII data representation enables use of highly efficient text
compression algorithms that can minimize storage requirements. Such
compression algorithms are only typically applied to session records
so as to enable implementation of interim data overwrite.
2.2.4. State on the accounting management system
In order to keep track of received accounting data, accounting
management systems may need to keep state on managed devices or
concurrent sessions. Since the number of devices is typically much
smaller than the number of concurrent sessions, it is desirable to
keep only per-device state if possible.
2.2.5. CPU requirements
CPU consumption of the managed and managing nodes will be
proportional to the complexity of the required accounting processing.
Operations such as ASN.1 encoding and decoding,
compression/decompression, and encryption/decryption can consume
considerable resources, both on accounting clients and servers.
The effect of these operations on accounting system reliability
should not be under-estimated, particularly in the case of devices
with moderate CPU resources. In the event that devices are over-
taxed by accounting tasks, it is likely that overall device
reliability will suffer.
2.2.6. Efficiency measures
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| Resource | Efficiency measures |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| Network | Batching |
| Bandwidth | Compression |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| Memory | Compression |
| | Interim accounting overwrite |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| Non-volatile | Compression |
| Storage | Interim accounting overwrite |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| System | Per-device state |
| state | |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | |
| CPU | Hardware assisted |
| requirements | compression/encryption |
| | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2.3. Data collection models
Several data collection models are currently in use today for the
purposes of accounting data collection. These include:
Polling model
Event-driven model without batching
Event-driven model with batching
Event-driven polling model
2.3.1. Polling model
In the polling model, an accounting manager will poll devices for
accounting information at regular intervals. In order to ensure
against loss of data, the polling interval will need to be shorter
than the maximum time that accounting data can be stored on the
polled device. For devices without non-volatile stage, this is
typically determined by available memory; for devices with non-
volatile storage the maximum polling interval is determined by the
size of non-volatile storage.
The polling model results in an accumulation of data within
individual devices, and as a result, data is typically transferred to
the accounting manager in a batch, resulting in an efficient transfer
process. In terms of Accounting Manager state, polling systems scale
with the number of managed devices, and system bandwidth usage scales
with the amount of data transferred.
Without non-volatile storage, the polling model results in loss of
accounting data due to device reboots, but not due to packet loss or
network failures of sufficiently short duration to be handled within
available memory. This is because the Accounting Manager will
continue to poll until the data is received. In situations where
operational difficulties are encountered, the volume of accounting
data will frequently increase so as to make data loss more likely.
However, in this case the polling model will detect the problem since
attempts to reach the managed devices will fail.
The polling model scales poorly for implementation of shared use or
roaming services, including wireless data, Internet telephony, QoS
provisioning or Internet access. This is because in order to
retrieve accounting data for users within a given domain, the
Accounting Management station would need to periodically poll all
devices in all domains, most of which would not contain any relevant
data. There are also issues with processing delay, since use of a
polling interval also implies an average processing delay of half the
polling interval. This may be too high for accounting data that
requires low processing delay. Thus the event-driven polling or the
pure event-driven approach is more appropriate for usage sensitive
billing applications such as shared use or roaming implementations.
Per-device state is typical of polling-based network management
systems, which often also carry out accounting management functions,
since network management systems need to keep track of the state of
network devices for operational purposes. These systems offer
average processing delays equal to half the polling interval.
2.3.2. Event-driven model without batching
In the event-driven model, a device will contact the accounting
server or manager when it is ready to transfer accounting data. Most
event-driven accounting systems, such as those based on RADIUS
accounting, described in [4], transfer only one accounting event per
packet, which is inefficient.
Without non-volatile storage, a pure event-driven model typically
stores accounting events that have not yet been delivered only until
the timeout interval expires. As a result this model has the
smallest memory requirements. Once the timeout interval has expired,
the accounting event is lost, even if the device has sufficient
buffer space to continue to store it. As a result, the event-driven
model is the least reliable, since accounting data loss will occur
due to device reboots, sustained packet loss, or network failures of
duration greater than the timeout interval. In event-driven
protocols without a "keep alive" message, accounting servers cannot
assume a device failure should no messages arrive for an extended
period. Thus, event-driven accounting systems are typically not
useful in monitoring of device health.
The event-driven model is frequently used in shared use networks and
roaming, since this model sends data to the recipient domains without
requiring them to poll a large number of devices, most of which have
no relevant data. Since the event-driven model typically does not
support batching, it permits accounting records to be sent with low
processing delay, enabling application of fraud prevention
techniques. However, because roaming accounting events are
frequently of high value, the poor reliability of this model is an
issue. As a result, the event-driven polling model may be more
appropriate.
Per-session state is typical of event-driven systems without
batching. As a result, the event-driven approach scales poorly.
However, event-driven systems offer the lowest processing delay since
events are processed immediately and there is no possibility of an
event requiring low processing delay being caught behind a batch
transfer.
2.3.3. Event-driven model with batching
In the event-driven model with batching, a device will contact the
accounting server or manager when it is ready to transfer accounting
data. The device can contact the server when a batch of a given size
has been gathered, when data of a certain type is available or after
a minimum time period has elapsed. Such systems can transfer more
than one accounting event per packet and are thus more efficient.
An event-driven system with batching will store accounting events
that have not yet been delivered up to the limits of memory. As a
result, accounting data loss will occur due to device reboots, but
not due to packet loss or network failures of sufficiently short
duration to be handled within available memory. Note that while
transfer efficiency will increase with batch size, without non-
volatile storage, the potential data loss from a device reboot will
also increase.
Where event-driven systems with batching have a keep-alive interval
and run over reliable transport, the accounting server can assume
that a failure has occurred if no messages are received within the
keep-alive interval. Thus, such implementations can be useful in
monitoring of device health. When used for this purpose the average
time delay prior to failure detection is one half the keep-alive
interval.
Through implementation of a scheduling algorithm, event-driven
systems with batching can deliver appropriate service to accounting
events that require low processing delay. For example, high-value
inter-domain accounting events could be sent immediately, thus
enabling use of fraud-prevention techniques, while all other events
would be batched. However, there is a possibility that an event
requiring low processing delay will be caught behind a batch transfer
in progress. Thus the maximum processing delay is proportional to
the maximum batch size divided by the link speed.
Event-driven systems with batching scale with the number of active
devices. As a result this approach scales better than the pure
event-driven approach, or even the polling approach, and is
equivalent in terms of scaling to the event-driven polling approach.
However, the event-driven batching approach has lower processing
delay than the event-driven polling approach, since delivery of
accounting data requires fewer round-trips and events requiring low
processing delay can be accommodated if a scheduling algorithm is
employed.
2.3.4. Event-driven polling model
In the event-driven polling model an accounting manager will poll the
device for accounting data only when it receives an event. The
accounting client can generate an event when a batch of a given size
has been gathered, when data of a certain type is available or after
a minimum time period has elapsed. Note that while transfer
efficiency will increase with batch size, without non-volatile
storage, the potential data loss from a device reboot will also
increase.
Without non-volatile storage, an event-driven polling model will lose
data due to device reboots, but not due to packet loss, or network
partitions of short-duration. Unless a minimum delivery interval is
set, event-driven polling systems are not useful in monitoring of
device health.
The event-driven polling model can be suitable for use in roaming
since it permits accounting data to be sent to the roaming partners
with low processing delay. At the same time non-roaming accounting
can be handled via more efficient polling techniques, thereby
providing the best of both worlds.
Where batching can be implemented, the state required in event-driven
polling can be reduced to scale with the number of active devices.
If portions of the network vary widely in usage, then this state may
actually be less than that of the polling approach. Note that
processing delay in this approach is higher than in event-driven
accounting with batching since at least two round-trips are required
to deliver data: one for the event notification, and one for the
resulting poll.
2.3.5. Data collection summary
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | | |
| Model | Pros | Cons |
| | | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Polling | Per-device state | Not robust |
| | Robust against | against device |
| | packet loss | reboot, server |
| | Batch transfers | or network |
| | | failures* |
| | | Polling interval |
| | | determined by |
| | | storage limit |
| | | High processing |
| | | delay |
| | | Unsuitable for |
| | | use in roaming |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Event-driven, | Lowest processing | Not robust |
| no batching | delay | against packet |
| | Suitable for | loss, device |
| | use in roaming | reboot, or |
| | | network |
| | | failures* |
| | | Low efficiency |
| | | Per-session state |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Event-driven, | Single round-trip | Not robust |
| with batching | latency | against device |
| and | Batch transfers | reboot, network |
| scheduling | Suitable for | failures* |
| | use in roaming | |
| | Per active device | |
| | state | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Event-driven | Batch transfers | Not robust |
| polling | Suitable for | against device |
| | use in roaming | reboot, network |
| | Per active device | failures* |
| | state | Two round-trip |
| | | latency |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Key
* = addressed by non-volatile storage
3. Review of Accounting Protocols
Accounting systems have been successfully implemented using protocols
such as RADIUS, TACACS+, and SNMP. This section describes the
characteristics of each of these protocols.
3.1. RADIUS
RADIUS accounting, described in [4], was developed as an add-on to
the RADIUS authentication protocol, described in [3]. As a result,
RADIUS accounting shares the event-driven approach of RADIUS
authentication, without support for batching or polling. As a
result, RADIUS accounting scales with the number of accounting events
instead of the number of devices, and accounting transfers are
inefficient.
Since RADIUS accounting is based on UDP and timeout and retry
parameters are not specified, implementations vary widely in their
approach to reliability, with some implementations retrying until
delivery or buffer exhaustion, and others losing accounting data
after a few retries. Since RADIUS accounting does not provide for
application-layer acknowledgments or error messages, a RADIUS
Accounting-Response is equivalent to a transport-layer acknowledgment
and provides no protection against application layer malfunctions.
Due to the lack of reliability, it is not possible to do simultaneous
usage control based on RADIUS accounting alone. Typically another
device data source is required, such as polling of a session MIB or a
command-line session over telnet.
RADIUS accounting implementations are vulnerable to packet loss as
well as application layer failures, network failures and device
reboots. These deficiencies are magnified in inter-domain accounting
as is required in roaming ([1],[2]). On the other hand, the event-
driven approach of RADIUS accounting is useful where low processing
delay is required, such as credit risk management or fraud detection.
While RADIUS accounting does provide hop-by-hop authentication and
integrity protection, and IPSEC can be employed to provide hop-by-hop
confidentiality, data object security is not supported, and thus
systems based on RADIUS accounting are not capable of being deployed
with untrusted proxies, or in situations requiring auditability, as
noted in [2].
While RADIUS does not support compression, IP compression, described
in [5], can be employed to provide this. While in principle
extensible with the definition of new attributes, RADIUS suffers from
the very small standard attribute space (256 attributes).
3.2. TACACS+
TACACS+ offers an accounting model with start, stop, and interim
update messages. Since TACACS+ is based on TCP, implementations are
typically resilient against packet loss and short-lived network
partitions, and TACACS+ scales with the number of devices. Since
TACACS+ runs over TCP, it offers support for both transport layer and
application layer acknowledgments, and is suitable for simultaneous
usage control and handling of accounting events that require moderate
though not the lowest processing delay.
TACACS+ provides for hop-by-hop authentication and integrity
protection as well as hop-by-hop confidentiality. Data object
security is not supported, and therefore systems based on TACACS+
accounting are not deployable in the presence of untrusted proxies.
While TACACS+ does not support compression, IP compression, described
in [5], can be employed to provide this.
3.3. SNMP
SNMP, described in [19],[27]-[41], has been widely deployed in a wide
variety of intra-domain accounting applications, typically using the
polling data collection model. Polling allows data to be collected
on multiple accounting events simultaneously, resulting in per-device
state. Management applications are able to retry requests when a
response is not received, providing resiliency against packet loss or
even short-lived network partitions. Implementations without non-
volatile storage are not robust against device reboots or network
failures, but when combined with non-volatile storage they can be
made highly reliable.
SMIv1, the data modeling language of SNMPv1, has traps to permit
trap-directed polling, but the traps are not acknowledged, and lost
traps can lead to a loss of data. SMIv2, used by SNMPv2c and SNMPv3,
has Inform Requests which are acknowledged notifications. This makes
it possible to implement a more reliable event-driven polling model
or event-driven batching model. However, we are not aware of any
SNMP-based accounting implementations currently built on the use of
Informs.
3.3.1. Security services
SNMPv1 and SNMPv2c support per-packet authentication and read-only
and read-write access profiles, via the community string. This
clear-text password approach provides only trivial authentication,
and no per-packet integrity checks, replay protection or
confidentiality. View-based access control [40] can be supported
using the snmpCommunityMIB, defined in [11], and SNMPv1 or SNMPv2c
messages. The updated SNMP architecture [rfc2571] supports per-
packet hop-by-hop authentication, integrity and replay protection,
confidentiality and access control.
The SNMP User Security Model (USM) [38] uses shared secrets, and when
the product of the number of domains and devices is large, such as in
inter-domain accounting applications, the number of shared secrets
can get out of hand. The localized key capability in USM allows a
manager to have one central key, sharing it with many SNMP entities
in a localized way while preventing the other entities from getting
at each other's data. This can assist in cross-domain security if
deployed properly.
SNMPv3 does not support end-to-end data object integrity and
confidentiality; SNMP proxy entities decrypt and re-encrypt the data
they forward. In the presence of an untrusted proxy entity, this
would be inadequate.
3.3.2. Application layer acknowledgments
SNMP uses application-layer acknowledgment to indicate that data has
been processed. SNMP Responses to get, get-next, or get-bulk
requests return the requested data, or an error code indicating the
nature of the error encountered.
A noError SNMP Response to a SET command indicates that the requested
assignments were made by the application. SNMP SETs are atomic; the
command either succeeds or fails. An error-response indicates that
the entity received the request, but did not succeed in executing it.
Notifications do not use acknowledgements to indicate that data has
been processed. The Inform notification returns an acknowledgement
of receipt, but not of processing, by design. Since the updated SNMP
architecture treats entities as peers with varying levels of
functionality, it is possible to use SETs in either direction between
cooperating entities to achieve processing acknowledgements.
There are eighteen SNMP error codes. The design of SNMP makes
service-specific error codes unnecessary and undesirable.
3.3.3. Proxy forwarders
In the accounting management architecture, proxy forwarders play an
important role, forwarding intra and inter-domain accounting events
to the correct destinations. The proxy forwarder may also play a
role in a polling or event-driven polling architecture.
The functionality of an SNMP Proxy Forwarder is defined in [39]. For
example, the network devices may be configured to send notifications
for all domains to the Proxy Forwarder, and the devices may be
configured to allow the Proxy Forwarder to access all MIB data.
The use of proxy forwarders may reduce the number of shared secrets
required for inter-domain accounting. With Proxy Forwarders, the
domains could share a secret with the Proxy Forwarder, and in turn,
the Proxy Forwarder could share a secret with each of the devices.
Thus the number of shared secrets will scale with the sum of the
number of devices and domains rather than the product.
The engine of an SNMP Proxy Forwarder does not look inside the PDU of
the message except to determine to which SNMP engine the PDU should
be forwarded or which local SNMP application should process the PDU.
The SNMP Proxy Forwarder does not modify the varbind values; it does
not modify the varbind list except to translate between SNMP
versions; and it does not provide any varbind level access control.
3.3.4. Domain-based access controls in SNMP
Domain-based access controls are required where multiple
administrative domains are involved, such as in the shared use
networks and roaming associations described in [1]. Since the same
device may be accessed by multiple organizations, it is often
necessary to control access to accounting data according to the
user's organization. This ensures that organizations may be given
access to accounting data relating to their users, but not to data
relating to users of other organizations.
In order to apply domain-based access controls, in inter-domain
accounting, it is first necessary to identify the data subset that is
to have its access controlled. Several conceptual abstractions are
used for identifying subsets of data in SNMP. These include engines,
contexts, and views. This section describes how this functionality
may be applied in intra and inter-domain accounting.
3.3.4.1. Engines
The new SNMP architecture, described in [27], added the concept of an
SNMP engine to improve mobility support and to identify which data
source is being referenced. The engine is the portion of an SNMP
entity that constructs messages, provides security functions, and
maps to the transport layer. Traditional agents and traditional
managers each contain an SNMP engine. engineID allows an SNMP engine
to be uniquely identified, independent of the address where it is
attached to the network.
A securityEngineID field in a message identifies the engine which
provides access to the security credentials contained in the message
header. A contextEngineID field in a message identifies the engine
which provides access to the data contained in the PDU.
The SNMPv3 message format explicitly passes both. In SNMPv1 and
SNMPv2c, the data origin is typically assumed to be the
communications endpoint (SNMP agent). SNMPv1 and SNMPv2c messages
contain a community name; the community name and the source address
can be mapped to an engineID via the snmpCommunityTable, described in
[11].
3.3.4.2. Contexts
Contexts are used to identify subsets of objects, within the scope of
an engine, that are tied to instrumentation. A contextName refers to
a particular subset within an engine.
Contexts are commonly tied to hardware components, to logical
entities related to the hardware components, or to logical services.
For example, contextNames might include board5, board7, repeater1,
repeater2, etc.
An SNMP agent populates a read-only dynamic table to tell the manager
what contexts it recognizes. Typically contexts are defined by the
agent rather than the manager since if the manager defined them, the
agent would not know how to tie the contexts to the underlying
instrumentation. It is possible that MIB modules could be defined to
allow a manager to assign contextNames to a logical subset of
instrumentation.
While each context may support instances of multiple MIB modules,
each contextName is limited to one instance of a particular MIB
module. If multiple instances of a MIB module are required per
engine, then unique contextNames must be defined (e.g. repeater1,
repeater2). The default context "" is used for engines which only
support single instances of MIB modules, and it is used for MIB
modules where it only makes sense to have one instance of that MIB
module in an engine and that instance must be easy to locate, such as
the system MIB or the security MIBs.
SNMPv3 messages contain contextNames which are limited to the scope
of the contextEngineID in the message. SNMPv1 and SNMPv2c messages
contain communities which can be mapped to contextNames within the
local engine, or can be mapped to contextNames within other engines
via the snmpCommunityTable, described in [11].
3.3.4.3. Views
Views are defined in the View-based Access Control Model. A view is
a mask which is used to determine access to the managed objects in a
particular context. The view identifies which objects are visible,
by specifying OIDs of the subtrees included and excluded. There is
also a mechanism to allow wildcards in the OID specification.
For example, it is possible to define a view that includes RMON
tables, and another view that includes only the SNMPv3 security
related tables. Using these views, it is possible to allow access to
the RMON view for users Joe and Josephine (the RMON administrators),
and access to the SNMPv3 security tables for user Adam (the SNMP
security Administrator).
Views can be set up with wildcards. For a table that is indexed
using IP addresses, Joe can be allowed access to all rows in given
RMON tables (e.g. the RMON hostTable) that are in the subnet
10.2.x.x, while Josephine is given access to all rows for subnet
10.200.x.x.
Views filter at the name level (OIDs), not at the value level, so
defining views based on the values of non-index data is not
supported. In this example, were the IP address to have been used
merely as a data item rather than an index, it would not be possible
to utilize view-based access control to achieve the desired objective
(delegation of administrative responsibility according to subnet).
View-based access control is independent of message version. It can
be utilized by entities using SNMPv1, SNMPv2c, or SNMPv3 message
formats.
3.3.5. Inter-domain access-control alternatives
As the number of network devices within the shared use or roaming
network grows, the polling model of data collection becomes
increasingly impractical since most devices will not carry data
relating to the polling organization. As a result, shared-use
networks or roaming associations relying on SNMP-based accounting
have generally collected data for all organizations and then sorted
the resulting session records for delivery to each organization.
While functional, this approach will typically result in increased
processing delay as the number of organizations and data records
grows.
This issue can be addressed in SNMP using the event-driven, event-
driven polling or event-driven batching approaches. Traps and
Informs permit SNMP-enabled devices to notify domains that have
accounting data awaiting collection. SNMP Applications [39] defines
a standard module for managing notifications.
To use the event-driven approaches, the device must be able to
determine when information is available for a domain. Domain-
specific data can be differentiated at the SNMP agent level through
the use of the domain as an index, and the separation of data into
domain-specific contexts.
3.3.5.1. Domain as index
View-based access control [40] allows multiple fine-grained views of
an SNMP MIB to be assigned to specific groups of users, such that
access rights to the included data elements depend on the identity of
the user making the request.
For example, all users of bigco.com which are allowed access to the
device would be defined in the User-based security MIB module (or
other security model MIB module). For simplicity in administering
access control, the users can be grouped using a vacmGroupName, e.g.
bigco. A view of a subset of the data objects in the MIB can be
defined in the vacmViewFamilyTreeTable. A vacmAccessTable pairs
groups and views. For messages received from users in the bigco
group, access would only be provided to the data permitted to be
viewed by bigco users, as defined in the view family tree. This
requires that each domain accessing the data be given one or more
separate vacmGroupNames, an appropriate ViewTable be defined, and the
vacmAccessTable be configured for each group.
Views filter at the name (OID) level, not at the data (value) level.
When using views to filter by domain it is necessary to use the
domain as an index. Standard view-based access control is not
designed to filter based on the values on non-indexed fields.
For example, a table of session data could be indexed by record
number and domain, allowing a view to be defined that could restrict
access to bigco data to the administrators of the bigco domain.
An advantage of using domains as an index is that this technique can
be used with SNMPv1 and SNMPv2c agents as well as with SNMPv3 agents.
A disadvantage is that the MIB modules must be specifically designed
for this purpose. Since existing MIB modules rarely use the domain
as an index, domain separation cannot be enabled within legacy MIB
modules using this technique.
SNMP does support a RowPointer convention that could be used to
define a new table, indexed by domain, which holds tuples between the
domain and existing rows of data. This would introduce issues of
synchronization between tables.
3.3.5.2. Contexts
ContextNames can be used to differentiate multiple instances of a MIB
module within an engine.
Individual domains, such as bigco.com, could be mapped to logical
contexts, such as a bigco context. The agent would need to create
and recognize new contexts and to know which instrumentation is
associated with the logical context. The agent needs to collect
accounting data by domain and make the data accessible via distinct
contexts, so that access control can be applied to the context to
prevent disclosure of sensitive information to the wrong domain. The
VACM access control views are applied relative to the context, so an
operation can be permitted or denied a user based on the context
which contains the data.
Domain separation is handled by using contextName to differentiate
multiple virtual tables. For example, if accounting data has been
collected on users with the bigco.com and smallco.com domains, then a
separate virtual instance of the accounting session record table
would exist for each domain, and each domain would have a
corresponding contextName. When a get-bulk request is made with a
contextName of bigco, then data from the virtual table in the bigco
context, i.e. corresponding to the bigco.com domain, would be
returned.
There are a number of design approaches to creating new contexts and
associating the contexts with appropriate instrumentation, most
notably a sub-agent approach and a manager-configured MIB approach.
AgentX [51], which standardizes a registration protocol between sub-
agents and master agents to simplify SNMP agent implementation,
allows for the creation and recognition of new contextNames when a
subagent registers to provide support for a particular MIB subtree
range. The sub-agent knows how to support a particular
functionality, e.g. instrumentation exposed via a range of MIB
objects. Based on values detected in the data, such as
source=bigco.com, the sub-agent could determine that a new domain
needed to be tracked and create the appropriate context for the
collection of the data, plus the appropriate access control entries.
The determination could be table-driven, using MIB configuration.
A manager-driven approach could use a MIB module to predefine
contextNames corresponding to the domains of interest, and to
indicate which objects should be collected, how to differentiate to
which domain the data should be applied based on a specified
condition, and what access control rules apply to the context.
Either technique could associate existing MIB modules to domain-
specific contexts, so domain separation can be applied to MIB modules
not specifically designed with domain separation in mind. Legacy
agents would not be designed to do this, so they would need to be
updated to support inter-domain separation and VACM access control.
The use of contextNames for inter-domain separation represents new
territory, so careful consideration would be needed in designing the
MIB modules and applications to provide domain to context and context
to instrumentation mappings, and to ensure that security is not
weakened.
3.3.6. Outstanding issues
There are issues that arise when using SNMP for transfer of bulk
data, including issues of latency, network overhead, and table
retrieval, as discussed in [49].
In accounting applications, management stations often must retrieve
large tables. Latency can be high, even with the get-bulk operation,
because the response must fit into the largest supported packet size,
requiring multiple round-trips. Transfers may be serialized and the
resulting latency will be a combination of multiple round-trip times,
possible timeout and re-transmission delays and processing overhead,
which may result in unacceptable performance. Since data may change
during the course of multiple retrievals, it can be difficult to get
a consistent snapshot.
For bulk transfers, SNMP network overhead can be high due to the lack
of compression, inefficiency of BER encoding, the transmission of
redundant OID prefixes, and the "get-bulk overshoot problem". In
bulk transfer of a table, the OIDs transferred are redundant: all OID
prefixes up to the column number are identical, as are the instance
identifier postfixes of all entries of a single table row. Thus it
may be possible to reduce this redundancy by compressing the OIDs, or
by not transferring an OID with each variable.
The "get-bulk overshoot problem", described in reference [50], occurs
when using the get-bulk PDU. The problem is that the manager
typically does not know the number of rows in the table. As a
result, it must either request too many rows, retrieving unneeded
data, or too few, resulting in the need for multiple get-bulk
requests. Note that the "get-bulk overshoot" problem may be
preventable on the agent side. Reference [41] states that an agent
can terminate the get-bulk because of "local constraints" (see items
1 and 3 on pages 15/16 of [41]). This could be interpreted to mean
that it is possible to stop at the end of a table.
3.3.6.1. Ongoing research
To address issues of latency and efficiency, the Network Management
Research Group (NMRG) was formed within the Internet Research Task
Force (IRTF). Since the NMRG work is research and is not on the
standards track, it should be understood that the NMRG proposals may
never be standardized, or may change substantially during the
standardization process. As a result, these proposals represent
works in progress and are not readily available for use.
The proposals under discussion in the IRTF Network Management
Research Group (NMRG) are described in [46]. These include an SNMP-
over-TCP transport mapping, described in [47]; SNMP payload
compression, described in [48]; and the addition of a "get subtree"
PDU or the subtree retrieval MIB [50].
The SNMP-over-TCP transport mapping may result in substantial latency
reductions in table retrieval. The latency reduction of an SNMP-
over-TCP transport mapping will likely manifest itself primarily in
the polling, event-driven polling and event-driven batching modes.
Payload compression methods include compression of the IP packet, as
described in [5] or compression of the SNMP payload, described in
[48].
Proposed improvements to table retrieval include a subtree retrieval
MIB and the addition of a get-subtree PDU. The subtree retrieval MIB
[50] requires no changes to the SNMP protocol or SNMP protocol
engine, so it can be implemented and deployed more easily than a
change to the protocol. The addition of a get-subtree PDU implies
changes to the protocol and to the engines of all SNMP entities which
would support it. Since it may be possible to address the "get-bulk
overshoot problem" without changes to the SNMP protocol, the
necessity of this modification is controversial.
Reference [49] also discusses file-based storage of SNMP data, and
use of an FTP MIB, to enable storage of SNMP data in non-volatile
storage, and subsequent bulk transfer via FTP. This approach would
require implementation of additional MIB modules as well as FTP, and
requires separate security mechanisms such as IPSEC to provide
authentication, replay, integrity protection and confidentiality for
the data in transit. The file-based transfer approach has an
important benefit - compatibility with non-volatile storage.
Issues of legacy support exist with the NMRG proposals. Devices
which do not implement the new functionality would need to be
accommodated. This is especially problematic for proxy forwarders,
which may need to act as translators between new and legacy entities.
In these situations, the overhead of translation may offset the
benefits of the new technologies.
3.3.6.2. On-going security extension research
In order to simplify key management and enable use of certificate-
based security in SNMPv3, a Kerberos Security Model (KSM) for SNMPv3
has been proposed in [44]. This memo is not on the standards track,
and therefore is not yet readily available for use.
Use of Kerberos with SNMPv3 requires storage of a key on the KDC for
each device and domain, while dynamically generating a session key
for conversations between domains and devices. In terms of stored
keys, the KSM approach scales with the sum of devices and domains; in
terms of dynamic session keys, it scales as the product of domains
and devices.
As Kerberos is extended to allow initial authentication via public
key, as described in [42], and cross-realm authentication, as
described in [43], the KSM inherits these capabilities. As a result,
this approach may have potential to reduce or even eliminate the
shared secret management problem. However, it should also be noted
that certificate-based authentication can strain the limits of UDP
packet sizes supported in SNMP implementations, so that alternate
transport mappings may be required to support this.
An IPSEC-based security model for SNMPv3 has been discussed.
Implementation of such a security model would require the SNMPv3
engine to be able to retrieve the properties of the IPSEC security
association used to protect the SNMPv3 traffic. This would include
the security services invoked, as well as information relating to the
other endpoint, such as the authentication method and presented
identity and certificate. To date such APIs have not been widely
implemented, and in addition, most IPSEC implementations only support
machine certificates, which may not provide the required granularity
of identification. Thus, an IPSEC-based security model for SNMPv3
would probably take several years to come to fruition.
3.3.7. SNMP summary
Given the wealth of existing accounting-related MIB modules, it is
likely that SNMP will remain a popular accounting protocol for the
foreseeable future.
Support for notifications makes it possible to implement the event-
driven, event-driven polling and event-driven batching models. This
makes it possible to notify domains of available data rather than
requiring them to poll for it, which is critical in shared use
networks and roaming.
Given the SNMPv3 security enhancements, it is desirable for SNMP-
based intra-domain accounting implementations to upgrade to SNMPv3.
Such an upgrade is virtually mandatory for inter-domain applications.
In inter-domain accounting, the burden of managing SNMPv3 shared
secrets can be reduced via the localized key capability or via
implementation of a Proxy Forwarder. In the long term, alternative
security models such as the Kerberos Security Model may further
reduce the effort required to manage security and enable streamlined
inter-domain operation.
SNMP-based accounting has limitations in terms of efficiency and
latency that may make it inappropriate for use in situations
requiring low processing delay or low overhead. This includes usage
sensitive billing applications where fraud detection may be required.
These issues can be addressed via proposals under discussion in the
IRTF Network Management Research Group (NMRG). The experimental SNMP
over TCP transport mapping may prove helpful at reducing latency.
Depending on the volume of data, some form of compression may also be
worth considering. However, since these proposals are still in the
research stage, and are not on the standards track, these
capabilities are not readily available, and the specifications could
change considerably before they reach their final form.
SNMP supports separation of accounting data by domain, using either
of two general approaches with the VACM access control model. The
domain as index approach can be used if the desired MIB module
supports domain indexing, or it can implemented using an additional
table. The domain-context approach can be used in agents which
support dynamic logical contexts and a domain-to-context and
context-to-instrumentation mapping mechanism. Either approach can be
supported using SNMPv1, SNMPv2c, or SNMPv3 messages, by utilizing the
snmpCommunitytable [11] to provide a community-to-context mapping.
4. Review of Accounting Data Transfer
In order for session records to be transmitted between accounting
servers, a transfer protocol is required. Transfer protocols in use
today include SMTP, FTP, and HTTP. For a review of accounting
attributes and record formats, see [45].
Reference [49] contains a discussion of alternative encodings for SMI
data types, as well as alternative protocols for transmission of
accounting data. For example, [49] describes how MIME tags and XML
DTDs may be used for encoding of SNMP messages or SMI data types.
This enables data from SNMP MIBs to be transported using any protocol
that can encapsulate MIME or XML, including SMTP and HTTP.
4.1. SMTP
To date, few accounting management systems have been built on SMTP
since the implementation of a store-and-forward message system has
traditionally required access to non-volatile storage which has not
been widely available on network devices. However, SMTP-based
implementations have many desirable characteristics, particularly
with regards to security.
Accounting management systems using SMTP for accounting transfer will
typically support batching so that message processing overhead will
be spread over multiple accounting records. As a result, these
systems result in per-active device state. Since accounting systems
using SMTP as a transfer mechanism have access to substantial non-
volatile storage, they can generate, compress if necessary, and store
accounting records until they are transferred to the collection site.
As a result, accounting systems implemented using SMTP can be highly
efficient and scalable. Using IPSEC, TLS or Kerberos, hop-by-hop
security services such as authentication, integrity protection and
confidentiality can be provided.
As described in [13] and [15], data object security is available for
SMTP, and in addition, the facilities described in [12] make it
possible to request and receive signed receipts, which enables non-
repudiation as described in [12]-[17]. As a result, accounting
systems utilizing SMTP for accounting data transfer are capable of
satisfying the most demanding security requirements. However, such
systems are not typically capable of providing low processing delay,
although this may be addressed by the enhancements described in [20].
4.2. Other protocols
File transfer protocols such as FTP and HTTP have been used for
transfer of accounting data. For example, Reference [9] describes a
means for representing ASN.1-based accounting data for storage on
archival media. Through the use of the Bulk File MIB, accounting
data from an SNMP MIB can be stored in ASN.1, bulk binary or Bulk
ASCII format, and then subsequently retrieved as required using the
FTP Client MIB.
Given access to sufficient non-volatile storage, accounting systems
based on record formats and transfer protocols can avoid loss of data
due to long-duration network partitions, server failures or device
reboots. Since it is possible for the transfer to be driven from the
collection site, the collector can retry transfers until successful,
or with HTTP may even be able to restart partially completed
transfers. As a result, file transfer-based systems can be made
highly reliable, and the batching of accounting records makes
possible efficient transfers and application of required security
services with lessened overhead.
5. Summary
As noted previously in this document, accounting applications vary in
their security and reliability requirements. Some uses such as
capacity planning may only require authentication, integrity and
replay protection, and modest reliability. Other applications such
as inter-domain usage-sensitive billing may require the highest
degree of security and reliability, since in these cases the transfer
of accounting data will lead directly to the transfer of funds.
Since accounting applications do not have uniform security and
reliability requirements, it is not possible to devise a single
accounting protocol and set of security services that will meet all
needs. Rather, the goal of accounting management should be to
provide a set of tools that can be used to construct accounting
systems meeting the requirements of an individual application. As a
result, it is important to analyze a given accounting application to
ensure that the methods chosen meet the security and reliability
requirements of the application.
Based on an analysis of the requirements, it appears that existing
deployed protocols are capable of meeting the requirements for
intra-domain capacity planning and non-usage sensitive billing. In
these applications efficient transfer of bulk data is useful although
not critical. Thus, it is possible to use SNMPv3 to satisfy these
requirements, without the NMRG extensions. These include TCP
transport mapping, sub-tree retrieval, and OID compression.
In inter-domain capacity planning and non-usage sensitive billing,
the security and reliability requirements are greater. As a result,
no existing deployed protocol satisfies the requirements. For
example, existing protocols lack data object security support and
extensions to improve scalability of inter-domain authentication are
needed, such as the Kerberos Security Model (KSM) for SNMPv3.
For usage sensitive billing, as well as cost allocation and auditing
applications, the reliability requirement are greater. Here
transport layer reliability is required to provide robustness against