Internet DRAFT - draft-anderson-req-dyn-part
draft-anderson-req-dyn-part
Internet Draft T. Anderson
Expiration: August 2001 Intel
File: draft-anderson-req-dyn-part-00.txt A. Doria
Nortel Networks
February 2001
Requirements for the Dynamic Partitioning of Network Elements
Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026. Internet-Drafts are
working documents of the Internet Engineering Task Force (IETF), its
areas, and its working groups. Note that other groups may also
distribute working documents as Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other documents
at any time. It is inappropriate to use Internet-Drafts as
reference material or to cite them other than as ``work in
progress.''
To view the current status of any Internet-Draft, please check the
``1id-abstracts.txt'' listing contained in an Internet-Drafts
Shadow Directory, see http://www.ietf.org/shadow.html.
Conventions used in this document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in [RFC2119].
Abstract
This document identifies a set of requirements for the mechanisms
used to dynamically reallocate the resources of a partitionable
network element (NE). These requirements are particularly critical
in the case of an operator creating a virtual NE (by partitioning a
physical NE) and then leasing control of that virtual NE to a third
party.
1. Definitions
In this document, the following definitions will be used.
Anderson et. al. Expires August 2001 [Page 1]
Internet Draft Requirements for Dynamic Partitioning Feb 2001
Partition - A partition is a defined set of a physical network
element (NE) resources that can be used to create a virtual NE.
Active Partition - An active partition is a partition in which the
resources are in use; either under the direct control of a separate
controller or under internal policy based control.
Controller - The entity responsible for controlling the operations
of an active partition.
Static Partitioning - In static partitioning, no changes can be made
to any active partitionÆs resources without requiring a restart of
that partition. Instances of repartitioning in which connections to
controllers are disconnected before resources are reallocated
therefore fall into this category.
Dynamic Partitioning - In dynamic partitioning, an active
partitionÆs resources can be reapportioned without requiring a
restart of the partition.
Frozen Partition - A frozen partition is an active partition which
is in the process of shutdown. A frozen partition's unused
resources are relinquished, but all current connections are allowed
to remain until removed by the controller. As connections close the
resources are returned to the NE.
Deterministic Partitioning - In deterministic partitioning, each
active partition is given an allotted quantity of each resource.
The usage of resources in one active partition do not influence the
resources available to another active partition. All discussions in
these requirements presuppose the use of deterministic partitioning.
Statistical Partitioning - In statistical partitioning, some or all
resources are pooled among the active partitions, and allocations
may be based on percentages or on some other metric. Discussion of
statistical partitions is outside the scope of these requirements.
Proactive Notification - A proactive notification is a message sent
from a NE to its controller at the time an event occurs.
Specifically, if a NE asynchronously sends the controller a message
when it is dynamically partitioned, we say that the NE has
proactively notified its controller of the resource reapportionment.
Explicit Reactive Notification - In explicit reactive notification,
the NE does not asynchronously send a message when dynamic
partitioning occurs. Instead, the NE includes a "resource changed"
error code in the response to a subsequent request by the
controller.
Implicit Reactive Notification - This is similar to an Explicit
Reactive Notification except that the protocol does not contain an
explicit "resource changed" error. In this case, all that the NE
Anderson et. al. Expires August 2001 [Page 2]
Internet Draft Requirements for Dynamic Partitioning Feb 2001
can do is to indicate that some unspecified error has occurred when
the controller attempts to use non-allocated resources.
2. Introduction
Several logical entities are involved in the partitioning and
control of a NE. First, there is the physical NE itself that is
capable of having its resources partitioned. (This also implies the
ability to enforce this division of resources between competing
partitions). The partition manager (PM) is the management entity
that specifies the number of virtual NEs, partitions, in which the
physical NE should be partitioned. The PM then allocates the
resources of the physical NE to those virtual NEs. Subsequently,
one or more controllers would direct the use of the resources of
that now active partition.
In many cases, the physical realm reflects this logical division of
functionality. For example, MEGACO [RFC3015] and GSMP [GSMPv3] are
examples of protocols that allow control functionality to be
physically separated from switching/forwarding functionality.
Recently, some regulatory environments have mandated multi-provider
access to a single physical infrastructure. To satisfy these
regulations, a common use of partitioning will be for the owner of
the physical NE to partition the NE into several virtual NEs,
partitions, and then to lease these to third parties. In this case,
the PM must be physically separate from all of the controllers.
Since the physical NE must also be remotely configurable, the PM
will also be physically separate from the physical NE. The
following illustration depicts this arrangement. The dashed lines
indicate potential interactions.
------------------ -------------------
| | 3 | |
| Partition |-------------| Controller |
| Manager | | |
------------------ -------------------
\ /
\ /
\ 1 2 /
\ /
-----------------
| |
| Network |
| Element |
-----------------
The interaction labeled "1" is one in which the PM partitions the NE
and allocates resources to the partitions. In order to support
dynamic partitioning, this document will place certain requirements
on proposed (or new) solutions in this space such as [MSF-SPMIB].
The interaction labeled "2" is one by which the controller
configures and manages an active partition of the physical NE.
Anderson et. al. Expires August 2001 [Page 3]
Internet Draft Requirements for Dynamic Partitioning Feb 2001
Proposed solutions in this space include GSMP [GSMPv3] and MEGACO
[RFC3015].
The interaction labeled "3" is one by which a PM and a controller
could communicate to alter the nature of an active partition.
Possible interactions include:
- A controller could request that the resources of one
of its active partitions be altered; either increased
or decreased.
- The PM could respond to a controller request for
altered resource levels.
- The PM could request that a controller release
resources currently allocated to one of its active
partitions. This could involve the following types of
request:
- A request to relinquish allocated but currently unused
resources. That is to put a freeze on additional use
of the specified resources.
- A request to relinquish used resources.
- A request to relinquish an active partition. That is
a request that a controller shut down an active
partition.
- The controllerÆs response to a PM request.
As far as the authors know, no proposed standard solutions currently
exist for type 3 interactions.
3. Dynamic Partitioning
Static repartitioning of a NE can be a costly and inefficient
process. First, before static repartitioning can take place, all
existing connections with controllers must be severed. When this
happens, the NE will typically release all the state configured by
the controller. Then, the virtual NE must be placed in the "down"
state while the repartitioning takes place. Once the repartitioning
is completed, the partitions are placed in the "up" state and the
controllers are allowed to reconnect to the partitions. Then, the
controllers can reestablish state in the active partition. Thus,
static repartitioning results in a period of downtime and a period
in which the controllers are reestablishing state. This is the case
even if resources that are not currently in use in one partition,
either and active or an inactive partition, are intended for a fully
loaded active partition.
Therefore, dynamic partitioning is to be preferred to static
partitioning since it avoids the downtime and loss of state
associated with static partitioning. However, a different set of
potential problems exist for dynamic partitioning. Some questions
to be answered include the following:
- Who initiates the repartitioning on the NE?
- How is the controller notified of an increase or
decrease in resources?
- What should happen when the PM would like to decrease
the resources allocated to a partition but those
Anderson et. al. Expires August 2001 [Page 4]
Internet Draft Requirements for Dynamic Partitioning Feb 2001
resources are in use?
4. Requirements
This document does not attempt to answer the preceding questions but
instead defines a set of requirements that any solution to these
problems MUST satisfy.
1. If a PM instructs a NE to release resources allocated to an
active partition and if any of those resources are currently in
use, the NE MUST deny the PMÆs request.
2. During dynamic repartitioning, a NE MUST maintain all existing
connection state.
3. If a NE denies a repartitioning request due to resources being in
use, the PM MAY contact the controller to ask it to reduce its
resource utilization.
4. If a PM has requested that a controller reduce resource
utilization so that a partition can be downsized and that
controller is not cooperating, the PM MUST be able to "down" the
virtual NE, thereby disconnecting the controller, and then reduce
the partitionÆs resources. In other words, the PM must be able
to resort to static partitioning when a controller is
uncooperative.
5. Control protocols SHOULD NOT include any mechanism by which a NE
can ask its controller to reduce its resource usage.
6. Because controllers cannot be trusted to use only those resources
allocated to their active partitions, the NE MUST reject all such
attempts. Preferably, the control protocol would allow the NE to
do so with an explicit reactive notification although implicit
reactive notifications are also permitted.
7. Control protocols MAY contain proactive resource notification
messages by which a NE could instantaneously inform the
controller of an increase or decrease in resources. When
present, dynamic partitioning solutions MAY make use of proactive
notifications. However, we do not specifically require control
protocols to contain proactive notifications because all control
protocols must already have explicit or implicit reactive
notifications as mentioned in requirement #6.
8. A PM MAY directly inform a controller of a change in virtual NE
resources rather than rely on the implicit resource exhaustion
mechanism of the control protocol.
9. NEs MAY inform the PM of resource exhaustion on a particular
partition.
10. A controller MAY ask the PM for further resources or a reduction
in existing resources.
11. To support the automation of interaction between the PM and
attached controllers, the PM MUST be able to determine from the
NE the addresses of the controllers that are currently attached
to a virtual NE.
5. Security Considerations
Only authorized PMs MUST be allowed to dynamically repartition a NE.
Similarly, only the PM (or an authorized agent of the PM) that is
Anderson et. al. Expires August 2001 [Page 5]
Internet Draft Requirements for Dynamic Partitioning Feb 2001
authorized to partition a NE MUST be allowed to contact controllers
to request that they decrease their resources or inform them that
their resources have been increased. Likewise, the PM MUST verify
and authenticate that any requests for additional/fewer resources
for a virtual NE have come from a controller authorized to control
the specified virtual NE.
6. Intellectual Property Considerations
The IETF is being notified of intellectual property rights claimed
in regard to some or all of the specification contained in this
document. For more information, consult the online list of claimed
rights.
7. References
[GSMPv3] A. Doria, et. al, "Draft-ietf-gsmp-08.txt", work in
progress.
[SPMIB] T. Anderson, et. al, "draft-anderson-partitioning-mib-
00.txt", work in progress, February 2001.
[RFC2119] S. Bradner, "Key words for use in RFCs to Indicate
Requirement Levels", RFC 2119, BCP 14, March 1997.
[RFC2297] P. Newman, et. al., "IpsilonÆs General Switch Management
Protocol Version 2.0," RFC2297, March 1998.
[RFC3015] F. Cuervo, et. al., "Megaco Protocol 1.0," RFC3015,
November 2000.
8. Author Information
Todd A. Anderson
Intel
2111 NE 25th Avenue
Hillsboro, OR 97124 USA
+1 503 712 1760
todd.a.anderson@intel.com
Avri Doria
Nortel Networks
600 Technology Park Drive
Billerica, MA 01821
Phone: +1 401 663 5024
Email: avri@nortelnetworks.com
Anderson et. al. Expires August 2001 [Page 6]