Internet DRAFT - draft-mjsraman-rtgwg-inter-as-psp
draft-mjsraman-rtgwg-inter-as-psp
RTGWG Working Group Shankar Raman
Internet-Draft Balaji Venkat Venkataswami
Intended Status: Experimental RFC Gaurav Raina
Expires: October 30, 2013 IIT Madras
April 28, 2013
Reducing Power Consumption using BGP
draft-mjsraman-rtgwg-inter-as-psp-06
Abstract
In this paper, we propose a framework to reduce the aggregate power
consumption of the Internet using a collaborative approach between
Autonomous Systems (AS). We identify the low-power paths among the AS
and then use Traffic Engineering (TE) techniques to route the packets
along the paths. Such low-power paths can be identified by using the
consumed-power-to-available-bandwidth (PWR) ratio as an additional
constraint in the Constrained Shortest Path First (CSPF) algorithm.
For re-routing the data traffic through these low-power paths, the
Inter-AS Traffic Engineered Label Switched Path (TE-LSP) that spans
multiple AS can be used. Extensions to the Border Gateway Protocol
(BGP) can be used to disseminate the PWR ratio metric among the AS
thereby creating a collaborative approach to reduce the power
consumption. Since calculating the low-power paths can be
computationally intensive, a graph-labeling heuristic is also
proposed. This heuristic reduces the computational complexity but may
provide a sub-optimal low-power path. The feasibility of our
approaches is illustrated by applying our algorithm to a subset of
the Internet. The techniques proposed in this paper for the Inter-AS
power reduction require minimal modifications to the existing
features of the Internet. The proposed techniques can be extended to
other levels of Internet hierarchy, such as Intra-AS paths, through
suitable modifications.
Status of this Memo
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79.
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.
Shankar Raman.et.al Expires October 30, 2013 [Page 1]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
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."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/1id-abstracts.html
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
Copyright and License Notice
Copyright (c) 2013 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1 Low-power routers and switches . . . . . . . . . . . . . . . 4
1.2 Power reduction using routing and traffic engineering . . . 4
1.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . . 5
2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Pre-requisites for the Proposed Method . . . . . . . . . . . 6
2.1.1 Constructing network topology using BGP strands . . . . 6
2.1.2 PWR ratio calculation . . . . . . . . . . . . . . . . . 7
2.1.2.1 Earlier method of computing numerator of PWR
ratio. . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Explicit routing using TE-LSPs . . . . . . . . . . . . . 10
2.2 LOW-POWER PATHS . . . . . . . . . . . . . . . . . . . . . . 11
2.2.0.1 Algorithm 1 ASBR low-power path algorithm . . . . . 12
2.2.0.2 Algorithm 2 PCE low-power path algorithm . . . . . . 12
2.2.1 Illustration . . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Equivalence class with total ordering . . . . . . . . . 13
2.2.3.1 Algorithm 3 PCE low-power path algorithm with
Shankar Raman.et.al Expires October 30, 2013 [Page 2]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
graph labeling . . . . . . . . . . . . . . . . . . . 14
2.3 Implementation notes and Discussion . . . . . . . . . . . . 15
2.4 Applicability within ASes within a single Admin Domain . . . 18
2.4.1 PWR_SESSION . . . . . . . . . . . . . . . . . . . . . . 18
2.4.2 Power profiles of Routers and Switches . . . . . . . . . 19
2.4.2.1 Concave and Convex power curves . . . . . . . . . . 21
2.4.2.3 Need to advertise both available power and
consumed power . . . . . . . . . . . . . . . . . . . 22
2.5 Conclusion and Future Work . . . . . . . . . . . . . . . . . 23
2.6 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 26
3 Security Considerations . . . . . . . . . . . . . . . . . . . . 27
4 IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 27
5 References . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.1 Normative References . . . . . . . . . . . . . . . . . . . 27
5.2 Informative References . . . . . . . . . . . . . . . . . . 27
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 28
Shankar Raman.et.al Expires October 30, 2013 [Page 3]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
1 Introduction
Estimates of power consumption for the Internet predict a 300%
increase, as access speeds increase from 10 Mbps to 100 Mbps [3],
[8]. Access speeds are likely to increase as new video, voice and
gaming devices get added to the Internet. Various approaches have
been proposed to reduce the power consumption of the Internet such as
designing low-power routers and switches, and optimizing the network
topology using traffic engineering methods [2].
1.1 Low-power routers and switches
Low-power router and switch design aim at reducing the power consumed
by hardware architectural components such as transmission link,
lookup tables and memory. In [4] it is shown that the router's link
power consumption can vary by 20 Watts between idle and traffic
scenarios. Hence the authors suggest having more line cards and
running them to capacity: operating the router at full throughput
will lead to less power per bit, and hence larger packet lengths will
consume lower power. The two important components in routers that
have received attention for high power consumption are buffers and
TCAMs. Buffers are built using dynamic RAM (DRAM) or static RAM
(SRAM). SRAMs are limited in size and consume more power, but have
low access times. Guido [1] states that a 40Gb/s line card would
require more than 300 SRAM chips and consume 2:5kW. DRAM access times
prevent them from being used on high speed line cards. Sometimes the
buffering of packets in DRAM is done at the back end, while SRAM is
used at the front end for fast data access. But these schemes cannot
scale with increasing line speeds. Some variants of TCAMs have been
proposed for increasing line speeds and for reduced power consumption
[7].
1.2 Power reduction using routing and traffic engineering
At the Internet level, creating a topology that allows route
adaptation, capacity scaling and power-aware service rate tuning,
will reduce power consumption. In [8] the author has proposed a
technique to traffic engineer the data packets in such a way that the
link capacity between routers is optimized. Links which are not
utilized are moved to the idle state. Power consumption can be
reduced by trading off performance related measures like latency. For
example, power savings while switching from 1 Gbps to 100 Mbps is
approximately 4 W and from 100 Mbps to 10 Mbps around 0:1 Watts.
Hence instead of operating at 1 Gbps the link speed could be reduced
to a lower bandwidth under certain conditions for reduced power
consumption.
Multi layer traffic engineering based methods make use of parameters
Shankar Raman.et.al Expires October 30, 2013 [Page 4]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
such as resource usage, bandwidth, throughput and QoS measures, for
power reduction. In [6] an approach for reducing Intra-AS power
consumption for optical networks that uses Djikstra's shortest path
algorithm is proposed. The input to this method assumes the existence
of a network topology using which an auxiliary graph is constructed.
Power optimization is done on the auxiliary graph and traffic is
routed through the low-power links. However, the algorithm expects
the topology to be available for getting the auxiliary graph. This
topology is easy to obtain for Intra-AS scenario, but not for Inter-
AS cases. In our approach, we propose a collaborative approach by AS
in power reduction. The core of the Internet at the Inter-AS level,
uses the Multi-Protocol Label Switching (MPLS) technology. MPLS label
switched paths that traverse multiple AS carry traffic from a head-
end to a tail-end. The AS use the Border Gateway Protocol (BGP) for
exchanging routing and topology related information. One of the
attributes of BGP namely, AS-PATH-INFO is used to derive the topology
of the Internet at the AS level. The CSPF algorithm is run on this AS
level topology with the consumed-power-to-available-bandwidth (PWR)
ratio as a constraint, to determine the low-power path from the head-
end to the tail-end. The PWR ratio can be exchanged among the
collaborating AS using BGP. Explicit routing can be achieved between
the head-end and the tail-end through the low-power paths connecting
the AS using the Inter-AS Traffic Engineered Label Switched Path (TE-
LSP) that span multiple AS.
Calculation of such low-power paths can be computationally intensive
and hence certain heuristics may be needed to reduce the computation
time. A graph-labeling heuristic is proposed to reduce the
computation time, which may lead to sub-optimal low-power paths. We
illustrate our approaches by applying it to a subset of the Internet
topology. The rest of the paper is organized as follows: In Section
II, we discuss in detail the pre-requisites for the algorithm.
Section III introduces the proposed technique which uses the CSPF
algorithm to calculate the low-power paths. We also show that by
using a graph-labeling technique, we can reduce the computational
complexity of the low-power path algorithm, but may obtain a sub-
optimal low-power path. In Section IV, we discuss the implementation
issues. We present our conclusion and future work in Section V.
1.1 Terminology
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 RFC 2119 [RFC2119].
2. Methodology
Shankar Raman.et.al Expires October 30, 2013 [Page 5]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
<Document text>
2.1 Pre-requisites for the Proposed Method
In this section we discuss the pre-requisites for the implementation
of the proposed scheme.
2.1.1 Constructing network topology using BGP strands
The Inter-AS topology can be modeled as a directed graph G = (V; E;
f) where the vertices (V) are mapped to AS and the edges (E) map the
link that connect the neighboring AS. The direction (f) on the edge,
represents the data flow from the head-end to the tail-end AS. To
obtain the Inter-AS topology, the approach proposed in [5] is used.
In this approach, it is shown that a sub-graph of the Internet
topology, can be obtained by collecting several prefix updates in
BGP. This is illustrated in Figure 1 which shows the different graph
strands of AS that are recorded from the BGP packets. Each vertex in
this graph is assigned a weight according to the consumed-power-to-
available-bandwidth (PWR) ratio of the AS, as seen by an Autonomous
System Border Router (ASBR) that acts as an entry point to the AS.
Figure 2 shows the strands merged together to form the topology sub-
graph. In this figure, the weight of the vertices are mapped to the
ingress edges. A reference AS level topology derived from 100 strands
of AS-PATH-INFO received by an AS in the Internet is presented in
Figure 3 in [9].
0.2 0.05 0.1
(A) ----> (B) ----> (D)
0.1 0.03 0.2
(D) ----> (G) ----> (H)
0.03 0.5 0.3
(G) ----> (E) ----> (X)
0.5 0.5 0.5 0.1
(C) ----> (B) ----> (H) ----> (X)
0.05 0.5 0.3
(B) ----> (E) ----> (X)
Figure 1: Different strands obtained from BGP updates, where vertices
A,B,C,D and G represent the head-end AS. D,H and X form the tail-end
AS. The vertex weights refer to the PWR ratio of the AS, and the
direction of the link shows the next AS hop.
Shankar Raman.et.al Expires October 30, 2013 [Page 6]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
0.5
(C) +----------------+
0.5| / |
| / |
0.05 V/ 0.1 0.03 0.2 V
(A)--->(B)--->(D)--->(G)--->(H)
| | |
| 0.5| | 0.1
| V V
+----------->(E)--->(X)
0.5 0.3
Figure 2:Combining the strands to get the topology of the Internet.
The PWR ratio is mapped to the the ingress link of the ASBR and not
to the AS.
2.1.2 PWR ratio calculation
In the topology sub-graph, each AS is expected to share its PWR
ratio. In order to calculate this ratio we need to calculate the
consumed power in the AS and the maximum bandwidth available with an
ASBR.
In this proposal each AS is expected to share its PWR ratio from as
many ASBRs (Autonomous System Border Routers) that it has.
Intuitively in order to calculate this ratio we need to calculate the
consumed power representative of the AS and the maximum bandwidth
available with an ASBR on its egress links into the AS. The entry
point to the AS is through the ASBRs that advertise the prefixes
reachable through the AS. Hence the numerator of the PWR ratio is
calculated for the AS at each ingress ASBR. We first obtain the
summation of power consumed at the Provider (P) and the Provider Edge
(PE) routers within an AS. The numerator of the PWR ratio is
calculated by summing up the consumed power of all the routers to be
taken into account and then dividing this sum by the number of
routers. A more intuitive approach would be to use a weighted average
method by assigning routers to categories and having appropriate co-
efficients for each of these categories, thus arriving at a weighted
average which is more accurate. One of these alternatives can be used
to arrive at the numerator of the PWR ratio. Yet another alternative
would have been to sum up the total consumed power of all routers in
the AS and represent that as the numerator of the PWR ratio.
This average consumed power is divided by the maximum bandwidth
available at each of the ASBR's egress link. This step is necessary
Shankar Raman.et.al Expires October 30, 2013 [Page 7]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
as the requested bandwidth for any path from the head-end to the
tail-end using the ASBR is limited by the bandwidth available in the
ASBR's egress links. The highest available bandwidth amongst the
egress links of the ASBR is used as the denominator in the PWR ratio
computation. If the entry point to the AS is through a different ASBR
then the PWR ratio assigned to the ingress link of the ASBR might
vary. Hence, an head-end AS might see different PWR ratios for an
intermediate AS, if the intermediate AS has different ASBRs as its
entry point.
The PWR ratio must be computed and disbursed much ahead of time
before the Inter-AS TE-LSP explicit path or route is computed using
the CSPF algorithm. The correctness of this ratio is of importance to
compute the Inter-AS TE-LSP route through the green AS. If the entry
point to the AS is through a different ASBR then the PWR ratio
assigned to the ingress link of the ASBR might vary. Hence, an head-
end AS might see different PWR ratios for an intermediate AS, if the
intermediate AS has different ASBRs as its entry point.
We now illustrate the PWR ratio calculation. Consider an AS X which
is one of the AS in the vicinity of another AS Y . Let this ASBR of X
have 3 egress links into X denoted as E(1), E(2) and E(3), and 2
ingress links labeled I(1) and I(2). We now calculate the PWR ratio
for I(1) and I(2). Assume that the routers in X have average consumed
power of 200K Watts per hour. From figure 4 we can calculate the PWR
ratio for I(1) and I(2) as 200K Watts / (60 * 60 * 1.5 Gigb = 3.7037
* (10 raised to -8) We could scale this to 0.37087 by multiplying
with a base value of 10 raised to the 7th power.
Shankar Raman.et.al Expires October 30, 2013 [Page 8]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
.__________________
(
( E(1)
\ ( +--------->(Core router)
\ +-------+ / 1Gb
+------>| |/ E(2)
200KW / (60*60*1.5Gb) | ASBR |------------>(Core router)
+------>| of |\ 1Gb
/ | AS 100| \
/ +-------+ \ E(3)
( +-------->(Core router)
( 1.5Gb
.__________________
Figure 4:Calculation of PWR ratio by an ASBR associated with an AS.
The I represents ingress links and E represents egress links. 200KW
is the average consumed power in the AS. 1.5Gb is the maximum
available bandwidth of the egress link in an ASBR.
Note that this ratio is actually a mapping function that is defined
for each of the ingress links of the ASBR associated with an AS. For
the head-end AS this mapping function does not exist as there is no
ingress link. The PWR ratio can then be advertised to the other
neighboring AS using the control plane through BGP extensions. BGP
ensures that the information is percolated to other AS beyond the
immediate neighbors. On receipt of these power metrics to the AS at
the far-ends of the Internet, the overall AS level PWR ratio based
Internet topology can be constructed. This view of the Internet is
available with each of the routers without using any other complex
discovery mechanism. Some sample link weights shown in Figure 2 is
obtained by using such a mapping function on the ingress links.
2.1.2.1 Earlier method of computing numerator of PWR ratio.
Earlier in the previous versions of this document in order to
calculate this PWR ratio we needed to calculate the available power
and the maximum bandwidth available with an ASBR. The entry point to
the AS is through ASBRs that advertise the prefixes reachable through
the AS. Hence, the numerator of the PWR ratio is calculated for the
AS at each ingress ASBR. We first obtained the summation of power
consumed at the major Provider (P) and Provider Edge (PE) routers
within an AS. The average available power is obtained by subtracting
the consumed power from the maximum power rating and summing the
values for all the routers and then dividing the result by the number
of routers. As an alternative, one could use a weighted average for
more accuracy depending on the category of the router advertising the
consumed power. Yet another alternative is to take the average or sum
of the maximum power rating of all the routers within an AS without
Shankar Raman.et.al Expires October 30, 2013 [Page 9]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
taking into account the consumed power. One of these alternatives was
chosen to calculate the numerator of the PWR ratio.
Intuition however drives us towards consumed power as a better
numerator since the lesser the power consumed the lesser the
numerator and hence lesser the ratio if enough bandwidth is available
at the ingress ASBR. The amount of consumed power per bit of
information ought to be low for the shortest path to work out
properly. One more aspect is that lesser the power consumed per
available bit of bandwidth it could be a sign that routers are more
optimal in their power consumption as they take on more traffic. This
is a very crucial point to be considered.
However additional research seems to indicate that both Available and
Consumed Power for a router be advertised. The need that arises for
such a proposition is that there exist power profiles of routers
which is dealt in later sections (section 2.4.2). Please refer
section 2.4.2.1 onwards for more analysis and research on this
subject.
2.1.3 Explicit routing using TE-LSPs
We assume that the head-end and the tail-end may reside in different
AS and the path is along multiple intervening AS. The way to generate
this path is by using Traffic Engineered Label Switched Paths (TE-
LSPs). TE-LSPs can influence the exact path (at the AS level) that
the traffic will pass through. This path can then be realized by
providing these set of low-power consuming AS to a protocol like
Resource Reservation Protocol (RSVP). RSVP-TE then creates TE-LSPs or
tunnels, using its label assigning procedure. The routers use these
low-power paths created by the explicit routing method rather than
using the conventional shortest path to the destination. By this way,
we can influence exclusion of a number of high power AS on the way
from the head-end to the tail-end AS. For example, the dotted line in
Figure 5 represents the explicit route that is chosen by making use
of such TE-LSPs from head-end AS A to the tail-end AS X. Note that if
number of hop was the metric used by CSPF, then the route chosen is
the path with 3 hops.
Shankar Raman.et.al Expires October 30, 2013 [Page 10]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
0.5
(C) +----------------+
0.5| / |
| / |
0.05 V/ 0.1 0.03 0.2 V
(A)...>(B)...>(D)...>(G)...>(H)
| | .
| 0.5| . 0.1
| V V
+----------->(E)--->(X)
0.5 0.3
Figure 5:Low-power path is represented by the dotted lines. This low-
power path has a longer number of hops than the conventional shortest
path.
2.2 LOW-POWER PATHS
In this section we present the low-power path algorithm. The
algorithm consists of two sub-algorithms: the first algorithm is
executed by all the ASBRs in the network and the second by all the
Path Computation Elements (PCEs) in their respective AS. The
algorithms for the ASBRs and PCEs are given as Algorithm 1, 2 and 3.
Shankar Raman.et.al Expires October 30, 2013 [Page 11]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
2.2.0.1 Algorithm 1 ASBR low-power path algorithm
Require: Weighted Topology Graph T=(AS, E, f)
1: Begin
2: if ROUTER == ASBR then
3: /* As part of IGP-TE */
4: Trigger exchange of available bandwidth on bandwidth change,
to the AS internal neighbors;
5: BEGIN PARALLEL PROCESS 1
6: while PWR ratio changes do
7: Assign the PWR ratio to the Ingress links;
8: Exchange the PWR ratio with its external neighbors;
9: Exchange the PWR ratio with AS's (internal) ASBRs;
10: end while
11: END PARALLEL PROCESS 1
,br 12: BEGIN PARALLEL PROCESS 2
13: while RSVP packets arrive do
14: Send and Receive TE-LSP reservations in the explicit path;
15: Update routing table with labels for TE-LSP;
16: end while
17: END PARALLEL PROCESS 2
18: end if
19: End
2.2.0.2 Algorithm 2 PCE low-power path algorithm
Require: Weighted Topology Graph T=(AS, E, f)
Require: Source and Destination for Inter-AS TE LSP with sufficient
bandwidth
1: Begin
2: if ROUTER == PCE then
3: Calculate the shortest paths from the head-end to the
tail-end using CSPF with PWR ratio as the metric;
4: if no path available then
5: Signal error;
6: end if
7: if path exists then
8: Send explicit path to head-end to construct path;
9: end if
10: Continue passively listening to BGP updates to update
T=(AS, E, f);
11: end if
12: End
2.2.1 Illustration
We now illustrate the proposed technique with a simple example.
Shankar Raman.et.al Expires October 30, 2013 [Page 12]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
Consider the AS level topology sub-graph shown in Figure 5
constructed using the strands shown in Figure 1. The PWR ratio
calculated at an ASBR which represents the metric for the AS is
assigned to the ingress link. For example, AS H has two edges coming
into it: one from B and the other from G. Note that the power metrics
for the two strands are different as G to H is lower than that of B
to H. This means that the lower power metric into H is better if the
path from G to H is chosen rather than the one from B to H. This is
illustrated in the Figure 5 using dotted lines. To construct a path
with A as the head-end AS and X as the tail-end AS, from the AS level
topology we see that the path A, B, H, X and A, B, E, X have the
shortest number of hops. However by using CSPF with the PWR ratio
metric as the constraint, we see that the path A, B, D, G, H, X is
power efficient. The routing choice will however be based on the
reservation of the bandwidth on this path. Given that available
bandwidth exists to setup a TE-LSP, the explicit path A, B, D, G, H,
X is chosen. The Resource Reservation Protocol (RSVP) adheres to its
usual operation and tries to setup a path. If bandwidth is not
available in the low-power path thus calculated, then we may fall
back to other paths like A, B, H, X or A, B, E, X provided there is
available bandwidth in these paths. The low-power path algorithm
given as Algorithm 2 is executed by the PCE. Algorithm 1 prepares the
topology and feeds it as input to the PCE as a weighted topology
graph. Using the CSPF algorithm to calculate a route from a source to
destination could be time consuming for a large networks. But the
topology is dynamically updated and hence the computation of the
shortest paths can be triggered based on need. We now give a
heuristic method based on graph-labeling that reduces the computation
time but could trade-off the optimal low-power path.
2.2.3 Equivalence class with total ordering
The heuristic is based on avoiding high PWR ratios. The approach
partitions the weighted links into equivalence classes based on a
range of PWR values. For each partition a labeling is applied such
that each link in the partition has the same label. A total ordering
relationship is then defined on the equivalence class. The heuristic
then starts including partitions with minimum label value iteratively
until we get a connected component, which includes the head-end and
tail-end AS. We apply the CSPF algorithm with the weights as label
values on this sub-graph to obtain the low-power path. The modified
algorithm which uses this scheme is given in Algorithm 3. It should
be noted that this algorithm could provide sub-optimal power paths as
the intermediate steps carry incomplete Internet topology
information.
Shankar Raman.et.al Expires October 30, 2013 [Page 13]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
2.2.3.1 Algorithm 3 PCE low-power path algorithm with graph labeling
Require: Weighted Topology Graph T=(AS, E, f)
Require: Source and Destination for Inter-AS TE LSP with
sufficient bandwidth
1: Begin
2: if ROUTER == PCE then
3: Group the links into N partitions with a label for
each partition depending on the PWR ratio
4: Sort the labels in ascending order.
5: repeat
6: Include the links that have the least label value;
7: Remove the partition with this label;
8: until there is a path from the head-end to tail-end AS
9: Calculate the low-power path using labels from the
head-end to the tail-end using CSPF ;
10: if no path available then
11: Signal error;
12: end if
13: if path exists then
14: Send explicit path to head-end to construct path;
15: end if
16: Continue passively listening to BGP updates to
update T=(AS, E);
17: end if
18: End
Shankar Raman.et.al Expires October 30, 2013 [Page 14]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
+------(1245)-----+
/ G / \G
/ / Y \
V / Y \
(1339) + (6785)<------(1377)----+
| | | / |
| G | G | + G V R
| | V V (2485)
V | (1006) (3426)--+ |
(34234) | Y | G | | | R
| | V V | V
| G | (11229) (4563) | (5677)
V | Y | | V | R
(23411) | V Y +-->(7786)<-+
| +-->(1485)<--------/ |
| Y | | R
| V V
| (15467) (8298)
| G G | | R
| | |
| V V
+--------->(16578)<-------(9732)
Figure 6:Application of the graph-labeling heuristic. We consider 3
labels "G" < "Y" < "R". Using algorithm 3 the "G" path from the head-
end AS 1245 to the tail-end AS 16578 is chosen in the first
iteration.
2.2.4 Illustration of graph labeling
We briefly illustrate the graph-labeling algorithm using Figure 6. In
this diagram we have categorized the links into three partitions
based on the PWR ratio. PWR ratio less than 0:1 are labeled as G,
between 0:1 to 0:3 are labeled as Y and the rest as R. The total
ordering is defined as G < Y < R, where the G links have low PWR
ratios than the Y links. The path could be established through the AS
that have G as the ingress link; the path being 1245, 1339, 34234,
23411 and 16578.
2.3 Implementation notes and Discussion
In this section we present some notes on feasibility of
implementation of our scheme in a live network. First, the requested
bandwidth should be available on the low-power path, but the CSPF
algorithm is run with multiple constraints, one of which is the
bandwidth requirement for the flows to be transported through the TE-
LSP. The PWR ratio can then be applied to the available paths thus
computing the low-power paths. Second, as we are using traffic
Shankar Raman.et.al Expires October 30, 2013 [Page 15]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
engineering with link state routing protocols, there is a reliable
flooding process that are triggered when updates about the change in
characteristic arise. We propose addition of some attributes with no
change to the protocol implementation. There may be a time lag when
the far ends of the Internet receive the attribute and the time it
originated. This however cannot be avoided as with other attributes
and metrics.
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
+-------------------------------------------------------------+
| Owning 32 bit Autonomous System Number |
+-------------------------------------------------------------+
| Other 32 bit Autonomous System Number |
+-------------------------------------------------------------+
| PWR Ratio for the AS (Consumed PWR) |
+-------------------------------------------------------------+
| PWR Ratio for the AS (Available PWR) |
+-------------------------------------------------------------+
| Advertising ASBR's IP router ID |
+-------------------------------------------------------------+
| Peer ASBR's IP router ID |
+-------------------------------------------------------------+
| 64 bit sequence number for restarts, aging |
| and comparison of current PWR Ratio. |
+-------------------------------------------------------------+
Figure 7: Proposed PDU format with an added attribute for AS-PATH-
POWER-METRIC
The additions to the above Attribute have been added to optimize and
correctly correlate the connecting ASes and the inter-AS links among
them. For the traffic direction into the Advertising AS the above
information will be easier to correlate than the previous version
which did not advertise the peer AS which had the ingress links into
the advertising Router.
In MPLS-TE when the TE metrics are modified, there is a reliable
flooding process within an Interior Gateway Protocol (IGP). Such
triggered updates apply to the PWR ratio as well. The proposed PWR
ratio is advertised to the neighboring AS and the information
percolated to all the AS, in a AS-PATH-POWER-METRIC attribute. This
attribute can be implemented as shown in Figure 7. The frequency of
the updates for this attribute should be fixed to avoid network
flooding.
The AS-PATH-POWER-METRIC for each ASBR is calculated, and advertised
as the PWR ratio for the AS. This AS-PATH-POWER-METRIC is filled into
the appropriate optional transitive non-discretionary attribute and
Shankar Raman.et.al Expires October 30, 2013 [Page 16]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
inserted into a unique vector for a set of prefixes advertised from
the AS. Such advertised prefixes may have originated from the AS or
be the transit prefixes. The filled vector is sent to the ASBR of the
neighboring AS and the information propagates to all the ASBRs. If
the elements denoting AS in a vector of AS-PATH-INFO is not the same
as the ones that need to be advertised in a AS-PATH-POWER-METRIC,
then a suitable subset of AS-PATH-POWER-METRIC is identified and sent
in the BGP updates. A vector of size 1 also can be employed if the AS
in question is the only one for which PWR ratio has changed in the
originating AS. The collation can be done depending on availability
of such metrics and their mapping to a valid AS-PATH-INFO metric.
The power consumed by each router may fluctuate over short time
intervals. In order to dampen these fluctuations which can cause
unnecessary updates, power can be measured when falling within
intervals of suitable size (say a range of values). This is as
opposed to measuring power as a discrete quantity. This method of
power measurement reduces the frequency of triggered updates from the
routers due to power change.
0.1 0.2 0.1
(A) ---> (B) ---> (D)
0.1 0.2 0.02 0.2
(A) ---> (C) ---> (E) ---> (D)
0.1 0.2
(D) ---> (X)
Figure 8: Example of strands where more than one PWR ratio is
advertised by "D"
0.2 0.1 0.2
(A).....>(B).....>(D).....>(X)
| ^
|0.2 0.02 | 0.2
+--->(C)-------->(E)
Figure 9:Choice of low-power path derived using the algorithm which
uses lower value of the ingress link but through the same AS
A use case of multiple ASBRs advertising differing PWR ratio shows
that an AS may be seen as green through one ingress link and not
through the other. Consider the case of multiple ASBRs that belong to
the same AS, advertising PWR ratios that differ. This could lead to
power values that belong to different classes of ratios with many
intervening classes in between. These advertised PWR ratios could
Shankar Raman.et.al Expires October 30, 2013 [Page 17]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
lead to one ASBR being preferred over the other thus taking a
different path from head-end to tail-end. This also entails that
there may be multiple paths to the AS through these different ASBRs.
Consider Figure 8 which shows a set of strands that derive a topology
as in Figure 9. Here D is reachable via two paths but the PWR ratios
differ. This illustrates the case where the better metric wins out.
The average power consumed would not have an effect but the bandwidth
available on these ASBR egress links would definitely influence the
path.
2.4 Applicability within ASes within a single Admin Domain
As per [draft-ietf-idr-aigp] there are deployments in which a single
administration runs a network which has been sub-divided into
multiple, contiguous ASes, each running BGP. There are several
reasons why a single administrative domain may be broken into several
ASes (which, in this case, are not really "autonomous".) It may be
that the existing IGPs do not scale well in the particular
environment; it may be that a more generalized topology is desired
than could be obtained by use of a single IGP domain; it may be that
a more finely grained routing policy is desired than can be supported
by an IGP. In such deployments, it can be useful to allow BGP to
make its routing decisions based on the IGP metric, so that BGP
chooses the "shortest" path between two nodes, even if the nodes are
in two different ASes within that same administrative domain. The
authors refer to the set of ASes in a common administrative domain as
an "AIGP Administrative Domain".
A combination of the AIGP administrative metric and the graph
heuristic algorithm could be combined to arrive at a set of a
suitable number k power-shortest paths and then use a tie-break
amongst such k power-shortest-paths with the least AIGP metric. This
is provided the set of ASes where the decision is being made all fall
under a AIGP Administrative domain. This provides a trade-off of
power shortest paths and least number of hops (link wise) to get from
source to destination across these ASes.
2.4.1 PWR_SESSION
An implementation that supports the PWR attribute CAN support a per-
session configuration item, PWR_SESSION, that indicates whether the
PWR attribute is enabled or disabled for use on that session.
- The default value of PWR_SESSION, for EBGP sessions, between
providers (distinct operators) CAN be "disabled".
- The default value of PWR_SESSION, for IBGP and confederation-
Shankar Raman.et.al Expires October 30, 2013 [Page 18]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
EBGP sessions, MUST be "enabled."
The PWR attribute MUST NOT be sent on any BGP session for which
PWR_SESSION is disabled.
If an PWR attribute is received on a BGP session for which
PWR_SESSION is disabled, the attribute MUST be treated exactly as if
it were an unrecognized transitive attribute. That is, " The
handling of an unrecognized optional attribute is determined by the
setting of the Transitive bit in the attribute flags octet. Paths
with unrecognized transitive optional attributes SHOULD be accepted.
If a path with an unrecognized transitive optional attribute is
accepted and passed to other BGP peers, then the unrecognized
transitive optional attribute of that path MUST be passed, along with
the path, to other BGP peers with the Partial bit in the Attribute
Flags octet set to 1. If a path with a recognized, transitive
optional attribute is accepted and passed along to other BGP peers
and the Partial bit in the Attribute Flags octet is set to 1 by some
previous AS, it MUST NOT be set back to 0 by the current AS".
This helps in confining the distribution of the attribute and use in
calculation of the power shortest paths only amongst ASes that have
trust relationships with other ASes. Of course, this includes and
promotes the use of PWR attribute within a AIGP administrative
domain.
2.4.2 Power profiles of Routers and Switches
It has been experimented and from several sources found that there
exist routers which have different power profiles. The power profile
of a router is the curve of power consumption to available bandwidth.
Mentioned below are a few of these prominent ones that have to be
taken into consideration.
The first profile that we will consider is the flattening curve. The
power consumed to available bandwidth curve takes the shape of a
steep one initially and then tapers off to a plateau. The point at
which it begins to give a delta-C (delta in Power Consumed) to delta-
B (Available Bandwidth exhausted) is the inflection point that tapers
off to a plateau. Here the delta-C/delta-B begins to slow down or
decrease rapidly. The more the traffic that is added onto the device
the lesser it draws power.
Shankar Raman.et.al Expires October 30, 2013 [Page 19]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
^
|
P | .
o | .
w | .
e | .
r | .
| .
c | .
o | .
n | .
s | .
u.| .
------------------------------------>
| Available Bandwidth exhausted
The second profile that we will consider is the exponential curve.
The power consumed to available bandwidth curve takes the shape of an
ever increasing steep curve as shown below. Here the delta-C/delta-B
begins to increase as more traffic is thrown onto it as the Available
bandwidth exhausted increases. This power curve beyond a point is
intolerable with respect to power guzzling.
^
|
P | .
o | .
w | .
e | .
r | .
| .
c | .
o | .
n | .
s | .
u.| .
------------------------------------>
| Available Bandwidth exhausted
The third profile that we will consider is a linear curve. In other
words just a straight line. Here delta-C/delta-B is a constant.
Shankar Raman.et.al Expires October 30, 2013 [Page 20]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
^
|
P | .
o | .
w | .
e | .
r | .
| .
c | .
o | .
n | .
s | .
u.| .
------------------------------------>
| Available Bandwidth exhausted
2.4.2.1 Concave and Convex power curves
Given that there are 3 kinds of major profiles in the router power
consumption, what line would we like to pick. This is an important
point when choosing the metric to pick the low power paths.
(a) If the confrontation is between 2 first profile routers the lower
of the 2 would be considered as shown below. The lower curve offers
better power savings for each GB of bandwidth transported.
^
|
P | .
o | .
w | . .
e | . .
r | . .
| . .
c | . .
o | . .
n | . .
s | . .
u.| .
------------------------------------>
| Available Bandwidth exhausted
(b) If the confrontation is between 2 second profile routers the
upper curve offers more power savings per GB of bandwidth.
Shankar Raman.et.al Expires October 30, 2013 [Page 21]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
^
|
P | . .
o | . .
w | . .
e | . .
r | . .
| . .
c | . .
o | . .
n | . .
s | .
u.| .
------------------------------------>
| Available Bandwidth exhausted
(c) When the confrontation is between a first profile curve and a
second profile curve, it would be optimal to pick (as shown below)
the lower of the curves because it gives us lesser power consumed for
every GB of traffic routed / switched. Here the exponential curve is
the one that offers lesser amount of power consumed per GB of traffic
is chosen. But when it gets to a point that the two curves intersect
it would be more optimal to pick the tapering curve. Thus at the
meeting point of the 2 curves the exponential curve becomes more
costly and the tapering one gives us more GB for the power buck. Thus
this switchover from one curve to the other (in other words from the
exponential curve to the tapering one) does the trick in terms of
finding an optimal solution.
^ .
| .
P | . .
o | (*)
w | . .
e | . .
r | . .
| . .
c | . .
o | . .
n | . .
s | . .
u.| ..
------------------------------------>
| Available Bandwidth exhausted
(*) Metric switchover point from Consumed Power to Available Power.
2.4.2.3 Need to advertise both available power and consumed power
Shankar Raman.et.al Expires October 30, 2013 [Page 22]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
Thus the above sections have shown that both the available power and
the consumed power MUST be advertised so that case (c) can be
deciphered and the switchover of the curves be done and the
appropriate router be chosen for the rest of the bandwidth to be
switched over to.
Thus there will exist Consumed-Power to Available Bandwidth ratio and
the Available Power to Available Bandwidth ratio. Both the ratios are
computed and the lower value chosen. The Available Power can be
judged from the calibration process such as the one carried out by
independent test organizations as in [12]. An example of their
calibration is referred to in [12].
2.5 Conclusion and Future Work
In this paper, we proposed a scheme for reducing the power
consumption of the Internet using collaborative effort between AS.
The topology of the Internet is represented using a graph model and
derived using the strands obtained from the AS-PATH attribute of the
BGP updates. CSPF algorithm is run on this topology by using the PWR
ratio as a constraint. The PWR ratio is advertised through the
ingress links of the ASBRs associated with AS using BGP updates. The
CSPF algorithm finds out the low-power consuming AS that can route
data packets from a head-end to a tail-end. Explicit routing is
handled through the use of TE-LSPs. This entails adopting routes by
choosing entry points to an AS that give energy saving paths. Since
using CSPF can be time consuming a heuristic algorithm to derive the
low-power paths using graph-labeling was proposed. Our work
complements the current schemes for reducing power consumption within
a router such as switching off or bringing to power-idle-state
certain select components within the forwarding and lookup
mechanisms.
This Power shortest Path calculation can be taken care of a Path
Computation Element (PCE) unit that could be either be a process
running on a linecard on a ASBR, or even a core router or an offline
engine that is passively listening to the BGP updates within the AS
without spitting out any routes of its own. The PCE architecture has
already been proposed in the ietf and even has a separate working
group for itself.
These offline or linecard engines are currently being sold in the
market by the networking majors and other companies that develop
hardware and software for the PCE. All the PCE needs to do is to
accept configuration and passively listen to BGP updates from various
peers or even be a client for a route reflector, thus
Shankar Raman.et.al Expires October 30, 2013 [Page 23]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
a) Accepting these BGP updates
b) Extracting the AS PATH information from these updates
c) Then constructing the inter-AS topology
d) Apply the PWR metric that comes along in these BGP updates to the
edges of the graph
e) Then compute the power shortest path as required by the
configuration.
Normally the ASes have SLA agreements between each other to carry X
amount of traffic from say a provider A. If the AS representing the
ISP then advertises fake figures to carry more traffic than is
mandated by the SLA agreement with other providers, then it is to
that ISPs detriment since by advertising a better PWR ratio it
invites more traffic through it thus getting paid less and carrying
more traffic. This is not in the best interest of the ISP. This is so
because in the final analysis the Power Shortest Path computed would
include it regardless of the amount of traffic to be carried thus
causing it to invite more traffic through it than it has accepted,
even much more than its capacity. Hence it would be advisable for
that ISP to advertise proper PWR ratios and NOT on the lower side of
the spectrum. If it advertises HIGHER PWR ratios it would not be
chosen, and hence that could be a policy measure NOT to accept any
traffic at all since its capacity may be filled up with existing
traffic. So advertising on the LOWER side would lead to lesser amount
of benefit with respect to dollar per bit transported, and on the
HIGHER side would be to exclude it from carrying any traffic that
wanted to use the Power Shortest Path.
We also propose that there be a governing body in the IETF or outside
it or sponsored by the IETF to verify the power ratios advertised are
indeed valid or approximately closer to the actual consumption. A
link up for each ISP with a power application level gateway to ensure
proper ratios are advertised could be mandated amongst at least the
co-operating ISPs (ASes).
The points on which this proposal by us innovates is as follows.
a) There has been no effort prior to this to build an inter-AS
topology with a weighted graph based on a PWR ratio. On this point it
breaks a new path that would lead to inter-AS co-operation that
contributes to power reduction overall in the internet. The paper
suggested for OSPF by [10] deals with intra-Autonomous-system
scenario rather than an inter-AS one. It is also to be noted that the
Shankar Raman.et.al Expires October 30, 2013 [Page 24]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
IGP such as OSPF / IS-IS or any other link-state protocol for that
matter is expected to capture the energy consumption of each router
within the Autonomous system as in paper [10] to help get a hold on
the overall average within the AS, or even sum up the total of all
the power consumption within the AS with such intra-AS IGP LSA. This
contributes to the PWR ratio proposed in our idea. Thus the intra-AS
metric contributes to the PWR ratio. [10] proposal deals with
primarily paths setup within an AS and not inter-AS paths. Thus the
fundamental problem it solves is different while the problem we solve
relates to the inter-AS paths which run across ASes from a head-end
AS to a tail-end one.
b) The other aspect of innovation is to use BGP as the piggyback
protocol upon which this scheme stands. There has been no effort
earlier to approach the internet power reduction problem with BGP as
the mode of transport of the energy ratios and coupling it with the
inter-AS topology built with AS-PATH-INFO information.
The above 2 are key aspects of innovation.
When links and switches are gated or put into low-power state within
an AS, the power-consumption automatically drops at the aggregate
level, as a result of which the PWR ratio would be a lower figure
advertised through BGP and thus this AS would attract more Power
Shortest Path traffic through it. Thus the links within the AS and
the switches within it would function more optimally if it had more
traffic that went along paths that were originally put in low-power
state thus utilizing the paths more effectively, when attracting PSP
traffic.
There exist MIBs today that have object identifier for power consumed
in a router. Maybe all the related components within it may NOT be
listed with regards to power consumed. But the overall power consumed
by the Router / Switch is gettable. Once it is advertised in a opaque
Link-State-Advertisement say in the form of a TLV and the LSAs are
flooded through the network in an AS, all routers get a uniform
picture of which router consumes what power. This method already
exists for Traffic engineering Database LSAs that are advertised as
LSAs for the purpose of traffic engineering within an AS. We are
merely piggybacking on this capability to calculate the PWR ratio at
the ASBR which amongst others is yet another Router / Switch of the
AS.
Our future work includes looking into computing low-power paths
within AS as well. Further it can be noted that the proposed
algorithms might lead to increased latency as the number of hops
increase, which could be critical for time sensitive applications.
Since the PWR ratio could vary dynamically with traffic, the impact
Shankar Raman.et.al Expires October 30, 2013 [Page 25]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
of traffic on the algorithm would also be of interest.
2.6 Acknowledgements
Shankar Raman would like to acknowledge the support by BT Public
Limited (UK) under the BT IITM PhD Fellowship award. Balaji Venkat
and Gaurav Raina would like to acknowledge the UK EPSRC Digital
Economy Programme and the Government of India Department of Science
and Technology (DST) for funding given to the IU-ATC. We would like
to acknowledge that a version of this paper has been accepted in
IARIA conference ENERGY 2012.
Shankar Raman.et.al Expires October 30, 2013 [Page 26]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
3 Security Considerations
No specific security considerations apart from the usual
considerations with respect to authenticating BGP messages / updates
from BGP neighbors is necessary for this scheme.
4 IANA Considerations
A new optional transitive non-discretionary attribute needs to be
provided by IANA for carrying the PWR ratio across the Internet in
the specified format in BGP.
5 References
5.1 Normative References
5.2 Informative References
REFERENCES
[1] G. Appenzeller, Sizing router buffers, Doctoral
Thesis, Department of Electrical Engineering, Stanford
University, 2005.
[2] A. P. Bianzino, C. Chaudet, D. Rossi and J. L.
Rougier, A survey of green networking research, IEEE
Communications and Surveys Tutorials, preprint.
[3] J. Baliga, K. Hinton and R. S. Tucker, Energy
consumption of the internet, Proc. of joint international
conference on optical internet, June 2007, pp. 1-3.
[4] J. Chabarek, J. Sommers, P. Barford, C. Estan, D.
Tsiang and S. Wright, Power awareness in network design
and routing, Proc. of the IEEE INFOCOM 2008, April 2008,
pp. 457-465.
[5] B. Venkat et.al, Constructing disjoint and partially
disjoint InterAS TE-LSPs, USPTO Patent 7751318, Cisco
Systems, 2010.
[6] M. Xia et. al., Greening the optical backbone network:
A traffic engineering approach, IEEE ICC Proceedings, May
2010, pp. 1-5.
Shankar Raman.et.al Expires October 30, 2013 [Page 27]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
[7] W. Lu and S. Sahni, Low-power TCAMs for very large
forwarding tables, IEEE/ACM Transactions on Computer
Networks, June 2010, vol. 18, no. 3, pp. 948-959.
[8] B. Zhang, Routing Area Open Meeting, Proceedings of
the IETF 81, Quebec, Canada, July 2011.
[9] M.J.S Raman, V.Balaji Venkat, G.Raina, Reducing Power
consumption using the Border Gateway Protocol, IARIA
conferences ENERGY 2012.
[10] A.Cianfrani et al., An OSPF enhancement for energy
saving in IP Networks, IEEE INFOCOM 2011 Workshop on Green
Communications and Networking
[draft-ietf-idr-aigp] P. Mohapatra et.al, The Accumulated
IGP metric attribute for BGP,
https://datatracker.ietf.org/doc/draft-ietf-idr-aigp/,
November 2012.
Authors' Addresses
Shankar Raman
Department of Computer Science and Engineering
IIT Madras,
Chennai - 600036
TamilNadu,
India.
EMail: mjsraman@cse.iitm.ac.in
Balaji Venkat Venkataswami
Department of Electrical Engineering,
IIT Madras,
Chennai - 600036,
TamilNadu,
India.
EMail: balajivenkat299@gmail.com
Shankar Raman.et.al Expires October 30, 2013 [Page 28]
INTERNET DRAFT Reducing power consumption using BGP April 28, 2013
Prof.Gaurav Raina
Department of Electrical Engineering,
IIT Madras,
Chennai - 600036,
TamilNadu,
India.
EMail: gaurav@ee.iitm.ac.in
Shankar Raman.et.al Expires October 30, 2013 [Page 29]
