Internet DRAFT - draft-pskim-grasping-network-situation
draft-pskim-grasping-network-situation
Network Working Group P. Kim
Internet-Draft Korea Polytechnic University
Intended status: Experimental
Expires: May 3, 2020
November 4, 2019
Grasping Network Situation for Improving End-to-End Throughput
draft-pskim-grasping-network-situation-00.txt
Abstract
In this draft, a mechanism to grasp the network situation is proposed
for improving end-to-end path throughput. The proposed mechanism is
based on the active packet-train probing based estimation. The
proposed mechanism defines three cases of the difference between the
average output gap and the input gap, and then reflects fully them.
Since three cases are handled respectively by appropriate
corresponding manners, the proposed mechanism can be expected to
reduce the detection error for the turning point. Therefore, through
the proposed mechanism, the available bandwidth can be estimated more
reliably.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on May 3, 2020.
Copyright Notice
Copyright (c) 2019 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
(https://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
Kim Expires May 3, 2020 [Page 1]
Internet-Draft Grasping Network Situation November 2019
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 . . . . . . . . . . . . . . . . . . . . . . . . 2
2. A Grasping Network Situation via Bandwidth Estimation . . . . 3
2.1 Existing Active Packet-train Probing Based Estimation . . . . 3
2.2 An Alternative Active Packet-train Probing Based Estimation . 4
3. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 6
4. References . . . . . . . . . . . . . . . . . . . . . . . . . 6
Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 7
1. Introduction
Traffic jams on narrow roads can be one of the main causes of traffic
congestion, which also applies to communication networks. If there
are more data traffic than the available network bandwidth,
communication latency appears. This can adversely affect 5G-based
Internet services such as self-driving cars, autonomous robots, etc.
Communication latency is the term used to indicate any kind of delay
that happens in data communication over a network. In particular,
high latency creates bottlenecks in any network communication. It
prevents the data from taking full advantage of the network pipe and
effectively decreases the communication bandwidth. The impact of
latency on network bandwidth can be temporary or persistent based on
the source of the delays.
Recently, in order to reduce the communication latency, grasping the
network situation and adjusting the data transmission amount have
been researched as shown in BBR(Congestion-based congestion control)
[1] and ExLL(An ultra low-latency congestion protocol for mobile
cellular network)[2]. BBR has been designed to prevent bottlenecks
before they happen. For a given network connection, BBR uses recent
measurements of the network's transmission rate and round-trip time
to build an explicit model that includes both the maximum recent
bandwidth available to that connection, and its minimum recent
round-trip delay. ExLL has been designed to elaborate the allowed
network bandwidth for an efficient low-latency transmission protocol.
If data is sent only as much as the network bandwidth allowed by the
mobile communication terminal, the data will not be unnecessarily
accumulated. To do this, the pattern of packets received by the
mobile communication terminal is observed.
As shown in above observations, understanding the dynamic properties
of the end-to-end Internet performance metrics such as available
network bandwidth is beneficial for the proper resource management in
Kim Expires May 3, 2020 [Page 2]
Internet-Draft Grasping Network Situation November 2019
emerging wireless Internet services that required low-latency data
transmission. Therefore, the area of end-to-end available bandwidth
estimation has attracted considerable interest. As a result, several
mechanisms for the available bandwidth estimation have been developed
based on active measurements[3]. Among existing mechanisms for
available bandwidth estimation, the active packet-train probing
mechanism such as initial gap increasing and packet transmission rate
was used successfully. The ultimate objective is to experimentally
determine the input gap value at some point for which the average
output gap is equal to the input gap. At this point, the probing
packets are considered to interleave nicely with the competing
traffic, and the average rate of the packet train equals the
available bandwidth on the bottleneck link. This point is called the
"turning point". At the turning point, the input gap value for which
the average output gap is equal to the input gap is the right value
to use for estimating the available bandwidth. However, there are
some issues in the existing active packet-train probing mechanism.
After performing a measurement, three cases can be defined according
to the difference between the average output gap and the input gap.
These three cases have respectively different relationship between
the average rate of the probing packet train and the available
bandwidth. However, the existing mechanism did not reflect fully
these three cases in order to reduce the detection latency of the
turning point. That is, two of three cases are handled in the same
way, which can introduce the detection error for the turning point
since these two cases handled in the same way are absolutely
different. Thus, the available bandwidth can be estimated
inaccurately although the measurement latency can be reduced.
Therefore, to reduce the detection error of the turning point and
enhance the accuracy of the available bandwidth estimation, a new
mechanism is proposed based on the active packet-train probing
mechanism. The proposed mechanism reflects fully three cases, while
the existing mechanism reflected only two cases. Since three cases
are handled respectively by appropriate corresponding manners, the
proposed mechanism can be expected to reduce the detection error for
the turning point. Therefore, the end-to-end available bandwidth can
be estimated more reliably.
2. A Grasping Network Situation via Bandwidth Estimation
2.1 Existing Active Packet-train Probing Based Estimation
The active packet-train probing mechanism was proposed for the
available bandwidth estimation and shown to be much faster than
existing mechanisms with similar measurement accuracy but with
shorter measurement latency. This mechanism is based on a single-hop
gap model that captures the relationship between the competing
traffic and the probing packet train. As a sequence of probing packet
trains from the source travel through the network, packets belonging
to the competing traffic may be inserted between them, thus
Kim Expires May 3, 2020 [Page 3]
Internet-Draft Grasping Network Situation November 2019
increasing the gap at the destination. As a result, the average
output gap value at the destination may be a function of the
competing traffic rate, making it possible to estimate the amount of
competing traffic. That is, the average output gap can be used to
determine the competing traffic bandwidth and hence the available
bandwidth on the end-to-end path assuming that the bottleneck link
bandwidth along the end-to-end path is known. At some point, the
average output gap equals the input gap as gaps in a probing packet
train increase. This point is called the "turning point". At the
turning point, the input gap value for which the average output gap
is equal to the input gap is the right value to use for estimating
the available bandwidth.
However, there are some issues in the existing active packet-train
probing mechanism. After performing the measurement, three cases are
defined according to the difference between the average output gap
and the input gap. These three cases mean that the average output gap
at the destination is (a) larger than, (b) equal to, (c) less than
the input gap at the source. These three cases have respectively
different relationship between the average rate of the probing packet
train and the available bandwidth. However, the existing mechanism
did not reflect fully these three cases in order to reduce the
measurement latency. That is, both (b) and (c) cases are handled in
the same way, which can introduce the detection error for the turning
point since (b) and (c) cases are absolutely different. Therefore,
the available bandwidth can be estimated inaccurately although the
measurement latency can be reduced.
In this draft, a new mechanism for available bandwidth estimation
mechanism is proposed to improve the estimation accuracy compared
with the existing mechanism. As mentioned before, since (b) and (c)
cases handled in the same way are absolutely different, they should
be handled by respectively.
2.2 An Alternative Active Packet-train Probing Based Estimation
As shown in [3], the end-to-end available bandwidth is defined
as the difference between the bottleneck link bandwidth along an
end-to-end path and the competing traffic. The bottleneck link
bandwidth in the path determines the end-to-end bandwidth which is
the maximum IP layer rate that the path can transfer from source to
destination. In other words, the bandwidth of a path establishes an
upper bound on the IP layer throughput that a user can expect to get
from that path. There are diverse measurement mechanisms for the
bottleneck link bandwidth. Therefore, the bottleneck link bandwidth
can measured from one of existing mechanisms.
There are several important probing parameters such as probing
packet size, number of probing packet in packet train, and input gap
to get correct measurement. Among them, input gap in a probing
Kim Expires May 3, 2020 [Page 4]
Internet-Draft Grasping Network Situation November 2019
packet train is the most important parameter to control for accurate
available bandwidth estimation. The source sends a sequence of
probing packet trains with adjusting input gap. The difference
between the average output gap and the input gap is observed for
each train. Then, the turning point is detected for estimating the
available bandwidth.
After performing a measurement, three cases are defined according to
the difference between the average output gap and the input gap.
Three cases are called 'Red', 'Yellow', 'Green' cases which have
respectively different relationship between the average rate of the
probing packet train and the available bandwidth as follows:
- Red : The average rate of the packet train is more than the
available bandwidth with the following condition:
average output gap > input gap + delta/2.
- Yellow : The average rate of the packet train is similar to the
available bandwidth with the following condition:
|average output gap - input gap | < delta.
- Green : The average rate of the packet train is less than the
available bandwidth with the following condition:
average output gap < input gap - delta/2.
Above three cases are handled respectively as follows:
(1) Handling of 'Red' case
The measurement is repeated with the increased input gap. After
then, three cases observed once again. For each case, the
measurement is repeated with adjusting input gap as follows:
- Red : increased input gap
- Yellow : same input gap as previous measurement
- Green : decreased input gap
In the existing mechanism, the measurement is repeated with the same
input gap as previous measurement for 'Green' case.
(2) Handling of 'Yellow' case
The measurement is repeated with the same input gap as previous
measurement. After then, three cases are observed once again and
then handled respectively as follows:
- Red : measurement with increased input gap
- Yellow : measurement finished (turning point detected)
Kim Expires May 3, 2020 [Page 5]
Internet-Draft Grasping Network Situation November 2019
- Green : measurement with decreased input gap
In the existing mechanism, the measurement is finished for 'Green'
case.
(3) Handling of 'Green' case
The measurement is repeated with the decreased input gap. In the
existing mechanism, the measurement is repeated with the same input
gap in this case. After then, three cases are observed once again
and then handled respectively as follows:
- Red : measurement with increased input gap
- Yellow : measurement finished (turning point detected)
- Green : measurement with decreased input gap
In the existing mechanism, the measurement is finished for 'Green'
case.
As shown in three cases, the proposed mechanism handles 'Yellow' and
'Green' cases respectively while the existing mechanism handles them
in the same way.
When the turning point is detected, the measurement is finished and
then the end-to-end available bandwidth can be estimated as follows.
The end-to-end available bandwidth is obtained by subtracting the
competing traffic bandwidth from the bottleneck link bandwidth.
As mentioned before, the bottleneck link bandwidth can be measured
from one of existing mechanisms. Then, the competing traffic
bandwidth can be computed using the average output gap and the input
gap at the turning point, and the bottleneck link bandwidth.
3. IANA Considerations
This document has no IANA actions.
4. References
[1] N. Cardwell et al., "BBR v2 : A Model-based Congestion Control",
ICCRG, IETF 104, Mar 2019
[2] S. Park et al., "ExLL: An Extremely Low-latency Congestion
Control for Mobile Cellular Networks," Proc. of the 14th
International Conference on Emerging Networking EXperiments and
Technologies(CoNEXT'18) pp. 307-319, 2018.
[3] N. Hu and P. Steenkiste, "Evaluation and characterization of
available bandwidth probing techniques," IEEE JSAC, Vol. 21, No.
6, pp. 879-894, 2003.
Kim Expires May 3, 2020 [Page 6]
Internet-Draft Grasping Network Situation November 2019
Author's Address
Pyungsoo Kim
Department of Electronics Engineering,
Korea Polytechnic University,
2121 Jungwang-Dong, Shiheung City,
Gyeonggi-Do 429-793
KOREA
Phone: +82 31 8041 0489
EMail: pskim@kpu.ac.kr
Kim Expires May 3, 2020 [Page 7]