CC BY
48
THE METHOD OF LOAD BALANCING DISTRIBUTED MANAGEMENT IN LOW EARTH ORBIT TELECOMMUNICATION
SATELLITE SYSTEM
Ivanov Valery Igorevich,
MTUCI, NIO-48, research fellow, Moscow, Russia, [email protected]
Keywords: satellite telecommunication system, low earth orbit, routing, load balancing.
The method of load balancing distributed management in low earth orbit telecommunication satellite system (LEOTSS) is presented. The goal of load balancing management is to find for all data flows such routes that minimize the probability of packet discard and maximize the total throughput of LEOTSS. Methods of load balancing management are divided into centralized and distributed methods. In centralized methods, either one dedicated satellite or one earth station solves the problem of load balancing. In distributed methods, all satellites together solve the load balancing problem. Every satellite in the proposed method sends periodically special packets to discover routes. Packets are randomly forwarded to destination. The probability to choose next satellite in the packet route depends on mean packet loss probability of routes traversing through the next satellites. The less mean packet loss probability, the more probability to choose satellite. Information of line packet loss probability and delay is recorded in packet during it's forwarding. After the packet gets to destination, it is sent backwards by the same route. The satellites, which send packet back, update mean route packet loss of routes to the destination of packet. Based on information from returned packets, satellites define routes and distribution of flows on these routes to minimize packet discard probability and maximize LEOTSS throughput. The results of computer simulation showed that proposed method provides lower packet loss probability and higher LEOTSS throughput compared to other existing methods.
Для цитирования:
Иванов В.И. Метод распределённого управления балансировкой нагрузки в низкоорбитальной спутниковой системе связи // T-Comm: Телекоммуникации и транспорт. - 2015. - Том 9. - №12. - С. 67-71.
For citation:
Ivanov V.I. The method of load balancing distributed management in low earth orbit telecommunication satellite system. T-Comm. 2015. Vol 9. No.12, рр. 67-71.
г Г\
Introduction
The method of load balancing distributed management in low earth orbit telecommunication satellite system (LEOTSS) is presented. Due to low earth orbit, LEOTS5 has low propagation delay and low requirements for line budget compared to telecommunication satellite systems (TSS) located on other types of orbits [1]. Therefore, LEOTSS are more suitable then other types of TSS for real time and multimedia data transmission.
However, because of low orbit altitude of LEOTSS, coverage area of one satellite is very small. Hence, for global coverage, one needs great number of satellites. Satellites are usually connected via intersatellite links in a network. Data, in such networks, are in most cases transmitted through chain "sender - satellite of sender - chain of satellites - satellite of receiver - receiver" (picture 1).
As a result, the problem of routing in LEOTSS arises.
from packets and define routes and distribution proportions.
Let us consider the method in detail.
Method
Every satellite sends periodically in time tp « rservice packets to those satellites, to which at least one packet was sent during last /y periods of load balancing.
JiiSaJfc?
Jwf Шш
SSE
лважНй 2 ■ Jr^?^
Ш
1
Picture 1. Data transmission in LEOTSS
There are two types of routing methods: methods of QoS routing and methods for load balancing management [2]. Goal of QoS routing methods is to find best path or path that fulfills given requirements [3]. Goal of load balancing management is to find for all data flows such paths, with which the lowest probability of packet drop throughout all network is achieved, or, accordingly, maximal throughput of network is achieved [4].
In this article, the method of load balancing distributed management in LEOTSS is presented.
In the presented method, load balancing management is the process of finding for every source-destination pair multiple routes and proportions of data flow distribution on these routes. Data flow is a collection of all packets between source and destination. Load balancing is conducted periodically with period of length T. This period is named load balancing period.
The essence of method is following. To find routes, every satellite periodically sends service packets during load balancing period. Packets are randomly forwarded until they reach destination. Then destination satellites send service packets back by the same, but reversed route. After coming back, satellites extract information
Packet distribution. Start
l';.<"k;1 citation Start
Packet creation
1 r
Send packet
Г
f
Recced source satellite identifier
. .
Record destination satellite identifier
Retord identifierof source satellite into route vector
Creation of packet^ L End J
Packet distribution. End
Picture 2. Creation and sending of service packet
Table 1
The table of last packet transmission
Identifier of sate Hite-rece i ver Time of last packet
transmission
1 —
* *
15 16:32:28
* *
66 —
Sequence of packet distribution is following (picture 2), First, walk through the table of last packet transmission (table 1). If time passed from the last packet transmission to a given satellite is less then length of N periods of load balancing {N7), or no packets were sent, then we go to the next row of table. If time is less than NT, then satellite creates new service packet and send it to find destination satellite. Thus, satellite search routes only to those satellites, to which packets were recently sent. For other destination satellites, satellite take shortest by number of links path.
Service packet consists of the following fields:
1. Identifier of source satellite.
2. Identifier of destination satellite.
3. Vector M with sequence of satellites in packet route.
4. Vector P with probability of packet drop for each line packet have traversed.
5. Vector £>with delay of each line packet have traversed.
Before sending, satellite puts down into packet identifier of source satellite (which is its own identifier), identifier of destination satellite, and satellite puts its own Identifier into the beginning of vector A7(picture 2).
During forwarding process, satellites add their identifiers to vector M, probability of packet drop to vector /'and line delay to vector D of the line that packet was forwarded through. Vectors look like this:
• M = {5„52)...,5Af}, where Sl - source satellite,
Su - receiver satellite.
• P = {P]_2,P2_i.....PN.x.N}t where - probability
of packet drop of line between source satellite and second satellite in path, Ps_ ]N - probability of packet drop of line
between penultimate satellite in path and destination satellite.
• D={Dl_2,D2 where - delay of
line between source satellite and second satellite in path, Dn_- delay of line between penultimate satellite in
path and destination satellite.
Let us now consider the method of service packet forwarding (picture 3). This method is used for sending packet from source satellite and for forwarding packet from other satellites. First, satellite checks whether packet has traversed more than Nmia lines. If packet have moved
further, then satellite deletes packet. Without Nmax packets can travel almost forever. Packet can reach destination with outdated line Information,
If packet has traversed less than N lines, then satellite extracts identifier of source satellite and destination satellite from packet. Checks whether packet has arrived to destination. If so, then satellite delete loops from packet route in vector M(also satellite deletes from vectors /'and D information about deleted lines) and sends packet back to source satellite.
If packet has not arrived to destination, then satellite chooses randomly next line to forward packet through. Satellite adds to vector M identifier of satellite to which next line leads; to vector P, we add line drop rate; to vector D, we add line delay. After updating fields, satellite forwards packet to next satellite.
Let us consider the method of sending packet back (picture 4). Backward route is following:
^={5^-%,...A}'where sN -destination satellite,
S] - source satellite.
At the beginning, satellite defines identifier of source and destination satellite. Then satellite defines the probability of packet drop for route from intermediate satellite to destination satellite. Let the vector P = {PV2,...,PK_K+]PN_,_N} with line packet drop rates is
given. PK_K+I - Is packet drop rate of line between intermediate satellite and next satellite closer to destination. The probability of packet drop for route from intermediate satellite to destination satellite is following:
ош
Then route drop rate is added to statistics table (table 2).
Table 2
Statistics table
Identifier of source satellite Identifier of destination satellite Vector of route packet drop rates
23 48 0.05 0.01 0.07 0.03 ... 0.09
* * *
38 9 0.03 0.02 0.09 0.06 ... 0.09
Table 3
Preliminary routing table
Identifier of receiver satellite Route identifier Sequence of satellites in route Route packet drop rate Route packet delay, s
23 Ibc29b36f623ba82aaf 6724fd3bl6718 1 2 5 10 16 23 0.003 0.148
38 d41d 8 cd98f0 0 b204e98 00998ecf8427e 1 24 15 26 27 38 0.01 0.132
Then satellite checks whether packet has arrived to destination. If not, satellite forwards packet further. If packet has arrived to destination, then satellite updates preliminary routing table. Preliminary routing table is showed in table 3. First column of table is identifier of receiver satellite. Second column Is identifier of route. It is calculated by taking MD5 hash sum from sequence of satellites in route. Third column is sequence of satellites in route. Fourth column is the last known route packet drop rate. Fifth column is the last known route packet delay.
Preliminary routing table is updated as follows. At the beginning, satellite by vector M defines the route identifier. After that, the satellite searches route by identifier. If there is no route in table, new entry is created. If route is already in the table, then satellite updates route packet drop rate and delay, they are taken from vectors /'and D.
Residual routes are added to routing table. Distribution proportions are defined from the principle of equal packet drop intensities proposed in [5]:
P,r,R - pMrMR = ■■■=pMrMR^
p. - route packet drop rate, R - intensity of data flow between source satellite and destination satellite, rt -share of data flow for route i from the found route set.
Satellites use routing tables calculated during previous load balancing period in current load balancing period.
Numerical modelling
The numerical modelling of proposed method is conducted. Method is compared to ELB method [6] and shortest path method based on Dijkstra algorithm. The model of satellite system is similar to Iridium satellite system. Parameters of satellite system are presented in table 4.
Table 4
Parameters of satellite system
Orbit altitude 780 km
Number of planes 6
Number of satellites in planes 11
Orbit inclination 86,4°
Difference of longitude of ascending nodes between corotatinq planes 31,6°
Difference of longitude of ascending nodes between counterrotating planes 22°
Minimal elevation anqle 8,2°
Intersatelllte lines per satellite 4
Capacity of intersatellite line 25 Mbit/s
Capacity of uplink/downlink 1,5 Mbit/s
Capacity of line buffer 200 packets
Maximum latitude for interorbital lines 60°
Number of ground terminals 700
The next line is selected randomly with the following probability:
_ ' Ы'- л
p. is the mean value of packet drop rates of intermediate
routes. Drop rates are contained in statistics table (table 2).
At the end of load balancing period, routing table is defined based on preliminary routing table.
The method to calculate routing table for one destination is following. First satellite discards those routes whose delay Is larger than following: b = dnta+{dmB-dm.n)а, где
og[0,1] dmin - is minimal route delay, dmax - is maximal
route delay. Parameter a is defined empirically.
- Proposed method -ËLB Shortest path
--1—i—i—r
till! +—i —( —+ —Í--I I I I I I t A-"
J__1__L__L._J._J__— I__
1 I I A * ( I _ I _ [ J_ _ 1 I I ¡P 1 7
[ I 1
I t I I I J
__!__1__t-.J__.L_L.
i 1 I i 1 1 i I t 1 . ¿ T ■ - - тг - f - r-r ■
1 A T I I J
Plots of dependence of drop rate from flow Intensity from one terminal are showed on picture 5. Proposed method exhibits lowest drop rate. Under high flow intensity drop rate of proposed method and ELB are almost equal.
Plots of dependence of LEOTSS throughput from flow intensity from one terminal are showed on picture 6. Proposed method provides highest throughput. Under high flow intensity throughput of proposed method and ELB are almost equal.
0.81—:—:—-——r-r
0.6 0.7 0.8 0.9 1 1 1 1.2 1.3 1.4 1.5 1.6 1.7 1 a 1.9 2 2.1 2.2 2.3 2A 2.5 Data flow intensity from one terminal. Mbit/s
Picture S. LEOTSS packet drop rate
PUBLICATIONS IN ENGLISH
МЕТОД РАСПРЕДЕЛёННОГО УПРАВЛЕНИЯ БАЛАНСИРОВКОЙ НАГРУЗКИ В НИЗКООРБИТАЛЬНОЙ СПУТНИКОВОЙ СИСТЕМЕ СВЯЗИ
Иванов Валерий Игоревич, НИО-48, научный сотрудник, МТУСИ, Москва, Россия, [email protected]
Представлен метод распределенного управления балансировкой нагрузки в низкоорбитальной спутниковой системе связи (НССС). Цель управления балансировкой нагрузки - найти для всех потоков данных такие пути, при которых достигается наименьшая вероятность отбрасывания пакетов или соответственно максимальная пропускная способность всей НССС. Методы управления балансировкой нагрузки делятся на централизованные и распределённые. В централизованных методах балансировкой нагрузки управляет либо один выделенный спутник, либо центральная земная станция. В распределённых методах все спутники совместно решают задачу управления балансировкой нагрузки. В предложенном методе каждый спутник для определения маршрутов периодически рассылает служебные пакеты. Служебные пакеты случайно пересылаются, пока не доходят до получателя. Вероятность выбора следующего спутника в маршруте служебного пакета зависит от средней вероятности потери пакетов в маршрутах до получателя, проходящих через следующий спутник. Чем меньше средняя вероятность потери пакетов, тем выше вероятность выбора спутника. В процессе пересылки в пакеты записывается информация о задержке и вероятности отбрасывания пакетов в линиях. После того, как пакет дошёл до получателя, пакет отправляется пройденным маршрутом обратно. По данным из пакета спутники, которые пересылают пакет обратно, обновляют средние вероятности потери пакетов в маршрутах до получателя. После того, как пакеты возвращаются, спутники определяют по информации, содержащейся в пакетах, маршруты и пропорции распределения потоков данных по ним. Результаты имитационного моделирования показали, что созданный метод обеспечивает значительное снижение вероятности отбрасывания пакетов и повышение пропускной способности НССС по сравнению с другими существующими методами.
Ключевые слова: спутниковые системы связи, низкая земная орбита, маршрутизация, балансировка нагрузки.
Литература
1. Аболиц А.И. Системы спутниковой связи. Основы структурно-параметрической теории и эффективность. - М.: ИТИС, 2004.
2. Deyu Meng. Yongfa Ling A Genetic Optimization Algorithm to Solve the Problem of the Load-Balancing of Network Load / IJCSNS International Journal of Computer Science and Network Security, VOL.6 No. 7B, July 2006. Pp. 63-68.
3. Yan Chen, Toni Farley and Nong Ye. QoS Requirements of Network Applications on the Internet // Information, Knowledge, Systems Management 4 (2004), pp. 55-76.
4. Zhenyu Na, Zihe Gao, Yang Cui, Liming Chen, Qing Guo. Agent-Based Distributed Routing Algorithm with Traffic Prediction for LEO Satellite Network // International Journal of Future Generation Communication & Networks, June 2013, Vol. 6, Issue 3, p. 67.
5. N. Taft-Plotkin, B. Bellur, R. Ogier Quality of Service Routing using Maximally Disjoint Paths / IWQoS '99. 1999 Seventh International Workshop on Quality of Service, 1999. Pp. 119-128.
6. Taleb, T.; Mashimo, D.;Jamalipour, A; Kato, N. Explicit Load Balancing Technique for NGEO Satellite IP Networks With On-Board Processing Capabilities // IEEE/ACM Transactions on Networking, Vol. 17, Issue 1, Feb. 2009, pp. 281-293.
