CC BY
56
Algorithm for balancing the traffic of Ethernet ring networks, applied to execution to the software
At existing data networks released three logical architectural level. There are the core level, aggregation and access level. And if in the aggregation and core level operating with IP packets, and the total bandwidth calculated by summing all of the available communication channels between the points of exchange of information, at the access level, this rule does not operate. This is due to feature of ethernet switches work with frames composed in a lack of protection from "storms". Under the storm meant uncontrolled increase frames broudcast, in consequence of which is hammered all the available bandwidth and consumes all the power switch processor. On a result of the storm the whole network stops working. To solve this problem developed numerous solutions consists in building a tree network with the lack of redundant links. In consequence of that, of all available communication lines between the two switches is selected only one, and all other software disabled. And despite the possible overload of the main channel, the other link can only be used in cases of failure of the chosen path. Such distribution traffic at affordable communication lines very inefficiently. Based on this problem, we have developed a way to at overloads, or deterioration in the quality of traffic, convert some traffic to redundant channels, thus enhancing the quality of service provision to subscribers. For this purpose we offer use of software that continuously monitors switchs at the network with tools of SNMP, and if necessary, you can switch some traffic to the backup channel, thereby allowing efficient use of all existing bandwidth between switches. Due to the flexible mechanisms for determining
Kaywoids MSTP, RSTP, MSTP, BPDU, root, the quality of service delivery, our method allows to optimize the solution for any need of existing net-
switch, root and backup port, rinq topoloqy. . ij. . -l- . » j » i h.
r r or a/ works, and do not require serious business investments at due to the use of open software.
Lihtsinder Boris Yakovlevich,
DPhil, professor of MSIB PSATI. Russia, lixt@samtel.ru
Ryzhikh Sergey Vyacheslavovich,
graduate student of the Volga State University of Telecommunications and Informatics (PSUTI),
Russia, rrserq@mail.ru
Clitheroe Sean,
engineer of telecommunication,
"SilkSmith" company, United Kingdom, blacker.mini@qmail.com
Introduction
There are many ways to control load of data networks. Thus, when using IP packets [1] there is a possibility to use two or more ways of achieving the recipient that allows to balance the traffic. When working with frames [2] this approach is not possible, because the presence of two or more ways to cause loop traffic and as a result the "storm."
To prevent loop traffic to Ethernet networks used algorithms prevent loops, such as STP, RSTP and MSTP Algorithms are used to automatically build the spanning tree. They calculate the best routes between all the devices, according to certain rules, and automatically lock (disable) ports, in case of detection loops [3]. To determine the routes and control connections are constantly are sent service packs (Bridge Protocol Data Units, BPDU) [4]. In case of change structure of the network produced its reconfiguration.
Consider the principle of algorithm loop guard. Figure 1 shows a conventional ring topology of switches.
As the root, selected switch-1. STP protocol provides a logical ring break-2. Break made the switch-3 and fixed in his memory. Neighboring switch-10, located on the other side of the break, such information has not.
The presence logical break of ring divides ring into two branches, in which the information is to the side of the root switch-1 moves toward and against the direction of clockwise, respectively (arrows above). Switch ports are directed towards the main switch are considered active (blackened), and other ports of the ring — redundant. Information on the cost of paths send at ring in special fields "path cost" service frames (BPDU). Logical break don't obstructs passed along the network frames BPDU.
Serious shortcomings such use is the consideration of only the nominal bandwidth connections between switches and the number of switch between points. When the spanning tree completely, all backup links is disabled.
Backup links are not used for data transmission and included only in case of damage to the main channel. Such distribution channels may lead to the fact that one of the ports can be overloaded local traffic, and the real bandwidth in relation to the transit of traffic passing through the spanning tree may differ significantly from the nominal. As a result, the switches may have large queues and traffic delays.
In order to optimize traffic, are encouraged to consider the real load, each of the switch ports.
When overloaded, the whole (or a part of traffic) passing through the congested switch port adaptively switch to a backup path.
Under overload switch port refers to the availability of frames in the queue for this port. In noncongested condition, are no queue in the ports. The queues is checked using a standard mechanism inherent in the switch. This functionality is used in all modern switches.
Implementation of the migration process point break at overloads can be done in different ways. Propose changes in the cost way to implement the
Figure. 1. Segment of access data network Ethernet
transfer point break using the software installed on the management server.
Algorithm for balancing the traffic of Ethernet ring networks, applied to execution to the software.
Server implementation of the algorithm provides the following:
Network diagram represents access switches connected to the node aggregation-1 and a certain server-9 used in any existing network to monitor the status of communication nodes. This server can be located anywhere on the network but have access to the any IP address of any element of the network.
The algorithm considers an overload in the output ports of switches and provides opportunity reduce overload, due to the successive transfer point -2 logical break rings in direction to overloaded switch. For example, in the event of an overload -6 on the output switch port -7, the break moves to position -8. Thus, the traffic from switch -10 is transmitted toward the root switch of the backup path (anticlockwise), resulting in a discharge switch -7. If the overload is not eliminated, and on the left branch does not have any additional overload, then through the selected time interval timer "switch logical ring" (SLR), the next logical break point ring moves one step in the direction of an overloaded switch, etc. With the disappearance of overload break point moves to the starting position.
Consider the method of transfer point break during overloads, based on the use of software installed on the management server.
In the process of software configuration, are entered IP addresses of switches, indicating the address of the switch hook up directly to them. The survey is conducted by SNMP [5] protocol, which allows get information by sending control messages OID [6] on the switch. OID1 requests the status of the queues at the output ports of switches, OID2 requests numbers of logical rings for this port, OID3 determine whether this switch is root , OID4 changes the value cost for operating protocols of STP for this port, OID5 cancels the change of value cost of the port.
Survey switch occurs cyclically, with a specified interval of time, starting with the first entry of IP address management. The survey ends with the last of the list of addresses included in the list management server. The polling interval is determined by a timer query switches (QA), and can be changed by the operator network monitoring. Server contains a parameter determinative number of frames in a queue (NF) and is responsible for beginning of work traffic balancing algorithm. In switches available the counter that tracks queue sizes in each of the ports. In the survey, the server queries the queue sizes in each of the switch ports, and compares with the specified server value NF If the size of the queue in a ports on the switch greater than NF, then turned on traffic balancing algorithm.
QS and NF parameters allow you to use the algorithm for all types of traffic, and for any features of the existing network. For example, using a larger value of NF and long time at QS to provide ability ignore short-term overloads. Small as the value for NF and OK to avoid deterioration quality of service (QoS) for traffic, like delay and jitter.
After fixing using OID1 availability overload physical port (number of frames in the queue exceeds a set value NF), in this switch sent OID2, determines the number of of logical rings in which the overloaded port is active. for achieving root switch-1.
From the resulting list of logical rings selected logical ring with the highest number. Further, in the direction of the switch committed the overload sent questionnaire OID3. OID3 determines whether this switch is root or did this switch have blocked for selected logical ring.
If the switch -7 is not a boundary for the considered logical ring goes, further questioning switches. If the specified logical switch contains a blocked port, then it is the boundary, and to this switch sent OID4, changing the value of port cost in the selected logical ring to the maximum value.
Change of cost guaranteed to get you to change the path to root switch from the switch -10, and therefore move the break point from the position -2 of the ring in position -8. This method changes the value of the path allows use of the mechanisms incorporated in implementation of the standard protocol MSTR
Further, generated record at the server software, showing the switch numbers of switch, number of logical ring and the port whose value has been changed.
This record is stored in memory. Simultaneously, starts the timer of save the change path cost (CPC). At excess the time set in the timer, from the server side to the switch send OID5, returns previously modified path cost to its original state.
And at each step, using OID3 determined by two parameters: whether this logical switch have blocked port and did it is the root switch . If the logical switch is not the root switch or a boundary performed survey next switch.
If, in the case of the cycle survey boundary switch -3 determined before to identify the root switch, then the OID4 send to the previous switch -10, for change the path cost.
In the case of determine the main before determine the boundary switch, OID4 send on the next switch at list switch for change the path cost.
Such a selection switch for sending OID4, made mean the complexity of determining the direction of traffic in the boundary switch. Software has no information to what switch -4, -10 or -2 connected blocked port. Adding this information to the software is unduly time consuming and affects the scalability of the network.
Using the neighboring of the boundary switch allow switching of the traffic from port with overloaded on the backup channel. Amount switched traffic matches the traffic generated by users connected by no more than two access switches.
In the case with not termination overload, in accordance with timer SLR, switching logic switches to the backup channel will be in turn, starting with the highest numbers in the logical ring using the overload point as the primary path.
If, as a result of switching traffic overload is found in any other switch, different from switch the overload, the software is consistently delete existing records of change of cost path, thereby generating OID5 for sequential system returns to its original state.
Later, when the disappearance of overload occurs consistent traffic switches back to the "main" channels. Software, on the expiration of a timer SLR removes appropriate record stored in memory, and command OID5, removes highest bit values in the cost of the port. Thus, changes in highest bit field "path cost" guaranteed to get you change the path to the main switch, the switch from-10, and therefore move the break point in the ring make the initial position-2.
Conclusion
The developed algorithm improves the efficiency of data transmission networks, and improves traffic processing, such as latency and packet loss. The algorithm does not change the spanning tree protocol, but only uses the mechanisms available data protocols.
With the ability to take into account different index appeared processing engine traffic based on a number of criteria, such as the time of day or time of year, which allows for any distribution of traffic, conceived by the operator.
References
1. Victor and Natalia Olifer, information and analytical materials. http://www.opennet.ru/docs/RUS/ ip_network/glava_4.html, viewed 5 March 2014.
2. Base of Ethernet. Ethernet Layer 2 : Frame format and addressing at the link layer. http://admindoc.ru/ 774/ethernet-ethernet-layer-2/viewed 27 March 2014.
3. Loopdetect. http://animage.me/?p=137 viewed 27 March 2014.
4. Wikipedia.BPDU. http://ru.wikipedia.org/ wiki/BPDU viewed 27 March 2014.
5. Semenov YA Management protocol SNMP http://book.itep.ru/4/44/snmp_4413.htm viewed 27 March 2014.
6. Mark Silinio. Commands reference Net-SNMP http://www.opennet.ru/docs/RUS/net-snmp/ viewed 27 March 2014.
7. Net-SNMP Tutorial — Commands http://ru.wikipedia.ong/ wiki/BPDU, viewed 27 March 2014.
