Organization: China Interactive Publishing Network (http://www.china-pub.com/)
RFC Document Chinese Translation Program (http://www.china-pub.com/compters/emook/aboutemook.htm)
E-mail: Ouyang@china-pub.com
Translator: 15222775 @ 61. (15222775 @ 61. Hbzzx2001@yahoo.com.cn)
Translation time: 2002-1-18
Copyright: This Chinese translation copyright belongs to China Interactive Publishing Network. Can be used for non-commercial use free reprint, but must
Keep the translation and copyright information of this document.
EGP gateway under Berkeley Unix 4.2
(RFC911 - EGP GATEWAY Under Berkeley Unix 4.2)
Paul Kirton
University of Southern California Information Science
Accessing researcher from the Australian Telecom Research Laboratory
22 August 1984
Summary
This report describes the implementation of external gateway protocols running on a 4.2BSD UNIX operating system. Also on certain and local
The problem related to the network configuration has also been discussed.
This memo:
This memo describes the implementation of the External Gateway Protocol (EGP) (in this sense it is a status report). memorandum
Certain possible extensions and design issues are also discussed (in this sense, it is a more discussion of the invitation). This memission
The distribution of records is not limited.
This research fund fund is supplied by the National Defense Advanced Research Project Authority and Australian Telecom.
table of Contents
Introduction 2
1.1 Development Machine 3
1.2 External Gateway Protocol Summary 4
2. Gateway Design 4
2.1 Routing Table 5
2.1.1 Enter Update 5
2.1.2 Output Update 6
2.2 Neighbor Get 7
2.3 Hello and Poll Polling Interval 7
2.4 Neighborhood Statue 8
2.5 Neighbor Vacuum 8
2.6 serial 9
2.7 Treatment of Excessive Commands 9
2.8 Inappropriate Messages 9
2.9 Default Gateway 10
⒊ Test 10
⒋ future improvement 11
4.1 multi-independent system 11
4.2 Interface Monitoring 11
4.3 Network-level status information 11
4.4 Internal Gateway Protocol Interface 12
⒌ ⒌ 拓 问题 12
5.1 Topology Constraints and Routing Road 12
5.1.1 Background 12
5.1.2 Current Policy 13
5.2 Current ISI Configuration 14
5.2.1 EGP Crossing Apa Net 16
5.2.2 EGP Crossing ISI - NET 17
5.2.3 Potential Routing Road 17
5.3 Possible future configuration 18
5.3.1 Go to UCI - ICS Gateway 18
5.3.2 Dynamic to Backup Gateway 18
5.3.2.1 Commonly used operation 18
5.3.2.2 Host Initialization 19
5.3.2.3 When is the original and backup of both
5.3.2.4 UNIX 4.2 BSD 19
⒍ Thank you 20
⒎ Reference 20
1 Introduction
External Gateway Agreement (External Gateway Protocol) [Rosen 82; Seamonson & Rosen 84; Mills 84a]
Although the international internet route selection information has been allowed to maintain global distribution although independent development of different gateway systems.
EGP provides some ways to exchange information for different autonomous networks to provide information via their network they can reach. .
This report mainly describes the Berkeley UNIX 4.2 operating system running on a VAX computer as a user
Realization of the EGP of the process.
There is also a discussion of some issues related to local autonomous system configurations.
The EGP implementation is part of experimental rather than UNIX 4.2 BSD. It can be expected that Berke will join a future
EGP version.
The program is written in C language. The EGP part is a C-Gateway, which is written in Liza Martin in MIT Institute.
The order-based code is based, and the routing management section is based on the "ROUTED" of the management daemon "ROUTED" in Unix 4.2 BSD. The EGP function is consistent with the specification of [Mills 84a] unless specified.
It is assumed that EGP knowledge has been described in [Seamonson & Rosen 84; Mills 84a]. .
This chapter discusses the motivation of the project, Chapter 2, describes the gateway design, Chapter 3 is test, Chapter II recommends some improvements and 5th
Chapter on the topology problem.
Details of the EGP program and the software instructions are published in ISI Research Report ISI / RR - 84 - 145 [Kirton
84].
The request of COPIES and text of the EGP program should be sent to Joyce Reynolds (Jkreynolds @ _USC -
ISIF.ARPA). No software support is provided.
1.1 development
Consistent with the introduction of EGP, the Internet gateway is divided into "core" maintained by Bolt, Beranek, and Newman (BBN)
Autonomous Systems (AS) and many of the minimum maintained by different organizations have a network with a core AS gateway
"stub" AS.
This core AS may act as a hub that transmits route selection information
_______________
*
Unix is a trade Mark Of AT & T
* *
VAX is the trademark of Digital Equipment Corporation
Different stub as' s so that it only requires STUB AS to perform EGP with the core gateway. Further details
It is given in [ROSEN 82].
In this project, there have been 28 "Non - Routing" gateways in the Internet. Non - Routing Gateway cannot be exchanged
From the selection information, it is necessary to record static records in the core gateway route selection table. After August 1, 1984,
These static expressions have been deleted and the previous Non - Routing Gateway is required to be dynamically conveyed via EGP [Postel 84]
This message gives the core gateway.
In the USC Information Science Research Institute (ISI) has a Non - Routing Gateway to the Irvine Network (UCI - ICS)
University of California. Since the Non - Routing gateway is eliminated from the core network cluster, it is necessary to notify the route
The core ISI gateway to UCI-ICS uses EGP.
Also, in case the core ISI gateway, we want to have a backup network between ISI - NET and ARPANET.
turn off. In this way, the gateway may require route selection information via EGP transportation. Details of the ISI network configuration at 5.2
Discussion.
Of the 28 Non-Routing Gateway 23 Were Implement by UNIX Systems, Including
Isi's. Also, ISI's proposed backup gateway is a UNIX operating system. Therefore, there is a local and universal needed external network
The protocol is implemented under UNIX. The Unix Operation Department currently included in the Department of Defense (DOD) protocol
The version version is Berkeley UNIX 4.2, so it is selected.
1.2 Overview of External Gateway Protocol
This report is assumed to know the external gateway protocol, but see a short overview for the integrity. For the background of EGP
Please refer to [ROSEN 82], [Seamonson & Rosen 84] provides informal description and [Mills 84a] is provided.
Further format description and details.
EGP is typically transmitted between different AS and has a gateway that has a public network.
The public network is a neighbor gateway.
EGP contains three programs, neighbors, neighbor, and network accessibility.
Neighbor is a two-way handshake, in which the gateway pairs through exchange requests and confirm packets (including least Hello and
The POLL interval value) is consequential for transmission. Terminate the acquisition process by switching stop and stop confirmation packets.
Neighbor Dormy is a Hello Commands and ensuring that the I-H -U (I Heard You) responding to the gateway is in response to cyclical exchange. . The least 30 seconds minimal intervals are used to cross the arpanet. Only one gateway must send some commands,
Other gateways can use these commands to judge the achievability.
The gateway that is sending the reachang command is called at the active, and the gateway that is answering is called being dynamically.
Network accessibility is cycredient with poll commands and accepts the network via one or more networks on a shared network.
The update response that can be reached is judged. The minimum interval at the current 2 minutes is used to cross the Arpanet.
2. Gateway design
EGP is a polling agreement with loose timing constraints. Therefore, the only function that requires higher performance is a package
hair. UNIX 4.2 has embedded packet forwarding techniques to the internal core kernel, where best performance can be achieved.
When writing UNIX 4.2, there is no ICMP (Internet Control Message Protocol) redirector for the wrong route.
Text.
This is an essential Ofinternet gateway and will be added later in Berkeley.
EGP and routing updates are implemented as a user process. Because only minute changes need to join the UNIX kernel,
This simplifies development and distribution. This method is based on Xerox NS Routing Information Protocol (Routing Information
The method of the UNIX routing program "ROUTED" [Berkeley 83] of Protocol [Xerox 81]) is similar.
2.1 routing table
A routing consists of a destination network number, the address of the neighboring gateway is used on the network directly connected, and it is reached
The number of gateway stations of the network acts as a metric.
There are two sets of routing tables, kernel lists (used for package forwarding techniques) and EGP process tables. This kernel has been host and network
A separate list of separate destinations. The EGP process only maintains network routing tables. EGP when receiving the EGP update message
The list is updated. When a routing is changed, the core network list is passed through SiocadDRT and SiOCDELRT IOCTL.
System call is updated. In initialization, in order to maintain consistency, the kernel network routing table is imaged by the kernel store image file.
/ DEV / KMEM reads and copied to the EGP list.
This EGP implementation is designed to run on the gateway (also a host).
Because the route update is relatively slow by polling, the host passes the host through ICMP before the EGP process is notified.
A directionally received a routing change is possible. Redirect direct update kernel list. EGP process is in a sparse plug
Sales listen on the redirect message and update its routing table to maintain the consistency of them with the kernel.
The EGP process and routing tables are maintained for two separate lists, one for external routing (VIA DIFFERENT AS gateway)
One is the internal route (through the gateway of this AS). The external routing table is updated through the EGP update. Internal route
The table is currently static and set when initialized. It includes all direct connection networks through SiocgifConf
IOCTL system call and any internal non-Routing gateway read from the external gateway protocol initialization file, EGPinitFile is indeed
set. In the future, internal routing tables can be dynamically updated by the internal gateway protocol (IGP).
Separate listings for external and internal routing maintenance, simplifying only the outgoing updates containing only internal routing information [Mills 84b]
Preparation for packets. In the case of internal routing failure, it is also allowed to select an external route saved as an internal routing backup.
replace.
The candidate routing marks thwes in RTS_NotInstall indicates that the kernel route will not be updated. In the current implementation
Candidate routes are not used.
2.1.1 Enter Update
The EGP update is used to update the external routing table, if one of the following:
- There is no route entry for the destination network and the metric indicates that the route can be achieved.
- The reported gateway is the same as the current route.
- The reported distance metric is less than the current metric.
The current routing is old than that of all EGP neighbor limit polls. In short, the route is from last time
Update is ignored.
In addition to the provision of the route, if any external routing table is not updated by the EGP within 4 minutes or 3 times maximum polling interval (any larger) time, the external routing table is deleted.
If multiple EGP neighbors are obtained, the same processing is done in the previous order of the received.
In the worst case, a routing that is time to become time, and the older route is not announced.
When it is accessible, you can update a route from two polling intervals. With The Current Poll Interval this Could BE 4
Under UNIX 4.2 BSD, the TCP connection (Transfer Control Protocol) is automatically turned off after 6 minutes of idle. Instant
With the worst case, it will not cause the TCP connection to end. o
2.1.2 Output Update
Output updates include direct and static networks from internal routing tables, except for networks shared with the EGP neighbor. .
The network that allows notified in the update can be specified by EGPinitFile when initialization. These allow precise routes
Exclusion of external updates from (possible) routing circuits. Other cases of this option are, non-routing network
It is a different AS and this AS has not yGP. Its route may need to be classified into the core route
Table but they are not allowed to be told in the output update.
If the internal routing table includes other internal gateways on the network shared with the EGP neighbor, they will be as they
Contains a suitable initial station of the network is included in the update.
The distance to the network is set according to the internal routing table, unless the routing flag is Down, and the distance is set in this case.
Take 255. At present, the routing is only marked as Down when the output interface is output. Prepare for each output update,
First call the status of all interfaces with the Siocgifflags IOCTL system.
Unseave without request.
2.2 neighbor acquisition
EgpinitFile lists the address of the reliable EGP neighbor gateway and reads when initialized. These are usually core
Gateway, because only the core gateway provides complete Internet routing information. Present on Apa.com
Three core gateways support EGP CSS - GATEWAY, ISI - GATEWAY and PURDUE - CS - GW, and two
In Milnet, BBN - Minet - A - GW and Aeronet - GW.
EgpinitFile also contains the maximum value of the gateway you can obtained. The expected value is usually one. If this gateway is
Announced Down, the gateway in another list will be automatically obtained automatically in a sufficient time to ensure that the current route is not
time out.
The gateway will only accept acquisition from neighbors in the reliable list, and if it has got it to get its maximum amount of limit
I will accept them. This prevents acceptance from an update from a source of unreliable sources.
Because not all core gateways run EGP, it does not contain the core gateway is not in the trust list, but already
By updating the functionality of the core gateway indirectly informed of the message.
The new acquisition request is sent to the neighbor, in order to make them appear in the egpinitfile. If new request is not greater than
The maximum number of neighbors to be obtained is sent immediately. Many burst requests have been returned 5 times with 32 seconds.
At this time, the resend interval is increased to 4 minutes. As long as the maximum number of neighbors have been obtained, it has a prominent request.
UNACQUIRED neighbor is sent to CEASES. This method provides a quick response compromise, and when the neighbor does not respond
And the minimization of a neighbor may be a non-confirmation that has been issued but has not received a confirmation. If the designated neighbor
The maximum number of home has not been obtained, requested to be transferred to all UNACQUIRED neighbors.
2.3 Hello and Poll Polling Interval
The request and confirmation information contains the minimum of Hello and Poll intervals. The minimum number of words hello and the core gateway is currently respectively
It is 30 and 120 seconds.
The received interval is to check the upper bound to prevent its value from meaningless. The current upper industry is set in 120 and 480 seconds, respectively. if
A neighbor exceeds the upper bound, is considered illegal and no longer sends a request for one hour. This allows this situation to be corrected, and the gateway is running normally until the gateway automatically recovers without having to produce too many unnecessary network letters.
Interest stream.
The actual Hello and Poll intervals are determined by the first selection by the interval of the gateway with its peer gateway. consider
By the possible network delay transition 2 seconds, the difference is added to the Hello interval, and the POLL interval increases the integer multiple of the Hello interval.
This causes 32 seconds of Hello with 128 seconds of POLL interval.
If a response Poll update is not received, send a repoll (the same number of sequential numbers) instead of the next one.
Hello.
2.4 Neighborhood
If the EGP process sends a Sigterm signal through the kill command, all the neighbors have sent CEASE (stop)
command. Ceases repeats 3 times in Hello intervals. Once all have any of the Cease-ACKS
The response or a three-fold Ceases that has been sent, the process is terminated.
2.5 Neighbor Diability
Only the current reachability determination is implemented. Its behavior is tiny change in [Mills 84a]].
The response shift register is maintained. Regarding each Poll or Hello command, a zero change is sent to a shift register. in case
A response (I - H - U, update or error) get the correct serial number reaction, zero is replaced. Before each new command is sent
Cavity is judged by checking the upward shift registers. If the neighbor can be achieved and the received response <= 1, the
Neighbors are considered unreachable. If the neighbor is considered unreachable and received answered> = 3, it is now considered to achieve
of.
After a neighbor is acquired, it is immediately considered to be reached, so the first origin is from a core gateway (once it thinks
This gateway can be reached) Poll answers with an update. Polls does not send unless a neighbor has considered a possible
And it is not recommended in its last Hello I - H - U or Poll packets it thinks that this gateway is not reached. This one
Prevent the first POLL from being discarded after a DOWN / UP conversion. This is important because polling is used for accessibility determination.
Following at least one acquisition message must be received, before the first polling transmission. This will determine the associated body does not think this gateway
Stop. It usually requires at least one Hello before the first POLL. The discussion of this paragraph is different from [Mills 84a], it
It is recommended that a bobbin is considered downtime and once the osteotome is considered to be turned on Polls. This is only
An important part of [Mills 84a].
The Polls received by the bodies that are considered unreachable, it allows them to determine the correct place
Improve.
This is an [Mills 84A] option.
When a neighbor becomes unreachable, it is deleted from the routing table from the routing table. If you know that you can't get neighbors
This unreachable gateway is stopped and attempts to get a new neighbor.
If you know all the neighbors already obtained, the reachability decision is extended for 30 minutes ([Mills 84A] is recommended for 60 minutes), at timing
After the neighbor abort is aborted and Reacquisition tried every 4 minutes.
The purpose is to reduce unnecessary network information flow throughput.
If three consecutive polling do not receive the correct update answer, the neighbor is stopped and attempts to get or in 4 minutes.
Rechange in Recquisition. This provision prevents the incorrect update data format of the neighbors. These situations are once tested
The process appears in the process.
2.6 serial number
The serial number is processed in a manner recommended in [MILLS 84A]. Separate sequence numbers for each neighbor maintenance. hair
The sequence number is assigned to the first value and transmits 1 before each new Poll (non-Repoll) is not added before other time. Sequence
The column number is used for all commands. Receive sequence number by copying the last source from a neighbor request, Hello or Poll command
maintain.
This serial number is used to output updates. All responses (including the error response) returns the serial number of the message received.
2.7 Treatment of Excessive Commands
If you receive more than 20 commands from an neighbor within yourself within 8 minutes, the neighbor is considered illegal, suspended.
Reacquisition an hour.
Before the Poll Interval expire (less than 4 second network delay variable variable variable variable variable variable variable variable variable variable variable variable, with an update reply, other incorrect reactions are sent. When sending an updated repoll in the response,
Unsolicited [unsolicited bit] is not set, it is different from suggestions in [Mills 84a].
2.8 Inappropriate Messages
If you confirm that Hello, I - H - U, Poll or updates are received from any gateway (known or unknown).
The UNACQUIRED state, the synchronization is about to be submerged due to some reason. Send suspension (protocol illegal) packets to avoid and reduce
The network information flow of the network. This is an [Mills 84A] option.
2.9 Default Gateway
The default gateway is described in detail in Egpinitfile. By default route (NET 0 in UNIX 4.2 BSD) is used for kernel packages
Send, if the destination network has no specific route. It provides ending level if all the well-known EGP neighbors are not arrogant
Part. If there is only one available EGP EGP neighbor, this is especially useful, like this in ISI in 5.2.2.
The default route is installed at initialization and is deleted after a correct EGP update message is received. If all have been
The neighbors have no one can reach the reinstallation, if the route timeout is timeout before the process is termination, there is no
EGP neighbors obtained and can be achieved.
Because the default gateway does not know more about the more route selection information you have learned by EGP, a correct EGP update report
It was deleted after the text received. If it is not deleted, all traffic that can be transferred to the network may be sent to UNIX 4.2 forwarding
The default gateway below the strategy.
The default gateway will be set to a standard all-round core gateway rather than knowing the EGP GATEWAYS at the same time.
Another backup EGP neighbor gateway is available upon shutdown.
⒊ test
Some interesting situations occurring during testing.
Different implementations have different interpretations of the serial number. Therefore some packets that realize the rejection of incorrect serial numbers, leading to the right
The gateway was announced.
The problem is that this specification is completely text report, which is easy to generate inconsistencies, vague and imperfect. [MILLS 84A]
More format descriptions have been removed.
When the test response is sent to a neighbor gateway not in the package of the same network, the loop is generated because two gateways
Repeat the error message indicating an illegal interface. The problem is that after the two gateways are inspected, EGP is
The interest type is previously sent an error response. By do not send an error response unless it is confident that the message is not an error response
It is correct.
Once a core gateway has some form of data error in the update message, they have been rejected even if they are
Satisfactoryly transmitted. This leads to all routing times. The solution is that the calculation does not cause the connection to the correct update.
The number of continuous polls, if this number reaches 3 to abort EGP, try to get a replacement gateway.
Another feature of interest by Mike Karels in Burk, with more features in Milnet and Arpanet
Between both the gateway. Each ArpaNet host has a given Gateway to access the MILNET. Such as
If the EGP GATEWAY is a host is another gateway, and the EGP update message may indicate a different
Milnet / Arpanet Gateway from one of the given. When Host / Gateway occurs through the gateway routing reported by EGP
When a package, it will receive a redirection of a given gateway to it. So the MILNET gateway can be in EGP
The reported gateway and the assigned gateway remain exchanging. When using it to others
There is a similar thing in the network that Milnet / Arpanet Gateway.
⒋ future improvement
4.1 multi-independent system
Currently a maximum way to get a list of trusted list EGP neighbors suggest that all neighbors are located in the same AS
in. It means that they all belong to the core AS. When updating the routing table, the update is carried out independently has no distinction.
From the inside of the AS of the peer, it is still outside. Also, the route selection metric has no reference source AS.
If the EGP needs to be managed by additional core AS, you need to get all neighbors on the list to ensure that from two
The gateway of AS is always available. This leads to an unnecessary additional EGP traffic, if you get a lot of neighbors for security. A more desirable method will be to hold a separate trusted EGP GATEWAYS list and obtained for each AS.
Maximum number. Routing registration needs to hold some source AS to give priority to information from its own AS (see 5.1.2)
Festival).
4.2 Interface monitoring
The current interface status is only checked before the update of a response Poll. Interface status should be more frequently monitored and
And detect a changed update when a change is detected. At this time, I improve the slow aspect of the EGP polling response.
This has special importance for connecting the network through the Dial - IN line.
When the above network is dial, its related interface will be marked as UP but until the change has passed EGP
It won't receive a package before sending. This will be helpful for the situation of unshrouded packets, but for other non-nuclear
The heart gateway will still be delayed when the new route selection information is polled when the core EGP gateway.
Start thinking this situation may be helpful for the implementation of kernel EGP but the kernel does not join new
Device transfer interface status changes.
If this is done, it may be as simple as a user-level signal when an interface status changes.
4.3 Network-level status information
Current network-level status reports such as IMP destinations are unreachable, not to detect in EGP neighbors or other neighbors
Changes in the reachability. This information should be used to improve response time changes.
4.4 Internal Gateway Protocol Interface
All AS internal route selection information is static, and they are selected from the initialization file. Internal route
The management functions have been written so that it reasonably facilitates dynamic internal routing to connect to an IGP. By separating the outside and within
This is more convenient to be more convenient.
Output EGP Updates is prepared for internal routing tables via rt_nrnets (), whether it is static or dynamic
Ministry route.
The function also provides inspection, increase, change, and deletes internal routing, I.E., that is, rt_int_lookup (), RT_ADD (),
RT_Change () and rt_delete ().
One IgP is interactive with the current data structure, basically includes three functions: use a similar to ... -...
The function updates the internal routing table, and the output internal update is similar to rt_nrnets (), timeout interior routes is similar to
RT_TIME).
⒌ ⒌ ⒌ 问题
5.1 Topology Constraints and Route Background
5.1.1 background
EGP is not a path selection algorithm. It only enables external neighbors to exchange routing information, it is likely to pass
A path selection algorithm. It can't pass full information to prevent routing, if the loop is in the topology
[ROSEN 82].
The routing circuit may occur when the two gateways believe that when there is a candidate route that reaches one third gateway. Third network
When shutting down, they ended each other, forming a routing route loop. The current core system, the loop passes
From zero to unlimited (international internet gateway station diameter) release. This (usual) work is satisfactory because of GGP
The change is very rapid, and once the change occurs, the route selection update is sent.
The international internet network diameter is quite small (5) and uses a general distance metric, number of segments. But this may have
Change.
With EGP, change is slowly transmitted. Although a single unsolicited NR message may be sent, it does not have to directly
Passing to other gateways that need indirect listening. Also, the distance metric from different AS is completely independent, so from zero
Number to infinity cannot always apply.
The initial proposal will become a tree structure by limiting the topology of the AS so that there is no multi-routed by candidate AS.
From the way. The internally routed by the same AS is allowed, and if the internal routing policy is responsible for the control loop.
[Mills 84B] Note that even if the tree is constrained, "We must assume a short circuit sometimes in the core system.
Forming and this information may escape to other systems; however, people expect that this loop will not continue to last for a long time and at the core
The interior will not be released soon. So a non-core system circuit may continue until the first update newspaper is sent to it.
After his system, all loop detections have been cleared or until the reachability information page is timeout, no matter that one appears.
With the initial simple Stub EGP Systems, the tree configuration constraint can be used. But for a long time, this does not provide sufficient strength.
[Mills 83] It is recommended that the method AS can dynamically reconfigure themselves so that the topology constraint is always full.
Foot, there is no need for a single "Core" AS. An AS has a shared network and its neighbor AS is transmitted by himself.
Guide EGP. The owner passes the above information indirectly, and the core system is now doing. If you have AS being defined to get close
In this tree, all irregular interconnections can form a suitable tree structure topology. Topology through routing
Understanding topology notified in routing selection. There are other physical connections, but if they are not reported
Will not be used for routing. Each AS can only use an AS conduction EGP with a shared network. All do not pass another
The AS owned network conduction EGP AS is the root of the subtree. It may use other AS passions with a shared network
Guide EGP. This "attachment" combines two sub-trees into a single child, so the entire topology is still a tree. Topology violation
It can be judged because two different AS topologies will report they can reach the same network.
The above dynamic tree may have a preferred backup connection. In this case, the monitor must be a failed connection so that the route can be
Change back to a preferred connection and when the service is recovered.
On the other hand, it is considered that the possibility of detecting the routing circuit is then interrupted. Package Survival Time (TTL) expires for this
thing. If a diagnostic package suspects has a loop, such as ICMP Echo, you can send suspicious routing
Really it is a loop. If you detect a loop, you can send a special routing package through the route, which means
Remove the route after each gateway forwards the package. Acceptance of new route selection information may need to be postponed for a downtime
Time. This method needs to make wise choices for the initial TTL. But many hosts don't do this.
5.1.2 Current Policy
Given that the general trend is to increase the network interconnect and replace Long - Haul NetWorks such as Arpanet, WBNET (Broadband Guard
Star Networks) and the validity of the public data network, the tree topology constraint is usually unpopular. The current recommendation is limited
Less topology.
Here, you can take the [MILLS 84B].
EGP Topology Model:
- A independent system contains a set of gateways connected via a network. Each gateway in the system is for its system
Each other gateway in the middle must be achieved, the path is only included in the gateway in the system.
- One gateway in the system can run EGP with the gateway in any other system, as long as the pass
The path to the EGP does not contain a gateway that runs EGP in the third system.
- "Core System" is different from other core systems, because it is only allowed to distribute the system instead of its own
Real information.
- At least one gateway in each system must have a network like a gateway in the core system.
- In addition to the above implications, there is no topology or connectivity constraints.
A gateway will use it from its configuration (direct connection network), its system's IGP, called S, (within
Ministry Network) to construct its routing table with EGP (neighbor system of internal and external network). Information if with a special network
In addition to the conflict, they will be solved as follows:
If n is directly connected to the gateway, all IGPs and EGP reports about N are ignored.
If N is reported inside the IGP report, the EGP report is an external or inside of another system, and the IGP report is reported.
Prioritize.
- If N is not in the interior of the EGP as a system and the other system, the internal Report dominates.
- IF N is used as an internal, and the minimum number of stations specified by the report accounts for the main number of the same system of EGP.
Position.
- In all other situations, recently received reports dominate.
Old information may come from the outdated information of the table.
This temporary model provides an acceptable degree of organization.
A short routing circuit may appear between the system, but these finally discontinued by the old accessibility information, because of this information
Outtime from the list. A short circuit may appear due to the temporary core system loop, because the local network is in multiple lines
The additional loop may have a look that it does not increase much hazard.
5.2 Current ISI Configuring a streamlined version of ISI network configuration is displayed in Chart 5 - 1.
ISI - HobGoblin can provide a backup gateway work for the core ISI - gateway between Arpanet and ISI-Net
can. ISI - HobGoblin is a VAX 11/750 running Berkeley Unix 4.2. This description is in these reports EGP
Implement running on ISI - Hobgoblin.
ISI - Troll belongs to an ancient Biological Museum of California on an Irvine Network (UCI - ICS).
A complete logic gateway contains ISI-TROLL, the 9600 baud connection and UCI - 750A [ROSE 84]. ISI TROLL operation
Berkeley UNIX 4.1a This cannot be running the EGP program. So it is a non-rouled gateway. UCI - ICS network
There is a need to be notified to the core AS by ISI-HobGoblin. Contains an appropriate in EgpinitFile
The login can be done.
In ISI - NET, the host contains ISI-Troll, with static route login to point out ISI - GATEWAY as an except for UCI -
The first site of all the networks of ICS and ISI - NET.
-------------------------------------------------
/ /
/ Arpanet /
/ 10 /
/ /
-------------------------------------------------
| | | | |
| | | | |
| | | | |
----------- -------------------------------
| ISI-PNG11 | | | | | |
| ARPANET | | Isi-Gateway | | Isi-Hobgoblin |
| Address | | | | VAX 11/750 |
| Logical | | Core EGP | | UNIX 4.2 |
| MULTIPLEXER | | | | | |
----------- -------------------------------
| | | | |
| | | | |
| | | | |
------------------------------------------
/ / / /
/ 3 MB / S Ethernet / / ISI-NET / / NET 10 / / 128.9 /
/ / / /
------------------------------------------
|
|
|
------------
| ISI-TROLL |
| VAX 11/750 |
| UNIX 4.1A |
| Non-routing |
| | | | |
| | 9600 | ISI-TROLL, UCI-750A
| | BAUD | And the link Form A
| | LINK | Single Logical Gateway
| | | | |
| UCI-750A |
| VAX 11/750 |
| UNIX 4.2 |
------------
|
|
|
----------------------
/ /
/ UCI-ICS /
/ 192.5.19 /
/ /
----------------------
Figure 5-1: Simplified ISI Network Configuration, SIMPLIGIED ISI NetWork Configuration
EGP or ISI - Gateway or ISI - NET croses ArpaNet conduction.
5.2.1 EGP Crossing Apa
ISI - Hobgoblin is about to notify ISI - Gateway crossing Arpanet, where this core system can reach ISI - NET
And UCI - ICS.
From the outside of AS to ISI and bags to UCI - ICS may pass ISI Gateway, ISI - Hobgoblin and ISI -
Troll route. Additional route via ISI - Gateway (or other core EGP gateway) is because the core gateway is currently
Do not transfer the address in their IGP packets outside the external gateway
The gateway goes to the network connection protocol). Packages originated from UCI-ICS to external AS may pass ISI - Troll and ISI
Gateway routing. Therefore, the input and output packet paths are different.
The package originated from the ISI-Hobgoblin that is a host to the external AS may pass the appropriate gateway path on Apa.
by.
If ISI - TROLL, ISI - Hobgoblin and ISI - Gateway are all powered, UCI - ICS has to communicate with external AS.
Relying on ISI - Gateway may be eliminated, IF ISI - Troll By ISI - Hobgoblin routing in ISI - GATEWAY.
However, because ISI - HobGoblin is mainly a host instead of a gateway, it would rather ISI - GATEWAY
Routing package.
ISI - Hobgoblin may provide a spare gateway feature to ISI - Gateway because it may automatically transfer to a replacement
Core EGP peer, if ISI - Gateway Goes Down is stopped.
Even if ISI-Hobgoblin has normally accessed the core system, it may reach ISI - Net, because the core passes ISI - GATEWAY
Priority uses its internal ownership.
In order to route the output package correctly on the ISI-NET, they need static from ISI - Gateway to ISI -
Hobgoblin gateway login changes. These people are currently completed. If ISI - Gateway will stop
This is appropriate when it is time. 5.2.2 EGP Crossing ISI - NET
ISI - Hobgoblin is about to notify ISI - Gateway crossing its indirect neighboring ISI NET, ISI - TROLL, thus this core
Universal can reach ISI - NET and UCI - ICS.
All UCI - ICS external packet routing will pass the two-way no station number ISI - Gateway through ISI - Hobgoblin.
Organizing the package as a host ISI hobgoblin to the external AS will pass by ISI - GATEWAY two-way route, not
The Arpanet interface is therefore handled an additional route.
UCI - ICS has to communicate with Exterior AS, if ISI - Troll and ISI - Gateway is running and ISI -
Hobgoblin has reported UCI - ICS ISI - Gateway route. If ISI - HobGoblin is downtown, communication is still
It is possible for ISI-Gateway (and other core gateways) to travel to other neighbors.
If ISI - Gateway then stops, when it appears, it needs to pass UCI - ICS Route ISI - HobGoblin
New read.
Through ISI - NET rather than ARPANET conduction EGP will provide a more reliable service for UCI-ICS because the following
By :isi - Gateway is explicitly designed as a gateway, its runtime should be more than ISI - hobgoblin, remove
The additional routing site is desirable, and the external route selection information will continue after ISI-Hobgoblin is shut down. in case
ISI - HobGoblin is used for backup mode, and the new gateway route is installed after the host is installed in the host.
ARPANET restarts.
Therefore, EGP is classified by ISI-Net is selected as a preferred mode of operation.
5.2.3 Potential Routing
Because ISI - Gateway and ISI - Hobgoblin two jointly provided routes between Arpanet and ISI Net,
So there is a potential routing circuit. These topologies in fact violates the initial tree structure constraints. ISI provided
Hobgoblin is different from ISI - Gateway to conduct EGP through ISI NET and ARPANET, the gateway even knows
Self-sharing EGP network replacement routes, unknown from other networks. So a loop cannot appear. Lift
For example, if EGP is conducted by ISI - NET, ISI - Gateway and ISI - Hobgoblin will know from ISI from ISI
Net to Arpanet replacement, but they don't know the address of the gateway on Arpanet, so you can't
Enter ISI - NET from Arpanet. So they don't have sufficient routing data to route themselves.
5.3 Possible future configuration
5.3.1 Going to UCI - ICS Gateway
The services provided by the UCI ICS in terms of security and performance can be moved from ISI Troll to ISI -
HobGoblin's UCI ICS interface is completed. Safety will improve because the connection only requires ISI - Hobgoblin and its
The Arpanet interface is running; performance will improve because the additional gateway path segment may be removed.
These also will allow EGPs to pass the Arpanet to provide access to replacement core gateways running EGP. These ones
The increase is always reliably obtaining an opportunity to get an EGP neighbor. It will also serve from HobGoblin as a host
The bags to the external network are removed in addition to these additional routes.
These configuration changes will be completed in the future. I didn't do this, because ISI - hobgoblin is experimental
Moreover, the shutdown is more frequent than ISI-TROLL.
5.3.2 Dynamic to Backup Gateway
ISI - HobGoblin can provide a backup gateway work for the core ISI - gateway between Arpanet and ISI-Net
Can, this is in Section 5.2.1 to indicate. The above backup gateway can become a common method to provide greater security.
Current change in the backup gateway is required to join the new gateway route to the host of ISI - NET. This section describes the possible ways used in dynamically completing these conversions when the original network is shut down.
The goal is to detect when the original gateway is stopped, and the minimum additional network traffic will be on the part of the local network.
Convert to backup gateways. When it appears, the host should be restored as the original gateway.
The method of initiative is to monitor the original gateway status for the backup gateway. When there is a change, it will notify it all new networks.
Site host.
5.3.2.1 Commonly used operation
The backup gateway runs a process, which sends a reachability-probe packet every 30 seconds, such as ICMP ECHOES, to the original gateway and
Use this response to determine the EGP reach. If the original gateway goes Down
Point out the "gateway - address information" of the backup gateway address is broadcast
(More available multi-point transfer) to all hosts. When the original gateway appears another point to the original gateway address gateway message
When it is broadcast.
These broadcasts should be carried out four times in 30 seconds, in order to avoid confirmation and host address of the host address.
Each host will run a process, listening to gateway - address information. If a different gateway is suggested for new addresses
Default network access.
5.3.2.2 Host initialization
When a host appears, the original gateway may stop, so it needs to be able to decide whether it will use the backup gateway.
The host may read the original and backup gateway address from a static initialization file.
It will then set its default gateway as the original gateway and send one
"Gateway - Request Message" gives the backup gateway request the current gateway address. The backup gateway will respond to a gateway-address information.
If the no answer is received, the gateway - request should be repeated three times within 30 seconds. If there is no answer,
The backup gateway is assumed to stop and the original gateway remains as the default gateway.
Whenever the backup gateway appears, it broadcasts with a gateway-address information response.
In other words, a broadcast (or multi-point transfer) gateway - request message
The defined is only responded only by the gateway. Backup Gateway - address information needs to indicate that it is a backup gateway so that the future request does not have to be wide.
broadcast. Three retransmission should be used. However, the original gateway also needs to broadcast its address, whenever it appears.
5.3.2.3 When is the original and backup of both
If the original network is shut down and the backup knows it stops, it will broadcast gateway - address information points to the original gateway, in case
The start gateway appears.
In addition to the backup may not warn the shutdown, and the original appears before it. If the original gateway is broadcast when it appears
Gateway - address information is impossible. Otherwise, when the host uses a backup gateway, they should send every 10 minutes.
A gateway - request message. If the no answer is received, it should be repeated 3 times in 30 seconds and if still
No answer This backup is assumed to stop and the original gateway is restored.
Therefore, when the original gateway does not send gateway-address information when it is in appearing and is being used, the host needs to send messages periodically.
Sometimes, such as in ISI, the original gateway has been managed by different organizational management and the experimental characteristics cannot be increased at any time.
5.3.2.4 Unix 4.2 BSD
A difficulty is that there is no standard method that specifies the Internet broadcast or multi-point transmission address. Multi-point transmission address is more desirable,
Because only they need to handle the message (with hardware multi-point transmission detection interfaces). As for UNIX 4.2 BSD
The Internet address has a zero hook that is specified as an Internet broadcast address.
However, the ordinary Internet address policy is to use a full 1 to point out a broadcast function.
On UNIX 4.2 BSD Systems, both gateways and host processes may run in user-level so that they do not require the modification of the kernel.
User Data News Protocol (UDP) Set of words that may have a host-backup-gateway communication.
Super users to send and receive ICMP Echo packets to access the RAW setup requires a slightly modification of the International Internet-Tria Protocol
Convert table.
Acknowledge
I would like to thank all the people who help this project, especially: the proposed project Dave Mills, Jon
Postel discussion and encouragement, Liza Martin provides initial EGP code, Berkeley provides "ROUTED code, Mike Brescia Assist Test, Telecom Australia provides me with funds and ISI providers.
⒎ Reference
[Berkeley 83] "Unix Programmer's Manual", Vol. 1, 4.2 Berkeley Software
Distribution, University of California, Berkeley.
[Kirton 84] Kirton, P.a., "EGP GATEWAY Under Berkeley Unix 4.2", University
Of Southern California, Information Sciences Institute,
Research Report ISI / RR-84-145, TO BE PUBLISHED.
[Mills 83] Mills, D.L., "EGP MODELS AND SELF-Organizing Systems" Message
To EGP-people @ BBN-UNIX, Nov. 1983.
[Mills 84a] Mills, D.L., "Exterior Gateway Protocol Formal Specification",
NetWork Information Center RFC 904, April 1984.
[MILLS 84B] Mills, D.L., "Revised Egp Model Clarified and Discussed",
Message to egp-people @ bbn-unix, may 1984.
[Postel 84] Postel, J., "Exterior Gateway Protocol Implementation Schedule"
NetWork Information Center RFC 890, Feb. 1984.
[Rose 84] Rose, M.T., "Low-Tech Connection INTO THE ARPA-Internet: the
Raw-Packet Split Gateway ", Department of Information and Download and INFORMATION AND
Computer Science, University of California, Irvine, Technical
Report 216, Feb. 1984.
[Rosen 82] Rosen, E.c., "Exterior Gateway Protocol", Network Information
Center RFC 827, OCT. 1982.
[Seamonson & Rosen 84]
Seamonson, L.J. And Rosen, E.c., "Stub EXTERG GATEWAY
Protocol ", Network Information Center RFC 888, JAN. 84.
[Xerox 81] "Internet Transport Protocols", Xerox System Integration
STANDARD XSIS 028112, DEC. 1981.
RFC911 - EGP Gateway Under Berkeley Unix 4.2 Berkeley UNIX 4.2 EGP Gateway
1
RFC Document Chinese Translation Program
