Embodiment
As shown in Figure 1, the course of work of the cluster method between packet data serving node described in the invention is such:
At first, PDSN selects a cluster group that will add and adds this group, implements cluster strategy, exchange cluster information etc. later on and all will carry out in this group;
Secondly, the cluster information of each PDSN timing or not timing announcement oneself in the group, comprise the load-bearing capacity of oneself, current load condition, insert and the MS of off-line or the like in the recent period that PDSN is recorded in these information that other PDSN send in " cluster PDSN table " and " cluster MS table ".Wherein, the effect of " cluster PDSN table " is the load condition information of all PDSN in the record clustering group, from this table, can find out the lightest PDSN of load, this table is very useful when load balancing, the effect of " cluster MS table " is the basic condition that record is linked into all MS in this cluster group, and this table is very useful when Intelligence Selection;
When a MS is linked among the PDSN, PDSN at first inquires about cluster MS table, judge that this MS is that other PDSN switch the user or the new user who inserts that come in this cluster group, if in table, found this MS, illustrate that so this MS is that other PDSN switch the users of coming in this cluster group, otherwise illustrate that this MS is the new user who inserts;
Switch the user of coming for other PDSN in this cluster group, PDSN will recommend back the PDSN of access just now to this MS, avoid MS to produce unnecessary switching, and this strategy just is Intelligence Selection;
User for new access, PDSN will inquire about cluster PDSN table, from cluster PDSN table, find out a lightest PDSN of load, if the load of oneself is at present the lightest, PDSN will accept the access of this MS so, otherwise will recommend the lightest PDSN of load to the MS of this new access, this strategy just is load balancing, claims load sharing again.
The algorithm that PDSN butt joint access customer carries out Intelligence Selection or load balancing as shown in Figure 3, we will introduce this algorithm in detail the back.In the cluster group, the swarm algorithm of each PDSN is identical, and operation independent separately, has taken place also can not have influence on other PDSN unusually even participate in certain PDSN of cluster.
When MS is linked into certain PDSN, through after the Intelligence Selection or load balancing computing of this PDSN,, MS more suitably talks about if being linked into other PDSN, and this PDSN need recommend other more suitably PDSN to this MS so, and recommend method is as shown in Figure 2.PDSN is when recommending other PDSN to MS, in addition not newly-increased interface, but utilized A11 registration reply message (A11Registration RePly in the R-P interface of standard dexterously, abbreviation A11-RRP) (error code equals 136 in error code expression, implication is the IP address of unknown PDSN), do not need PCF to revise any interface in the time of can accomplishing like this to implement PDSN cluster strategy.
When the cluster that Fig. 2 describes, recommend in the A11 registration process of new PDSN, after having set up airlink between MS and the wireless network, PCF will send A11 register requirement (A11 RegistrationReQuest to PDSN_1, be called for short A11-RRQ), indicated home agent (Home Agent among the A11-RRQ, be called for short HA) equal the IP address of PDSN_1, PDSN will carry out load balancing and Intelligence Selection according to previously described strategy, if desired MS is recommended other PDSN in the group, PDSN_1 will send an A11-RRP to PCF so, the refusal code that comprises in the message equals 136, the best PDSN that provides access simultaneously is PDSN_2 (HA=2.1.1.1), PCF will send a new A11-RRQ to PDSN_2 again according to this recommendation results subsequently, wait PDSN_2 to respond the A11-RRP message after, the A11 passage is just set up and has been finished.
Fig. 3 (a) has provided the process chart of PDSN load balancing in the inventive method; Fig. 3 (b) provides the process chart of PDSN Intelligence Selection in the inventive method.
In the middle of practice, when a MS is linked into PDSN, what PDSN at first carried out will be Intelligence Selection, just carry out load balancing when Intelligence Selection is unsuccessful, introduce load balancing and Intelligence Selection respectively below in conjunction with Fig. 3.
The purpose of implementing load balancing in cluster group is to avoid some PDSN overload of power, and the too light situation of some PDSN load takes place, and utilizes Internet resources fully, for MS provides reliablely and stablely access service.
In conjunction with Fig. 3 (a), the work thinking of load balancing is as follows:
1. cluster PDSN is regularly to the Hostname of external world's notice (multicast and clean culture) oneself, IP address, present load condition, the information such as load-bearing capacity of oneself;
2. calculate the load percentage of each cluster PDSN, load percentage=present load/total load-bearing capacity;
When PCF be that the newly-built R-P of the new MS of access is when connecting, if find to have the load percentage of PDSN lower more than 10% than the load percentage of oneself, this PDSN just can recommend the minimum PDSN of that load percentage to MS so, otherwise just oneself admits the user of this new access;
4. for fear of the situation generation of overload suddenly of certain PDSN load, cluster PDSN also can be to the load condition of extraneous not timing ground proactive notification oneself;
5. if PDSN has a plurality of R-P network interfaces, cluster PDSN also can carry out load balancing between the several R-P network interfaces of oneself, promptly accept the request of setting up A10/A11 of PCF, but recommends the lighter R-P network interface of another one load simultaneously.
The purpose of Intelligence Selection is to avoid MS to produce unnecessary switching in cluster group, guarantees that the communication of MS is not interrupted.For example, PDSN1, PDSN2, PCF1, PCF2 totally 4 network elements are arranged, wherein PCF1 preferably is connected to PDSN1, and PCF2 preferably is connected to PDSN2, and PDSN1 and PDSN2 are in a cluster grouping.Original MS is linked into PDSN1 by PCF1, MS switched to PCF2 afterwards, but PCF2 has preferentially selected PDSN2, because PDSN1 and PDSN2 are same cluster groupings, so this moment, MS should be PDSN1 by the optimum PDSN that PCF2 connects, because MS can carry out PPP, mobile IP (Mobile IP more like this, be called for short MIP) etc. the reconstruction of session, can accelerate the switch speed of MS like this,, can also guarantee that like this communication of MS is not interrupted if the MS employing is the simple IP access way.
In conjunction with Fig. 3 (b), the work thinking of Intelligence Selection is as follows:
1. the PDSN of all participation clusters writes down the IMSI of interior all accesses of a period of time and off-line MS;
When timer to constantly just in group other PDSN notice insert recently and the IMSI of off-line MS, if when the MS of access or off-line reaches some, PDSN also can initiatively send to not timing this information;
3. each PDSN MS access information that other clusters PDSN is sent and the MS that is linked into oneself are recorded in the cluster MS table; All will inquire about this cluster MS table when having new MS to insert, if that find that MS inserted just now is other PDSN in the cluster PDSN, PDSN will recommend back old access PDSN to MS again so.
If the new MS that inserts does not appear in the cluster MS table, illustrate that this PDSN only need consider that load balancing is just passable, need not to consider Intelligence Selection.If the current load of PDSN has not surpassed the thresholding of cluster, PDSN will accept the A11-RRQ that PCF sends so, otherwise will inquire about cluster PDSN table, therefrom select a lightest PDSN of load, the possibility of result that chooses is oneself, also may be other clusters PDSN.Here the load balancing of carrying is not only the load balancing between PDSN, also comprises the load balancing between several R-P network interfaces of PDSN.After receiving the A11-RRQ message of PCF, when the load of several R-P network interfaces of PDSN discovery oneself is unbalanced, can accept the A11-RRQ of PCF, but in response message, recommend the R-P network interface that load is lighter simultaneously, will send to the new R-P network interface that PDSN recommends when PCF can transmit the A10 message afterwards.If the new MS that inserts has appeared in the cluster MS table, just need to have considered Intelligence Selection that because MS should be recommended back the PDSN that inserted just now this moment, MS need not to rebuild the PPP connection of previous foundation and MIP connection or the like like this this moment.If that previous access is PDSN oneself, the switching that MS has taken place between PCF, PDSN is interior is described so, PDSN will accept the A11 register requirement that PCF sends.If that previous MS inserts is other clusters PDSN, and this PDSN is working properly, PDSN will recommend back the previous PDSN that inserts with MS according to cluster IMSI table so.
After travelling carriage is linked into PDSN, PDSN will revise cluster MS table, simultaneously mobile station information be preserved, and in suitable the IMSI of the travelling carriage that inserts be announced to other clusters PDSN.When being linked into the travelling carriage off-line of this PDSN, PDSN also needs the information of travelling carriage is preserved, and in suitable the IMSI of the travelling carriage of recent off-line is also announced to other PDSN.
The present invention distinguishes different cluster groupings by different PORT COM, not communication mutually between the PDSN of different grouping, and the cluster of PDSN only carries out in this grouping, and a PDSN can not participate in two cluster groupings simultaneously.In addition, also provide two kinds of trunking modes: next-door neighbour's cluster and strange land cluster according to the difference of the position of the PDSN that participates in cluster.
So-called next-door neighbour's cluster is meant all PDSN that participate in cluster in a physical subnets, and in general, the PDSN of next-door neighbour's cluster is usually in a machine room.So-called strange land cluster is meant that the PDSN that participates in cluster can be from different local, and their IP address belongs to different physical subnets, each other a plurality of routers, switch etc. can be arranged.
Cluster between the PDSN adopts the bearing protocol of UDP as communication, next-door neighbour's cluster adopts the communication of multicast mode: IP=224.0.0.224, Port=10224+ (nGroup-1) * 10, the strange land cluster adopts the mode of unicast communication: the address of IP=opposite end PDSN, Port=10225+ (nGroup-1) * 10, wherein nGroup is the cluster packet numbering, and scope is generally 1~16.
The rudimentary algorithm of cluster has been described in the front, describes the cluster message that is used to exchange cluster information below again, and the reciprocal process of these cluster messages.
The PDSN cluster message that the present invention relates to mainly contains 8 classes, as shown in Table 1.In order to guarantee the fail safe of cluster message, the PDSN of all participation clusters of the present invention all disposes the key that identical being used to authenticates the cluster message, and identifying algorithm can adopt the HMAC-MD5 algorithm.The type of message and the explanation thereof of table one PDSN cluster
Type of message | The message function declaration |
Hello packet | Other PDSN in cluster group announce this PDSN and have started and begun to join in the cluster group. |
Announcement load-bearing capacity message | The load-bearing capacity of other PDSN announcements in cluster group oneself. |
Announcement inserts, the MS message of off-line | Other PDSN announcement in cluster group inserts in this PDSN and the user profile of off-line. |
Announce online message | Other PDSN announcements in cluster group are own online. |
The PDSN message of renaming | When PDSN changes the host name of oneself, by the newname of this message announcement oneself. |
Stop the cluster message | When PDSN need leave certain cluster group, by other PDSN of this message notifying. |
Ask for the cluster information request message | Initiatively other PDSN in cluster group ask for all information relevant with cluster of the other side. |
Ask for the cluster information response message | Ask for the response message of cluster information request message. |
Tell about the reciprocal process of cluster message below in conjunction with Fig. 4, Fig. 5, Fig. 6.
What Fig. 4 described is the transmission situation of message when PDSN has just started under the normal condition.
When PDSN just started, it also can't obtain the cluster information of other clusters PDSN, and this moment, PDSN will be initiatively to external world's issue hello packet, and the number of times of issue is generally continuous 3 times.Other PDSN produce a time delay (0~3 second) at random after receiving this hello packet, and then have inserted the information of MS to the load-bearing capacity of the PDSN of new startup announcement oneself and all on one's own initiative, and communication modes generally adopts the mode of clean culture.Why PDSN has a time delay when answering hello packet, mainly be for fear of the situation generation that all cluster PDSN respond hello packet simultaneously occurring.
Subsequently, PDSN of this firm startup will start 3 timers, be respectively the timer (t2) of the announcement online message timer of PDSN (t1), announcement PDSN load-bearing capacity message and the MS message timer (t3) of recent access of announcement and off-line.All PDSN in the cluster group all need the online message of timed sending oneself, each PDSN can judge which PDSN is working properly according to this message, which PDSN has lost contact, just can to judge that this cluster PDSN has worked undesired if (be generally 30 seconds) the online message of not receiving certain cluster PDSN in a period of time, and all later on Intelligence Selection and load balancing will foreclose the PDSN of this operation irregularity.If (being generally 20 minutes) this unusual PDSN does not also recover operate as normal in the longer time, all cluster informations of this unusual PDSN are with deleted, comprising all records that are linked into this PDSN in record in cluster PDSN table and the cluster MS table so.PDSN periodically outwards announces own present load condition, when carrying out load balancing, that preferentially select will be the lightest PDSN of cluster group internal loading, if certain PDSN finds that the load growth of oneself is too fast, it also can outwards issue the load condition message of oneself on one's own initiative, and will wait until that not necessarily timer arrives constantly just issue.When MS access or off-line were arranged, PDSN all can note the IMSI of these MS, and the MS message timer that waits announcement access in the recent period and off-line is to these IMSI being announced to other clusters PDSN constantly again.
The message interaction order under the normal condition has been described in the front, may also some abnormal conditions can occur in the middle of the practice process, below we discuss how to implement the mutual of PDSN cluster message again under abnormal conditions.
What Fig. 5 described is a kind of abnormal conditions, after (for example network communication is unusual) PDSN_1, PDSN_2 start a period of time promptly for a certain reason, all do not receive the other side's messages such as Hello, that is to say that PDSN_1 and PDSN_2 all do not know the other side's existence, although they are operated in the same cluster grouping.PDSN_1 has received any cluster message of PDSN_2 after a period of time, PDSN_1 does not find the relevant information of PDSN_2 in cluster PDSN table, PDSN_1 will initiatively send one to PDSN_2 and ask for the cluster information request message so, PDSN_2 will send one earlier to PDSN_1 and ask for the cluster information response message after receiving this message, PDSN_2 also can send announcement load-bearing capacity message and announce all and inserted the message of MS subsequently, hereafter, the cluster of PDSN_1 and PDSN_2 just recovers normal.
What Fig. 6 described is another kind of abnormal conditions, be that initial PDSN_1, PDSN_2 have all added a cluster group, and all set up normal trunking communication between them, PDSN_2 was restarted after occurring unusually afterwards, and outwards issue a hello packet, PDSN_1 is after receiving hello packet, all cluster informations of PDSN_2 will be removed, comprise that all are linked into MS information all among the PDSN_2 in the cluster MS table, and then the PDSN of PDSN_2, and to PDSN_2 transmission announcement load-bearing capacity message and announcement MS access information message as new startup.