5 Performance Evaluation

Transcription

1 5 Performance Evalation his chapter evalates the performance of the compared to the MIP, and FMIP individal performances. We stdy the packet loss and the latency to restore the downstream and pstream of packets dring intra-domain and inter-domain handover time. We consider an hipotetical network environment with five different domains interconnected by a backbone. One of the domains is restricted to the stationary C and the others are visited by the M dring a session connection time. Each domain visited by the M is composed of a border roter (MAP entity and three ARs connected to it. Each AR is connected to an IEEE b access point (AP that provides wireless network access to the M. All APs are placed in geographical positions that provide the necessary intersection between cells to make seamless handover possible. Figre 18 illstrates the network environment scenario sed for the evalation. LRS5 ER5 AR5 C LRS1/HRS MAP1 LRS2 MAP2 LRS3 MAP3 LRS4 MAP4 AR11 AR12 AR13 AR21 AR22 AR23 AR31 AR32 AR33 AR41 AR42 AR43 M M Figre 18 etwork environment scenario sed for the evalation We also assme that the Media Independent Event Service (MIES is available and ready to notify the pper layers abot the L2-GoingDown, L2-

2 Performance Evalation 60 Down and L2-UP triggers. So, the M doesn t need to wait for the Roter Advertisement message of the new AR to detect an L3 handover. he M will initiate the L3 handover when the L2-GoingDown trigger is detected. Once notified abot the L2-UP trigger, the M will se the stateless ato-configration mechanism to avoid contacting any entity to obtain a new co-located CoA. hese measres contribte to redce the handover process dration. In order to analyze the performance of the mobility management of the protocols, we need to model the transmission delay of the signaling messages. Depending on the protocol, the signaling messages go throgh wireless links and wired links or jst throgh wired links. Also, some messages are local to a domain and others are originated at a domain and forwarded by intermediary hops to another domain. able 5 introdces the parameters that will be sed in the analysis. able 5 Parameters sed in the analysis Parameters s r Average session connection time Average cell resident time m Average nmber of movements dring a session (i.e., m = s / r S D x-y B w L w B wl L wl P t R d R L2 Average size of a signaling message Average nmber of hops between x and y Bandwidth of the wired link Latency of the wired link (propagation delay and link layer delay Bandwidth of the wireless link Latency of the wireless link (propagation delay and link layer delay Roting table lookp and processing delay Downstream transmission rate (packet transmission rate Upstream transmission rate (packet transmission rate ime interval for L2-handover (starts at L2-Down and ends at L2-UP

3 Performance Evalation 61 Let (S,D x-y denote the transmission delay of a message of size S sent from x (always an M to y via the wireless and wired links, as sggested by [16]. (S,D x-y can be expressed as follows: S S, Dx y Lwl Dx y Lw + B + + wl B = + 1 w ( Dx y + Pt Let (S,D x-y denote the transmission delay of a message of size S sent from x (never an M to y via the wired links. (S,D x-y can be expressed as follows:, Dx y = ( Dx y + 1 S B w + L w + Pt 5.1 Packet loss dring a session Packet loss (Pkt_Loss dring a session is defined as the sm of packets lost dring all handovers procedres while the M is receiving or sending data packets. o redce packet loss, some bffering mechanisms may be sed by both the M and the AR to store in-flight packets, bt these mechanisms sally reqire more signaling messages to be added to the protocols. hen, not all of the protocols spport bffering control of packets dring handover time. his is the case of both the MIP and the. he FMIP and the se signaling messages to enable the bffering and forwarding of in-flight packets from the PAR to the AR. We will denote the bffer size reqirement for downstream and pstream flows as a fnction of the packet transmission rate in section 5.4. For now, we assme that the size of the bffer is appropriate to store the in-flight packets. We also assme that the M is able to store all of its pstream packets and no packets are lost in this direction. So, to denote the total packet loss for the MIP and, we consider that in-flight packets are lost dring the whole L2-handover and L3-handover. he L2- handover starts when the L2-Down trigger is detected and ends when the L2-UP trigger is detected. he L3-handover starts right after the L2-handover is completed and is composed of procedres that depend on the protocol. For the MIP,

4 Performance Evalation 62 these procedres are: (i discovering of the network prefix (Roter Solicitation/Roter Advertisement; (ii registration at the HA; and (iii establishment of the bi-directional tnnel between the M and the HA. For the, the procedres depend on the type of L3-handover. For the intra-domain handover, the procedres are: (i discovering of the network prefix (Roter Solicitation/Roter Advertisement; (ii registration at the MAP; and (iii establishment of the bidirectional tnnel between the M and the MAP. For the inter-domain handover, besides the procedres of the intra-domain handover, more two procedres are exected by the M: (i registration at the HA; and (ii establishment of the bidirectional tnnel between the HA and the MAP. he total packet loss for the MIP and can be expressed as follows: Pkt _ Loss( MIP = [ L2 + ( M AR M HA ] R d m Pkt _ Loss( Intra = [ L2 + ( M AR M MAP ] R d m Pkt _ Loss( Inter = [ L2 + ( (3 + 3 M HA HA MAP M AR M MAP ] R d In the case of the FMIP and the, in-flight packets are lost till the bffering and forwarding mechanisms are initiated. he initialization of these mechanisms is exected as part of the L3-handover, based on the mode of operation of the protocols. Usally, the L2-GoingDown trigger is sed to initiate the L3- handover before the L2-handover. When the icipated mode is sed, the bffering and forwarding mechanisms are initiated before the L2-handover. Otherwise, in the ive mode, these mechanisms are initiated after the L2-handover. For the icipated mode of the FMIP, these mechanisms are initiated when the FBU message, sent by the M, is received by the PAR before the L2- handover. So, packets will be lost only if the M is not able to send the FBU message before the L2-handover and, then, it will need to switch to operate in ive mode. In this circmstance, after the L2-handover, the M sends the FA message (encapslating the FBU to the AR over the wireless link and the AR m

5 Performance Evalation 63 sends the FBU message to the PAR over the wired links to initiate the forwarding mechanism. So, the total packet loss for the FMIP can be expressed as follows: Pkt _ Loss( FMIP Pkt _ Loss( FMIP = 0 = [ L2 + AR PAR M AR ] R d m he initialization of the bffering and forwarding mechanisms of the is very similar to the procedre exected by the FMIP. For the icipated mode of the, the bffering mechanism is initiated when the C2_HandoverInitiate message, sent by the M, is received by the PAR before the L2-handover starts (L2-Down trigger is fired. So, the packets are lost only if the M is not able to send the C2_HandoverInitiate message before the L2-handover starts. Under this circmstance, like FMIP does, the switches to operate in ive mode. his mode starts after the L2 handover is complete, when the M becomes able to send the C2_HandoverFinish message to the AR over the wireless link. In trn, the AR sends the 2_HandoverIndication message to the new LRS (LRS which relays the message to the previos LRS (PLRS. he PLRS relays the same message to the PAR which finally sends the 2_Handover Acknowledge message to the AR. o complete this phase, the AR also exchanges a pair of messages with the PAR over the wired links to reqest the packet forwarding. So, the total packet loss for the can be expressed as follows: Pkt _ Loss( Pkt _ Loss( = 0 = [ L LRS PLRS AR PAR M AR ] R d PLRS PAR m AR LRS 5.2 Average handover latency to restore the downstream flow dring a session Average handover latency to restore the downstream flow (Avg_Lat_Down dring a session is defined as the average time of all handovers of a session to restore the M s ability to receive packets of the ongoing downstream. o redce handover latency, some efforts were made to enhance the L2-handover and the L3-handover. Here, we focs on L3-handover analysis and assme a const vale L2 to express the L2-handover latency.

6 Performance Evalation 64 We assme that the roting optimization is sed to redce the latency of the packets sent/received by the M to/from the C. So, in order to restore the downstream, the M shold execte the retrn rotability procedre before sending the BU message to the C. his is the case of the MIP and inter. So, the latency to restore the downstream flow for the MIP and the inter can be expressed as follows: Lat _ Down( MIP = L2 Lat _ Down( + ( (2 + 2 int er + ( ( ( ( HA C = L2 HA C M AR M,C M HA HA MAP M C M HA M AR M MAP Becase of the hierarchical characteristic of the intra, the M does not need to execte the retrn rotability procedre to restore the downstream. his stream is restored jst after the bi-directional tnnel is established between the M and the MAP. So, the latency in this case can be expressed as follows: Lat _ Down( int ra = L2 + ( M AR M MAP Becase of the bffering and forwarding mechanisms of both the FMIP and, for these protocols, the M also does not need to execte the retrn rotability procedre to restore the downstream. he latency in these cases can be expressed as follows: Lat _ Down( FMIP Lat _ Down( FMIP Lat _ Down( = L2 + = L2 + = L2 + M AR AR PAR M AR M AR

7 Performance Evalation 65 Lat _ Down( = L LRS PLRS M AR AR PAR PLRS PAR AR LRS Finally, the average latency to restore the downstream dring a session can be expressed as follows: Avg _ Lat _ Down( MIP m Lat _ Downi ( MIP = 1 m Avg _ Lat _ Down( = p int ra + (1 p int ra int ra 1 Lat _ Down ( int er 1 i int ra i int er int ra Lat _ Down ( + int er where p intra is the intra-domain handover probability and m = intra + inter. Avg _ Lat _ Down( FMIP = p + (1 p 1 Lat _ Down ( FMIP 1 i i + Lat _ Down ( FMIP Avg _ Lat _ Down( = p + (1 p i= 1 Lat _ Down ( i= 1 i i Lat _ Down ( where p is the icipated mode probability, m = +, is the nmber of icipated operations and is the nmber of ive operations. 5.3 Average handover latency to restore the pstream flow dring a session Average handover latency to restore the pstream flow (Avg_Lat_Up dring a session is defined as the average time of all handovers of a session to restore

8 Performance Evalation 66 the M s ability to send packets of the ongoing pstream. o redce handover latency, some efforts were also made to enhance the L2-handover and the L3- handover. Here, we focs on L3-handover analysis and assme a const vale L2 to express the L2-handover latency. Once again, we assme that the roting optimization is sed to redce the latency of the packets sent/received by the M to/from the C. So, in order to restore the pstream, the M shold execte the retrn rotability procedre before sending the BU message to the C. his is the case of the MIP, inter and FMIP. So, the latency to restore the pstream of these protocols can be expressed as follows: Lat _ Up( MIP = Lat _ Down( MIP Lat _ Up( Lat _ Up( FMIP Lat _ Up( FMIP int er = Lat _ Down( = Lat _ Down( MIP = Lat _ Down( MIP Becase of the hierarchical characteristic of the intra, the M also does not need to execte the retrn rotability procedre to restore the pstream. his stream is also restored jst after the bi-directional tnnel is established between the M and the MAP. So, the latency in this case can be expressed as follows: Lat _ Up( = Lat _ Down( int ra he pstream restore procedre is one of the main advages of the. Dring the handover procedre, the network entities pdate the mobility binding cache of the AR of the C (or Cs on behalf of the M. So, when the M arrives at the new GA, it can restore the pstream jst after sending the C2_HandoverFinish message and receiving the 2C_HandoverAcknowledge message. he latency in this case can be expressed as follows: Lat _ Up( = L2 int er int ra M AR For the, the network entities can only pdate the mobility binding cache of the AR of the C (or Cs after the M has arrived at the new GA. In order to restore the pstream, the following procedre is exected: (i the M sends the C2_HandoverFinish message to the AR. (ii the AR sends the 2_HandoverIndication message to the LRS to reqest the handover context

9 Performance Evalation 67 of the M at the PAR; (iii the LRS relays the 2_HandoverIndication message to the PLRS; (iv the PRLS relays de 2_HandoverIndication message to the PAR; (v the PAR sends the 2_Handover Acknowledge message to the AR to transfer the handover context of the M; (vi the AR sends the 2_Bindingotification message to the LRS; (vii the LRS sends the 2_Bindingotification message to the LRS of the C; (viii the LRS of the C sends the 2_Bindingotification message to the AR of the C to pdate the binding of the M; and (ix the AR sends the 2_HandoverAcknowledge message to the M. So, the latency in this case can be expressed as follows: Lat _ Up( = L LRS PLRS PAR AR AR LRS C _ LRS C _ AR M AR PLRS PAR LRS C _ LRS Finally, the average latency to restore the pstream dring a session can be expressed as follows: Avg _ Lat _ UP( MIP = Avg _ Lat _ Down( MIP Avg _ Lat _ Up( = Avg _ Lat _ Up( FMIP = Avg _ Lat _ Down( Avg _ Lat _ Down( MIP Avg _ Lat _ UP( = p + (1 p i= 1 Lat _ Up ( i= 1 i i Lat _ Up ( where p is the icipated mode probability, m = +, is the nmber of icipated handovers and is the nmber of ive handovers.

10 Performance Evalation Bffer size reqirement for bffering and forwarding mechanisms Some of the mobility protocols spport bffering control of packets dring handover time. his is the case of both the icipated mode of operation of the FMIP and the. hese protocols se signaling messages to enable the bffering and forwarding of in-flight packets from the PAR to the AR. We will denote the bffer size reqirement for downstream (Bf_Size_Down and pstream (Bf_Size_Up as a fnction of the packet transmission rate, as follows. For the icipated mode of operation of the FMIP, both the PAR and the AR shold se bffers to store downstream in-flight packets. he PAR begins to store packets immediately after receiving the FBU message and remains bffering packets ntil the HAck message is received and the forwarding mechanism starts. In its trn, the AR begins to store packets when the first in-flight downstream packet (forwarded by the PAR is received and remains bffering packets ntil the FA message is received and the delivering of packets starts. In this case, the bffer size reqirement can be expressed as follows: Bf _ Size _ Down( FMIP = [ + [ L2 + M PAR AR PAR ] R M AR d + ] R d he bffering mechanism of the icipated mode of operation of the also stores in-flight downstream packets at both the PAR and the AR. he PAR begins to store packets immediately after receiving the C2_HandoverInitiate message and remains bffering packets ntil the 2_HandoverAcknowledge message is received, when the forwarding mechanism starts. In trn, the AR begins to store packets when the first in-flight downstream packet (forwarded by the PAR is received and remains bffering packets ntil the C2_HandoverFinish message is received, when the delivering of the bffered packets starts. So, for the, the bffer size reqirement can be expressed as follows: Bf _ SIze _ Down( = [ [ L2 + PLRS LRS M PAR AR PAR M AR ] R d + ] R d PAR PLRS LRS AR

11 Performance Evalation 69 In both the FMIP and the, the bffer size reqirement for the pstream flow shold be spported by the M. For the FMIP, the reqirement is independent from the mode of operation (icipated or ive and the pstream flow is restored after the retrn rotability and binding pdate procedres. For the, the bffer reqirement depends on the mode of operation. Under the icipated mode, the starts the bffering mechanism after the L2-Down trigger is fired by the link layer and finishes after the AR acknowledges the C2_HandoverFinish message sent by the M. his optimization is a reslt of the icipated registering of the M at the GA and the notification of the ARs of the Cs dring L2 handover. For the ive mode, the bffering mechanism also starts after the L2-Down trigger bt only finishes after the binding notification procedre finishes and the AR acknowledges the 2_HandoverFinish message sent by the M. So, for these protocols, the bffer size reqirement for the pstream can be expressed as follows: Bf _ Size _ Up( FMIP = [ L2 + ( ( HA C M C M AR M HA ] R Bf _ Size _ Up( = [ L2 ] R M AR Bf _ Size _ Up( = [ L2 + 2 AR LRS PLRS PAR LRS C _ LRS M AR PAR AR LRS PLRS C _ LRS C _ AR ] R 5.5 Reslt Analysis Or experiment ses the eqations of the performance parameters presented in sections 5.1, 5.2, 5.3 and 5.4 to analyze the mobility scenario illstrated in Figre 18. In this scenario, the M moves from the access network of the AR11 towards the access network of the AR43. We assme that the M establishes a bidirectional commnication session with the stationary C. his commnication session begins at the network of the AR11 and terminates when the M arrives at the network of the AR43. We also assme that the average movement speed of the

12 Performance Evalation 70 M is adeqate to allow the whole exection of the handover process between each access network. For each of the mobility protocols analyzed in or work, we sed the eqations to calclate the time to execte the intra-domain and inter-domain handover process. Along the path covered by the M, eleven handovers takes place, eight of which are intra-domain and three others are inter-domain. Or analysis of the handover process calclates the time to execte three parts of the handover: (i the time to start the bffering and forwarding mechanisms; (ii the time to restore the downstream flow of packets; and (iii the time to restore the pstream flow of packets. he time vales calclated for each part of the handover of the same mobility protocol considerably varies according to the location the handover takes place becase the signaling messages exchanged between the entities that controls the mobility of the M mst have to go throgh short or long paths to reach their destiny. So, for the first approach of or analysis, we calclated the time average vales to execte each of the three parts of the handover process, for each mobility protocol, along the path covered by the M. he vales of the parameters sed in the eqations are listed in able 6 and the D x-y vales (average nmber of hops between x and y sed are illstrated in Figre 19. he average time vales calclated are listed in able 7. he vale sed for the L2 handover time (L2 is the sm of the average time of the search and exection phases of the L2 handover process of for different commercial IEEE b cards with different chipsets, as described in [15]. he detection phase is discarded becase we assme that the MIES is available. So, this phase is sbstitted by the L2-GoingDown trigger. he L2-Down and L2- UP triggers are also available.

13 Performance Evalation 71 able 6 Parameters settings sed in the first approach of the analysis Parameter Vale S B w L w B wl L wl P t L2 48 bytes 100 Mbps 0.5 msec 11 Mbps 2 msec 0,001 msec 186 msec ARcn 0 C MAP MAP2 MAP3 MAP M 0 AR11 AR12 AR13 AR21 AR22 AR23 AR31 AR32 AR33 AR41 AR42 AR LRS1 LRS2 LRS3 LRS4 MAP4 0 LRS4 2 4 MAP3 0 LRS3 3 2 MAP1 2 MAP2 0 0 LRS1 LRS2 Figre 19 Dx-y vales sed in the analysis

14 Performance Evalation 72 he average time vales of able 7 demonstrate that the time to start the bffering and forwarding mechanisms in the FMIP and is qite similar in both modes of operation. his indicates (and will be frther demonstrated that the loss of downstream packets will be very similar. he reslts also demonstrate that the time to restore the downstream packets in the dring inter-domain handover is greater than the MIP. his is expected becase ses more signaling messages than the MIP to optimize the intra-domain handover, in which case the shows a better performance than the MIP. able 7 Average time vales calclated to the three parts of the handover process Average time vales calclated (ms Start bffering/ forwarding mechanism Restore downstream flow of packets Restore pstream flow of packets Intra Inter Intra Inter Intra Inter MIP A A A A FMIP icipated FMIP ive icipated ive Jst after the L2 handover, both the FMIP and are ready to restore the downstream flow. So, the average time of downstream reestablishment is almost the same as the time needed to complete L2 handover. heir performances are better than the MIP and becase they don t need to execte the retrn rotability procedre to restore the downstream flow. On the other hand, FMIP is reqired to execte this procedre to restore the pstream flow the same way they do. herefore, the FMIP pstream-flow restoration time is qite similar to the time calclated for the MIP and the. Here, we call attention to the first clear advage of the. When operating in the icipated mode, the registration

15 Performance Evalation 73 and binding notifications are exected by the AR and the LRS of the new GA on behalf of the M (dring the L2 handover process, which enables the M to restore the pstream flow of packets immediately after the L2 handover. Even when operating in the ive mode, this time is lower than the ones in the other protocols, which indicates that the binding notification procedre of the performs better than the binding pdate procedre. he second point of or analysis calclates the loss of packets of the downstream flow and the time reqired to restore the downstream and pstream flows for different transmission rate conditions. We also analyze the performance of these protocols for different probability occrrences of the intra-domain and interdomain handover, as well as of the icipated and ive mode of operations. We assme that the bidirectional commnication session between the M and the C takes 1000 seconds. he residence time of the M in each access network is 20 seconds. herefore, the nmber of movements between access networks dring a commnication session is 50. Each L3 handover between access networks is exected as an intra-domain handover with probability P intra and as an inter-domain handover with probability (1 P intra. In the same manner, when relev, the icipated mode of operation occrs with probability P ecipated while the ive mode of operation occrs with probability (1 P ecipated. able 8 Parameters settings sed in the second analysis Parameter Vale s r 1000 sec 20 sec m 50 R d R kbps kbps P intra 0 1 (defalt 0.8 P icipated 0 1 (defalt 0.8

16 Performance Evalation 74 he vales of the parameters sed in the eqations for this second analysis are the same sed for the first one pls the average time vales listed in able 7 and the parameters listed in able 8. Figre 20 shows the comparison of the total packet loss of the downstream flow for a commnication session. Becase of the absence of the bffering and forwarding mechanisms, MIP and loose a signific amont of packets if compared to FMIP and when the downstream transmission rate is above 64 Kbps. For the FMIP and, the loss of packets is eqivalent becase their bffering and forwarding mechanisms are essentially the same. hese protocols loose packets from the downstream flow dring handover only in the ive mode. In the icipated mode, we assme that the amont of bffers reqired to store packets are always available and none of the forwarding packets are lost. his analysis sed the following probabilities: P intra = 0.8 and P ecipated = 0.8. Downstream Packet Loss mber of Packets MIP FMIP Downstream ransmission Rate (Kbps Figre 20 Downstream flow packet loss verss transmission rate As shown in Figre 21, the average bffer size reqirement of the FMIP to store packets of the pstream flow is significly bigger than the one reqired for the. he FMIP reqires abot 38% more bffer area to store these packets becase, after the L2 handover, it is reqired to execte the retrn rotability procedre to notify the C abot the changing of the binding. So, it mst contine the bffering mechanism for an extra period ntil the procedre is finished. On the

17 Performance Evalation 75 other hand, the starts the forwarding mechanism immediately after the L2 handover, when the icipated mode of operation is in progress. Bffer Size Reqirement Bffer Size (KB FMIPdown down FMIPp p ransmission Rate (Kbps Figre 21 Bffer size reqirement for pstream/downstream flow Figre 22 shows the comparison of the latency to restore the downstream flow based on the probability of intra-domain handover and the operating probability of the icipated mode. his analysis demonstrates that most of the time the performs better than the MIP. he hierarchical fnctionality of the restores the commnication faster than the MIP becase, most of the time, the registering of the M and the establishment of the bidirectional tnnel is local to the access network. Unlike the, the MIP always registers the M and establishes the bidirectional tnnel with the HA. FMIP and perform significly better than both the other two. Basically, those protocols make se of the bffering and forwarding mechanisms to redce the latency. he performance of the FMIP is slightly better than the performance of the in the ive mode of operation. otice that the defines the protocol messages to se the service of the LRS to discover the address of the ARs based on the PoA-ID and to notify the PAR abot the handover process. hese messages take part in the handover process and are time consming. Althogh these tasks mst also be exected by the FMIP, there are no entities defined to execte them. So, they are implicitly exected by the AR itself which increases the complexity of this entity. Becase of this overbrden of the AR, the FMIP has fewer signaling messages than the which leads to a lower latency.

18 Performance Evalation 76 Average Downstream-Flow Restore Latency average latency (ms ,2 0,4 0,6 0,8 1 micromobility and ecipated mode probability MIP FMIP Figre 22 Average downstream-flow restore latency Average Upstream-Flow Restore Latency average latency (ms ,2 0,4 0,6 0,8 1 micromobility and ecipated mode probability MIP FMIP Figre 23 - Average pstream-flow restore latency In the following, we compare the latency to restore the pstream flow based on the probability of the intra-domain handover and the probability of the icipated mode. his analysis demonstrates that the MIP and the FMIP are eqivalent in this case. Here, the FMIP is reqired to register the M at the HA and to establish the bidirectional tnnel with the HA in order to execte the retrn rotability procedre. hese procedres are time consming and significly impact the performance of the FMIP. So, now, the reaches a better average performance

19 Performance Evalation 77 than the MIP and FMIP. Here, the advage of the icipated registering and binding notification procedres exected by the ARs and the LRSs on behalf of the M makes the difference in the performance of the.