Traffic Management for the Available Bit Rate (ABR) Service in ...

More documents

Recommendations

Info

dominated by factors such as the drop policy and TCP timer granularity. Intelligent drop policies can help improve the throughput slightly. 8.5 Source Model and TCP Options We use an in nite source model at the application layer running on top of TCP. This implies that TCP always has a packet to send as long as its window will permit it. Other parameters values used are: TCP maximum segment size MSS=512 bytes IP MTU size = 9180 bytes (no IP segmentation) TCP timer granularity =100 ms Delay-ack timer=0 (disabled) Packet processing time at the destination=0 We implemented the window scaling option so that the throughput is not limited by path length. Without the window scaling option, the maximum window sizeis2 16 bytes or 64 kB. We use a window of16 64 kB or 1024 kB. The network consists of three links of 1000 km each and therefore has a one-way delay of 15 ms (or 291 kB at 155 Mbps). In our simulations, we have not used \fast retransmit and recovery" used in pop- ular TCP Reno implementation. In a related work [36, 37] (TCP over UBR), we study the e ect of these algorithms in detail. Brie y, these algorithms have been designed to improve TCP performance when a single (isolated) segment is lost (due to errors). However, in high bandwidth links, network congestion results in several dropped segments (a burst loss). In this case, these algorithms are not able to recover 265
from the loss and they trigger the TCP timeout and the slow start algorithm leading to possibly worse performance. 8.6 ABR Source End System and ERICA Parameters The source end system parameters of ABR are selected to maximize the respon- siveness and throughput. The values of source parameters are: TBE = 128, 512 ICR = 10 Mbps ADTF = 0.5 sec CDF (XDF) = 0.5, CRM (Xrm) = TBE/Nrm PCR = 155.52 Mbps, MCR =0,RIF (AIR) = 1 Nrm = 32, Mrm = 2, RDF = 1/512, Trm = 100 ms, TCR =10c/s The ERICA switch algorithm parameters are chosen as follows. The target utilization parameter is chosen to be 90%. The overload and ABR capacity are measured at the switch over an interval of 100 cells or 1 ms (whichever is smaller). The bu er size at the bottleneck link is sized as TBE n 1, 2, or 4, where n is the number of ABR sources. 8.7 The n Source + VBR Con guration Figure 8.4 illustrates the general con guration we analyze, which we call \the n Source + VBR con guration." This con guration has a single bottleneck link 266
Page 1 and 2:
Tra c Management for the Available
Page 3 and 4:
ABSTRACT With the merger of telecom
Page 5 and 6:
Tra c Management for the Available
Page 7 and 8:
To my family iv
Page 9 and 10:
VITA April 30, 1971 :::::::::::::::
Page 11 and 12:
2.7 Switch Behavior . . . . . . . .
Page 13 and 14:
5.8 Proof: Fairness Algorithm Impro
Page 15 and 16:
8.15 Factors A ecting Bu ering Requ
Page 17 and 18:
Bibliography . . . . . . . . . . .
Page 19 and 20:
8.12 Maximum Queues for Satellite N
Page 21 and 22:
5.1 Transmitted cell rate (instanta
Page 23 and 24:
6.13 Results foratwo sources con gu
Page 25 and 26:
7.11 E ect of UILI on Medium Bursts
Page 27 and 28:
C.3 Flow Chart of Bi-Directional Co
Page 29 and 30:
header information which limits the
Page 31 and 32:
switches since a standard does not
Page 33 and 34:
congestion control deals with the c
Page 35 and 36:
hand side of the equation is the su
Page 37 and 38:
analyses which are used to validate
Page 39 and 40:
CHAPTER 2 THE ABR TRAFFIC MANAGEMEN
Page 41 and 42:
time (RTT). As we explain later, AB
Page 43 and 44:
feedback from nearby switches to re
Page 45 and 46:
Note that in-rate and out-of-rate d
Page 47 and 48:
Data cells also have an Explicit Fo
Page 49 and 50:
opportunity the source may nd that
Page 51 and 52:
Figure 2.7: Scheduling of forward R
Page 53 and 54:
as the timeout value) which reduces
Page 55 and 56:
It has been incorrectly believed th
Page 57 and 58:
NI CI Action 0 0 ACR Min(ER, ACR +
Page 59 and 60:
Source Rule 13: Sources can optiona
Page 61 and 62:
Switch Rule 1: This rule speci es t
Page 63 and 64:
Observe that the ABR tra c manageme
Page 65 and 66:
complex method like per-VC queuing
Page 67 and 68:
graphs have a steady state with con
Page 69 and 70:
Figure 3.2: Operating point between
Page 71 and 72:
The following example illustrates t
Page 73 and 74:
mention this concept of fairness fo
Page 75 and 76:
control problem was also simpler. S
Page 77 and 78:
of such an occurrence is expected t
Page 79 and 80:
y giving only one feedback per meas
Page 81 and 82:
CHAPTER 4 SURVEY OF ATM SWITCH CONG
Page 83 and 84:
The adaptive FCVC algorithm [63] co
Page 85 and 86:
np. Thus, long path VCs have fewer
Page 87 and 88:
networks like ATM can have complete
Page 89 and 90:
scheme followed by a discussion of
Page 91 and 92:
exibility of decoupling the enforce
Page 93 and 94:
4.6.1 Key Techniques In EPRCA, the
Page 95 and 96:
If the mean ACR is not a good estim
Page 97 and 98:
is the ratio of the input rate to t
Page 99 and 100:
which shares the link equally as al
Page 101 and 102:
proportional to the the unused ABR
Page 103 and 104:
The key technique in the scheme is
Page 105 and 106:
the bandwidth allocations to satis
Page 107 and 108:
4.10 DMRCA scheme The Dynamic Max R
Page 109 and 110:
on its CCR. Further MAX times out i
Page 111 and 112:
other words, a di erent set of para
Page 113 and 114:
A combination of several ideas in a
Page 115 and 116:
4.13 SP-EPRCA scheme The SP-EPRCA s
Page 117 and 118:
RTD and shuts o the source (for sta
Page 119 and 120:
4.14.1 Common Drawbacks Though the
Page 121 and 122:
The ATM Tra c Management standard a
Page 123 and 124:
5.1.1 Control-Cell Format The contr
Page 125 and 126:
The last two elds are used in the b
Page 127 and 128:
Figure 5.3: Flow chart for updating
Page 129 and 130:
LAF in cell Max(LAF in cell, z) The
Page 131 and 132:
the time of departure (instant mark
Page 133 and 134:
the steady state. The system operat
Page 135 and 136:
5.2.3 Use Measured Rather Than Decl
Page 137 and 138:
said about the maximum queue length
Page 139 and 140:
The key problem with some unipolar
Page 141 and 142:
Figure 5.6: Space time diagram show
Page 143 and 144:
As noted, these heuristics do not g
Page 145 and 146:
Transmitted Cell Rate ABR Queue Len
Page 147 and 148:
4. The RM cell contains a timestamp
Page 149 and 150:
1. The source o ered average cell r
Page 151 and 152:
Transmitted Cell Rate Link Utilizat
Page 153 and 154:
Page 155 and 156:
Page 157 and 158:
Figure 5.17: The parking lot fairne
Page 159 and 160:
Figure 5.19: Network con guration w
Page 161 and 162:
Page 163 and 164:
Page 165 and 166:
5.8 Proof: Fairness Algorithm Impro
Page 167 and 168:
Region 2: y s and x s and U(1 + ) x
Page 169 and 170:
eing divided into eight non-overlap
Page 171 and 172:
Proof for Region 2 Triangular regio
Page 173 and 174:
have: and y 0 = y(1 ; ) z x + y 0 =
Page 175 and 176:
The OSU scheme is, therefore, incom
Page 177 and 178:
workloads, where input load and cap
Page 179 and 180:
(a) Regions used to prove Claim C1
Page 181 and 182:
6.1 The Basic ERICA Algorithm The s
Page 183 and 184:
A ow chart of the basic algorithm i
Page 185 and 186:
efore competing sources can receive
Page 187 and 188:
that the latest CCR information is
Page 189 and 190:
the CCR value may not be an accurat
Page 191 and 192:
(Target Utilization) (Link Bandwidt
Page 193 and 194:
ABR Capacity Input Rate Overload Fa
Page 195 and 196:
of arithmetic averaging used above.
Page 197 and 198:
level of control. These parameters
Page 199 and 200:
6.14 ERICA+: Queue Length as a Seco
Page 201 and 202:
6.16 ERICA+: Maintain a \Pocket" of
Page 203 and 204:
hence, introduces a new parameter,
Page 205 and 206:
Figure 6.4: Linear functions for ER
Page 207 and 208:
and Figure 6.6: The queue control f
Page 209 and 210:
small averaging intervals. If the a
Page 211 and 212:
The maximum value of T 0 also depen
Page 213 and 214:
6.22 Performance Evaluation of the
Page 215 and 216:
espectively. The target delay T 0 i
Page 217 and 218:
6.22.4 Fairness Figure 6.9: Parking
Page 219 and 220:
st link, VC 15 is limited to a thro
Page 221 and 222:
From the gures, it is clear that ER
Page 223 and 224:
counts the bursty source as active
Page 225 and 226:
6.23 Summary of the ERICA and ERICA
Page 227 and 228:
(a) Transmitted Cell Rate (c) Link
Page 229 and 230:
Page 231 and 232:
Page 233 and 234:
(a) Transmitted Cell Rate (basic ER
Page 235 and 236:
(a) Transmitted Cell Rate (basic ER
Page 237 and 238:
Page 239 and 240:
Page 241 and 242: (a) Transmitted Cell Rate (c) Link
Page 249 and 250: (a) Transmitted Cell Rate (basic ER
Page 251 and 252: RM cells. In this chapter, we devot
Page 253 and 254: low. This tradeo was discovered and
Page 255 and 256: in the speci cation. b) ACR shall n
Page 257 and 258: 7.1.6 December 1995 Proposals There
Page 259 and 260: Figure 7.2: Multiplicative vsAdditi
Page 261 and 262: is never triggered. However, the PR
Page 263 and 264: simulation results of bursty source
Page 265 and 266: 1. The time-based proposal also ind
Page 267 and 268: The switch maintains a local alloca
Page 269 and 270: PNI = f0, 1g : f1 ) No rule 5b, 0 )
Page 271 and 272: after reaching the goal. The time-b
Page 273 and 274: Figure 7.8: Closed-Loop Bursty Tra
Page 275 and 276: The algorithm measures the load and
Page 277 and 278: Medium Bursts Medium bursts are exp
Page 279 and 280: count-based technique may be insu c
Page 281 and 282: Figure 7.13: An event trace illustr
Page 283 and 284: CHAPTER 8 SUPPORTING INTERNET APPLI
Page 285 and 286: u ering which does not depend upon
Page 287 and 288: 4 make it to the destination are ar
Page 289 and 290: Figure 8.3: At the ATM layer, the T
Page 291: issues and e ect of bursty applicat
Page 295 and 296: 8.8 Performance Metrics We measure
Page 297 and 298: the performance is fair. Also, the
Page 299 and 300: Figure 8.7(b) shows the rates (ACRs
Page 301 and 302: Window Size in bytes vfive-tcp/opti
Page 303 and 304: The e ect of large bu ers on CLR is
Page 305 and 306: 8.13 Summary of TCP/IP performance
Page 307 and 308: Feedback delay: Twice the delay fro
Page 309 and 310: on a link, two cells are expected a
Page 311 and 312: state only after the switch algorit
Page 313 and 314: queue length is less susceptible to
Page 315 and 316: to equalize rates for fairness, and
Page 317 and 318: QB = Link bandwidth (RT T ; T ) and
Page 319 and 320: u ered at the end-system, and not i
Page 321 and 322: Part b): When ABR load goes away, t
Page 323 and 324: All our simulations presented use t
Page 325 and 326: Averaging RTT(ms) Feedback Max Q Th
Page 327 and 328: a modi ed version of the ERICA algo
Page 329 and 330: ABR is better than UBR in these (en
Page 331 and 332: problems. During the ON time, the V
Page 333 and 334: hand, the frequency of the VBR is h
Page 335 and 336: like ERICA+ which uses the queueing
Page 337 and 338: minimum fairshare is low. This may
Page 339 and 340: 8.17 E ect of Long-Range Dependent
Page 341 and 342: terminology) are called \Presentati
Page 343 and 344:
The key point is that the MPEG-2 ra
Page 345 and 346:
the MPEG-2 Transport Stream, and st
Page 347 and 348:
8.17.4 Observations on the Long-Ran
Page 349 and 350:
For the video sources, we choose me
Page 351 and 352:
Video Sources ABR Metrics # Mean St
Page 353 and 354:
However, with modi cations to ERICA
Page 355 and 356:
Video Sources ABR Metrics # Avg Src
Page 357 and 358:
Video Sources ABR Metrics # Avg Src
Page 359 and 360:
On the other hand, if the applicati
Page 361 and 362:
of managing bu ers, queueing, sched
Page 363 and 364:
hence control the total load on the
Page 365 and 366:
call such a switch a \VS/VD switch"
Page 367 and 368:
Figure 9.2: Per-class queues in a n
Page 369 and 370:
which arises is where the rate calc
Page 371 and 372:
9.2 The ERICA Switch Scheme: Renota
Page 373 and 374:
The unknowns in the above equations
Page 375 and 376:
Figure 9.9: Two methods to measure
Page 377 and 378:
9.4 VS/VD Switch Design Options 9.4
Page 379 and 380:
# VC Rate VC Input Rate Input Rate
Page 381 and 382:
8 uses source rate measurement, we
Page 383 and 384:
The allocated rate update and the e
Page 385 and 386:
sources in chapter 6. We expect the
Page 387 and 388:
con guration mentioned in the table
Page 389 and 390:
can be very di erent for di erent V
Page 391 and 392:
CHAPTER 10 IMPLEMENTATION ISSUES At
Page 393 and 394:
With an enhanced UBR service, appli
Page 395 and 396:
2. Some switch schemes have a proce
Page 397 and 398:
4. Large legacy switches have a pro
Page 399 and 400:
section 9. Further, WAN switches wo
Page 401 and 402:
CHAPTER 11 SUMMARY AND FUTURE WORK
Page 403 and 404:
good transient performance. Since r
Page 405 and 406:
variant background tra c conditions
Page 407 and 408:
APPENDIX A SOURCE, DESTINATION AND
Page 409 and 410:
7. After following behaviors #5 and
Page 411 and 412:
set the QL and SN elds to zero, pre
Page 413 and 414:
d) VS/VD Control: The switch may se
Page 415 and 416:
5. Setting of other parameters at V
Page 417 and 418:
4. The averaging interval timer exp
Page 419 and 420:
1. Initialization: Target Cell Rate
Page 421 and 422:
THEN IF (OCR In Cell Fair Share Rat
Page 423 and 424:
APPENDIX C ERICA SWITCH ALGORITHM:
Page 425 and 426:
Number Active VCs In Last Interval
Page 427 and 428:
IF (NOT(Averaging VCs Option)) THEN
Page 429 and 430:
IF (Load Factor = In nity) THEN Loa
Page 431 and 432:
; (Contribution[VC] = Decay Factor)
Page 433 and 434:
Name Explanation Flow Chart (FC) or
Page 435 and 436:
Figure C.2: Flow Chart for Achievin
Page 437 and 438:
Figure C.4: Flow Chart of averaging
Page 439 and 440:
Figure C.6: Flow chart of averaging
Page 441 and 442:
C.3 Pseudocode for VS/VD Design Opt
Page 443 and 444:
(* | Bottleneck rate of next loop |
Page 445 and 446:
Follow SESRules 1-4 (see appendix A
Page 447 and 448:
CRM - Missing RM-cell Count DIR bit
Page 449 and 450:
TUB -Target Utilization Band Trm -
Page 451 and 452:
[12] J. Bennett and G. Tom Des Jard
Page 453 and 454:
[38] M. Grossglauser, S.Keshav, and
Page 455 and 456:
[64] H. T. Kung. Flow Controlled Vi
show all

Traffic Management for the Available Bit Rate (ABR) Service in ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?