-
Notifications
You must be signed in to change notification settings - Fork 1
/
draft-eckert-bier-cgm2-rbs-01.xml
1503 lines (1488 loc) · 79.2 KB
/
draft-eckert-bier-cgm2-rbs-01.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<?xml version='1.0' encoding='utf-8'?>
<!DOCTYPE rfc [
<!ENTITY nbsp " ">
<!ENTITY zwsp "​">
<!ENTITY nbhy "‑">
<!ENTITY wj "⁠">
]>
<?xml-stylesheet type="text/xsl" href="rfc2629.xslt" ?>
<!-- generated by https://github.com/cabo/kramdown-rfc2629 version 1.5.25 (Ruby 2.7.0) -->
<?rfc toc="yes"?>
<?rfc sortrefs="yes"?>
<?rfc symrefs="yes"?>
<?rfc comments="yes"?>
<rfc xmlns:xi="http://www.w3.org/2001/XInclude" ipr="trust200902" docName="draft-eckert-bier-cgm2-rbs-01" category="exp" obsoletes="" updates="" submissionType="IETF" xml:lang="en" tocInclude="true" sortRefs="true" symRefs="true" version="3">
<!-- xml2rfc v2v3 conversion 3.12.1 -->
<front>
<title abbrev="bier-cgm2-rbs">Carrier Grade Minimalist Multicast (CGM2) using Bit Index Explicit Replication (BIER) with Recursive BitString Structure (RBS) Addresses</title>
<seriesInfo name="Internet-Draft" value="draft-eckert-bier-cgm2-rbs-01"/>
<author initials="T." surname="Eckert" fullname="Toerless Eckert">
<organization>Futurewei Technologies USA</organization>
<address>
<postal>
<street>2220 Central Expressway</street>
<city>Santa Clara</city>
<code>CA 95050</code>
<country>USA</country>
</postal>
<email>tte@cs.fau.de</email>
</address>
</author>
<author initials="B." surname="Xu" fullname="Bing (Robin) Xu">
<organization>Huawei Technologies (2012Lab)</organization>
<address>
<postal>
<country>China</country>
</postal>
<email>bing.xu@huawei.com</email>
</address>
</author>
<date year="2022" month="February" day="09"/>
<workgroup>BIER</workgroup>
<abstract>
<t>This memo introduces the architecture of a multicast
architecture derived from BIER-TE, which this memo calls
Carrier Grade Minimalist Multicast (CGM2). It reduces
limitations and complexities of BIER-TE by replacing
the representation of the in-packet-header delivery tree
of packets through a "flat" BitString of adjacencies
with a hierarchical structure of BFR-local BitStrings
called the Recursive BitString Structure (RBS) Address.</t>
<t>Benefits of CGM2 with RBS addresses include smaller/fewer BIFT in BFR,
less complexity for the network architect and in the CGM2
controller (compared to a BIER-TE controller) and fewer packet
copies to reach a larger set of BFER.</t>
<t>The additional cost of forwarding with RBS addresses is
a slightly more complex processing of the RBS address
in BFR compared to a flat BitString and the novel per-hop rewrite of
the RBS address as opposed to bit-reset rewrite in BIER/BIER-TE.</t>
<t>CGM2 can support the traditional deployment model of BIER/BIER-TE
with the BIER/BIER-TE domain terminating at service provider
PE routers as BFIR/BFER, but it is also the intention of this document to
expand CGM2 domains all the way into hosts, and therefore
eliminating the need for an IP Multicast flow overlay,
further reducing the complexity of Multicast services using
CGM2. Note that this is not fully detailed in this version
of the document.</t>
<t>This document does not specify an encapsulation for CGM2/RBS
addresses. It could use existing encapsulations such as <xref target="RFC8296" format="default"/>,
but also other encapsulations such as IPv6 extension headers.</t>
</abstract>
</front>
<middle>
<section anchor="overview" numbered="true" toc="default">
<name>Overview</name>
<section anchor="introduction" numbered="true" toc="default">
<name>Introduction</name>
<t>Carrier Grade Minimalist Multicast (CGM2) is an architecture
derived from the BIER-TE architecture <xref target="I-D.ietf-bier-te-arch" format="default"/> with the following
changes/improvements.</t>
<t>CGM2 forwarding is based on the principles of BIER-TE forwarding:
It is based on an explicit, in-packet, "source routed" tree indicated
through bits for each adjacency that the packet has to
traverse. Like in BIER-TE, adjacencies can be L2 to a subnet
local neighbor in support of "native" deployment of CGM2
and/or L3, so-called "routed" adjacencies to support
incremental or partial deployment of CGM2 as needed.</t>
<t>The address used to replicate packets in the network is
not a flat network wide BitString as in BIER-TE, but a
hierarchical structure of BitStrings called a Recursive BitString Structure (RBS)
Address. The significance of the BitPositions (BP)
in each BitString is only local to the BIFT of the router/BFR
that is processing this specific BitString.</t>
<t>RBS addressing allows for a more compact representation of
a large set of adjacencies especially in the common case
of sparse set of receivers in large Service Provider Networks (SP).</t>
<t>CGM2 thereby eliminates the challenges in BIER <xref target="RFC8279" format="default"/> and BIER-TE having to
send multiple copies of the same packet in large SP
networks and the complexities especially for BIER-TE
(but also BIER) to engineer multiple set identifier (SI) and/or
sub-domains (SD) BIER-TE topologies for limited size BitStrings
(e.g.: 265) to cover large network topologies.</t>
<t>Like BIER-TE, CGM2 is intended to leverage a Controller
to minimize the control plane complexity in the network to
only a simple unicast routing underlay required only for
routed adjacencies.</t>
<t>The controller centric architecture
provides most easily any type of required traffic optimization
for its multicast traffic due to their need to perform
often NP-complete calculations across the whole topology:
reservation of bandwidth to support CIR/PIR traffic buffer/latency
to support Deterministic Network (DetNet) traffic, cost optimized
Steiner trees, failure point disjoint trees for higher resilience including
DetNet deterministic services.</t>
<t>CGM2 can be deployed as BIER/BIER-TE are specified today,
by encapsulating IP Multicast traffic at Provider Edge (PE)
routers, but it is also considered to be highly desirable
to extend CGM2 all the way into Multicast Sender/Receivers
to eliminate the overhead of an Overlay Control plane for
that (legacy) IP Multicast layer and the need to deal with
yet another IP multicast group addressing space. In this deployment option
Controller signaling extends directly (or indirectly via BFIR) into
senders.</t>
</section>
<section anchor="encapsulation-considerations" numbered="true" toc="default">
<name>Encapsulation Considerations</name>
<t>This document does not define a specific BIER-RBS encapsulation
nor does it preclude that multiple different encapsulations
may be beneficial to better support different use-cases
or operator/user technology preferences. Instead, it discusses
considerations for specific choices.</t>
<t>BIER-RBS can easily re-use <xref target="RFC8296" format="default"/> encapsulation. The
RBS address is inserted into the <xref target="RFC8296" format="default"/> BitString
field. The BFR forwarding plane needs to be configured
(from Controller or control plane) that the BIFT-id(s) used
with RBS addresses are mapped to BIFT and forwarding
rules with RBS semantic.</t>
<t>SI/SD fields of <xref target="RFC8296" format="default"/> may be used as in BIER-TE,
but given that CGM2 is designed (as described in the Overview
section) to simplify multicast services, a likely and
desirable configuration would be to only use a single
BIFT in each BFR for RBS addresses, and mapping these to
a single SD and SI 0.</t>
<t>IP Multicast <xref target="RFC1112" format="default"/> was defined as an extension
of IP <xref target="RFC791" format="default"/>, reusing the same network header, and
IPv6 multicast inherits the same approach. In comparison,
<xref target="RFC8296" format="default"/> defines BIER encapsulation as a
completely separate (from IP) layer 3 protocol,
and duplicates both IP and MPLS header elements into the
<xref target="RFC8296" format="default"/> header. This not only results in always
unused, duplicate header parameters (such as TC vs. DSCP), but
it also foregoes the option to use any non-considered
IPv6 extension headers with BIER and would require the
introduction of a whole new BIER specific socket API
into host operating systems if it was to be supported
natively in hosts.</t>
<t>Therefore an encapsulation of RBS addresses using an
IP and/or IPv6 extension header may be more desirable
in otherwise IP and/or IPv6 only deployments, for example
when CGM2 is extended into hosts, because it would allow
to support CGM2 via existing IP/IPv6 socket APIs as long as
they support extension headers, which the most important
host stacks do today.</t>
</section>
</section>
<section anchor="cgm2rbs-architecture" numbered="true" toc="default">
<name>CGM2/RBS Architecture</name>
<t>This section describes the basic CGM2 architecture
via <xref target="FIG-ARCH" format="default"/> through its key differences over the BIER-TE
architecture.</t>
<figure anchor="FIG-ARCH">
<name>CGM2/RBS Architecture</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
Optional
|<-IGMP/PIM-> multicast flow <-PIM/IGMP->|
overlay
CGM2 [CGM2 Controller]
control plane . ^ ^ ^
. / | \ BIFT configuration
.......... | | | per-flow RBS setup
. | | |
. v v v
Src (-> ... ) -> BFIR-----BFR-----BFER -> (... ->) Rcvr
|<----------------->|
CGM2 with RBS-address forwarding plane
|<.............. <- CGM domain ---> ...............|
|<--------------------->|
Routing underlay (optional)
]]></artwork>
</figure>
<t>In the "traditional" option, when deployed with a domain
spanning from BFIR to BFER, the CGM2 architecture is very
much like the BIER-TE architecture, in which the BIER-TE forwarding rules
for (BitString,SI,SD) addresses are replaced by the
RBS address forwarding rules.</t>
<t>The CGM2 Controller replaces the BIER-TE controller,
populating during network configuration the BIFT,
which are very much like BIER-TE BIFT, except that
they do not cover a network-wide BP address space, but
instead each BFR BIFT only needs as many BP in its BIFT
as it has link-local adjacencies, and in partial deployments
also additional L3 adjacencies to tunnel across non-CGM
capable routers.</t>
<t>Per-flow operations in this "traditional" option is very much as in
BIER/BIER-TE, with the CGM2 controller determining the
RBS address (instead of the BIER-TE (BitString,SI,SD)) to be
imposed as part of the RBS address header (compared
to the BIER encapsulation <xref target="RFC8296" format="default"/>) on the BFIR.</t>
<t>To eliminate the need for an IP Multicast flow overlays,
a CGM2 domain may extend all the way into Sender/Receiver
hosts. This is called "end-to-end" deployment model.
In that case, the sender host and CGM2 controller
collaborate to determine the desired receivers for
a packet as well as desired path policy/requirements,
the controller indicates to the sender of the packet
the necessary RBS address and address of the BFIR,
and the Sender imposes an appropriate RBS address header
together with a unicast encapsulation towards the BFIR.</t>
<t>CGM2 is also intended so especially simplify
controller operations that also instantiate QoS policies
for multicast traffic flows, such as bandwidth and
latency reservations (e.g.: DetNet). As in BIER-TE, this
is orthogonal to the operations of the CGM2/RBS address
forwarding operations and will be covered in separate documents.</t>
</section>
<section anchor="cgm2rbs-forwarding-plane" numbered="true" toc="default">
<name>CGM2/RBS forwarding plane</name>
<t>Instead of a (flat) BitString as in BIER-TE
that use a network wide shared BP address space for
adjacencies across multiple BFR, CGM2 uses a structured
address built from so-called RecursiveUnits (RU) that contain BitStrings,
each of which is to be parsed by exactly one BFR along the delivery
tree of the packet.</t>
<t>The equivalent to a BIER/BIER-TE BitString is
therefore called the RecursiveUnit BitString Structure (RBS) Address.
Forwarding for CGMP2 is therefore also called RBS forwarding.</t>
<section anchor="rbs-bift" numbered="true" toc="default">
<name>RBS BIFT</name>
<t>RBS BIFT as shown in <xref target="FIG-RBS-BIFT" format="default"/> are, like BIER-TE BIFT, tables that are indexed by
BP, containing for each BP an adjacency. The core difference over BIER-TE
BIFT is that the BP of the BIFT are all local to the BFR,
whereas in BIER-TE, the BP are shared across a BIER-TE domain,
each BFR can only use a subset the BP for its own adjacencies,
and only in some cases can BP be shared for adjacencies across
two (or more) BFR. Because of this difference, most of the complexities
of BIER-TE BIFT are not required with BIER-RBS BIFT, see <xref target="complexities" format="default"/>.</t>
<figure anchor="FIG-RBS-BIFT">
<name>RBS BIFT</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+-------------+
|BP|Recursive| Adjacency|
+--+---------+-------------+
| 1| 1|adjacenct BFR|
+--+---------+-------------+
| 2| 0| punt/host|
+--+---------+-------------+
| ..... ... |
+--+---------+-------------+
| N| ...| ... |
+--+---------+-------------+
]]></artwork>
</figure>
<t>An RBS BIFT has a configured number of N addressable BP entries.
When a BFR receives a packet with an RBS address,
it expects that the BitString inside the RBS address that
needs to be parsed by the BFR (see <xref target="RBS-address" format="default"/> has a length that matches N
according to the encapsulation used for the RBS address.
Therefore, N MUST support configuration in increments of the supported size
of the BitString in the encapsulation of the RBS Address.
In the reference encoding (see <xref target="RBS-address" format="default"/>), the increment for N is 1 (bit).
If an encapsulation would call for a byte accurate encoding of the
BitString, N would have to be configurable in increments of 8.</t>
<t>BFR MUST support a value of N larger than the maximum number of adjacencies
through which RBS forwarding/replication of a single packet is required,
such as the number of physical interfaces on BFR that are intended to be
deployed as a Provider Core (P) routers.</t>
<t>RBS BIFT introduce a new "Recursive" flag for each BP. These
are used for adjacencies to other BFR to indicate that the
BFR processing the packet RBS address BitString also has to
expect for every BP with the recursive flag set another
RU inside the RBS address.</t>
</section>
<section anchor="encoding" numbered="true" toc="default">
<name>Reference encoding of RBS addresses</name>
<t>Structure elements of the RBS Address and its components
are parameterized according to a specific encapsulation
for RBS addresses, such as the total size of the TotalLen
field and the unit in which it is counted (see <xref target="RBS-address" format="default"/>).
These parameters are outside the scope of this document. Instead,
this document defines example parameters that together form the
so called "Reference encoding of RBS addresses". This encoding
may or may not be adopted for any particular encapsulation
of RBS addresses.</t>
</section>
<section anchor="RBS-address" numbered="true" toc="default">
<name>RBS Address</name>
<t>An RBS address is structured as shown in <xref target="FIG-RBS" format="default"/>.</t>
<figure anchor="FIG-RBS">
<name>RBS Address</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+----------+-----+---------------+---------+
| TotalLen | Rsv | RecursiveUnit | Padding |
+----------+-----+---------------+---------+
. .
.... TotalLen .......
]]></artwork>
</figure>
<t>TotalLen counts in some unit, such as bits, nibbles or
bytes the length of the RBS Address excluding
itself and Padding. For the reference encoding, TotalLen
is an 8-bit field that counts the size of the RBS address
in bits, permitting for up to 256 bit long RBS addresses.</t>
<t>In case additional, non-recursive flags/fields are determined
to be required in the RBS Address, they should be encoded
in a field between TotalLen and RecursiveUnit, which is
called Rsv. In the reference encoding, this field has a length
of 0.</t>
<t>Padding is used to align the RBS address as required
by the encapsulation. In the reference encoding, this alignment
is to 8 bits (byte boundaries).
Therefore, Padding (bits) = (8 - TotalLen % 8).</t>
<section anchor="recursiveunit" numbered="true" toc="default">
<name>RecursiveUnit</name>
<t>The RecursiveUnit field is structured as shown in <xref target="FIG-RBS-RU" format="default"/>.</t>
<figure anchor="FIG-RBS-RU">
<name>RBS RecursiveUnit</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-+-+-+-+-+ -+-+-+-+-+-+-+-+-+ -+-+-+-+-+-+-+-+ -+
| BitString...| AddressingField...| RecursiveUnit 1...M|
+-+-+-+-+-+ -+-+-+-+-+-+-+-+-+ -+-+-+-+-+-+-+-+- -+
]]></artwork>
</figure>
<t>The BitString field indicates the bit positions (BPs)
to which the packet is to be replicated using the
BIFT of the BFR that is processing the Recursive unit.</t>
<t>For each of M BP set in the BitString of the RecursiveUnit for
which the Recursive flag is set in the BIFT of the BFR, the
RecursiveUnit contains a RecursiveUnit i, i=1...M, in order of increasing BP index.</t>
<t>If adjacencies between BFR are not configured as recursive in the BIFT,
this recursive extraction does not happen for an adjacency, no
RecursiveUnit i has to be encoded for the BP,
and BFRs across such adjacencies would have to share the
BP of a common BIFT as in BIER-TE. This option is not further
discussed in this version of the document.</t>
</section>
<section anchor="addressingfield" numbered="true" toc="default">
<name>AddressingField</name>
<t>The AddressingField of an RBS address is structured as shown in
<xref target="FIG-RBS-AF" format="default"/>.</t>
<figure anchor="FIG-RBS-AF">
<name>RBS AddressingField</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+
| L1 | L2 |...| L(M-1) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+
]]></artwork>
</figure>
<t>The AddressingField consists of one or more fields Li,
i=1...(M-1). Li is the length of RecursiveUnit i for the
i'th recursive bit set in the BitString preceding it.</t>
<t>In the reference encoding, the lengths are
8-bit fields indicating the length of RecursiveUnits in bits.</t>
<t>The length of the M'th RecursiveUnit is not explicitly
encoded but has to be calculated from TotalLen.</t>
</section>
</section>
</section>
<section anchor="bier-rbs-example" numbered="true" toc="default">
<name>BIER-RBS Example</name>
<t><xref target="FIG-E-TOPO" format="default"/> shows an example for RBS forwarding.</t>
<figure anchor="FIG-E-TOPO">
<name>Example Network Topology</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-+ +-+ +-+
| |-----| |------|C|-=> Client2
+-+ +-+ +-+
/ \ \ /=>/ \
/ \ \ / |
+-+ +-+ +-+ +-+
Client1 =>-|B|-=>-|R|-=>-|S|-=>-|D|-=> Client3
+-+ +-+ +-+ +-+
\ /
\ +-+
\-=>-|E|-=> Client4
+-+
]]></artwork>
</figure>
<t>A packet from Client1 connected to BFIR B is intended to be
replicated to Client2,3,4. The example initially assumes
the traditional option of the architecture, in which the imposition of the
header for the RBS address happens on BFIR B, for example
based on functions of an IP multicast flow overlay.</t>
<t>A controller determines that the packet should be forwarded
hop-by-hop across the network as shown in <xref target="FIG-E-TREE" format="default"/>.</t>
<figure anchor="FIG-E-TREE">
<name>Desired example forwarding tree</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
Client 1 ->B(impose BIER-RBS)
=>R(
=> E (dispose BIER-RBS)
=> Client4
=> S(
=>C (dispose BIER-RBS)
=> Client2
=>D (dispose BIER-RBS)
=> Client3
)
)
]]></artwork>
</figure>
<section anchor="bfr-b" numbered="true" toc="default">
<name>BFR B</name>
<t>The 34 bit long (without padding) RBS address shown in <xref target="FIG-E-B" format="default"/>
is constructed to represent the desired tree from <xref target="FIG-E-TREE" format="default"/> and is
imposed at B onto the packet through an appropriate header supporting the
reference encoding of RBS addresses.</t>
<figure anchor="FIG-E-B">
<name>RBS Address imposed at BFIR-B</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
.............. RecursiveUnit .................
. .
+-------+----+-----+-----+-----+----+-----+------+-----+-----+
|Tlen:34|B:01|R:011|L1:10|S:011|L1:3|C:001|D:0001|E:001|Pad:6|
+-------+----+-----+-----+-----+----+-----+------+-----+-----+
8bit 2bit 3bit 8bit 3bit 8bit 3bit 4bit 3bit 6bit
]]></artwork>
</figure>
<t>In <xref target="FIG-E-B" format="default"/> and further the illustrations of RBS addresses, BitStrings are
preceded by the name of the BFR for whom they are destined
and their values are shown as binary with the lowest BP 1
starting on the left. TotalLength (Tlen:),
AddressingField (L1:) and Padding (Pad:) fields are shown
with decimal values.</t>
<t>RBS forwarding on B examines this address based
on its RBS BIFT with N=2 BP entries, which is shown in
<xref target="FIG-E-B-BIFT" format="default"/>.</t>
<figure anchor="FIG-E-B-BIFT">
<name>BIER-RBS BIFT on B</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+---------+
|BP|Recursive|Adjacency|
+--+---------+---------+
| 1| 0| client1 |
+--+---------+---------+
| 2| 1| R |
+--+---------+---------+
]]></artwork>
</figure>
<t>This results in the parsing of the RBS address as shown in
<xref target="FIG-E-B2" format="default"/>, which shows that B does not need (nor can)
parse all structural elements, but only those relevant
to its own RBS forwarding procedure.</t>
<figure anchor="FIG-E-B2">
<name>RBS Address as processed by BFIR-B</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
......... RecursiveUnit ...............
. .
. ......,.. RecursiveUnit 1 .........
. . .
+-------+----+----------------------------------+-----+
|Tlen:34|B:01|R:01100001010011000000110010001001|Pad:6|
+-------+----+----------------------------------+-----+
8bit 2bit 32bit 6bit
]]></artwork>
</figure>
<t>There is only one BP towards BFR R set in the BitString B:01,
so the RecursiveUnit 1 follows directly after the end
of the BitString B:01 and it covers the whole Tlen - length
of BitString (34 - 2 = 32 bit).</t>
<t>B rewrites the RBS address by replacing the RecursiveUnit with RecursiveUnit 1
and adjusts the Padding to zero bits. The resulting RBS address
is shown in <xref target="FIG-E-R" format="default"/>. It then sends
the packet copy with that rewritten RBS address to BFR R.</t>
</section>
<section anchor="bfr-r" numbered="true" toc="default">
<name>BFR R</name>
<t>BFR R receives from BFR B the packet with that RBS address
shown in <xref target="FIG-E-R" format="default"/>.</t>
<figure anchor="FIG-E-R">
<name>RBS Address processed by BFR-R</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
.............. RecursiveUnit ............
. .
+-------+-----+-----+-----+----+-----+------+-----+
|Tlen:32|R:011|L1:18|S:011|L1:3|C:001|D:0001|E:001|
+-------+-----+-----+-----+----+-----+------+-----+
8bit 3bit 8bit 3bit 8bit 3bit 4bit 3bit
. . .
. RecursiveUnit 1...... .....
.
RecursiveUnit 2 ...
]]></artwork>
</figure>
<t>BFR R parses the RBS Address as shown in <xref target="FIG-E-R2" format="default"/> using its
RBS BIFT of N=3 BP entries shown in <xref target="FIG-E-R-BIFT" format="default"/>.</t>
<figure anchor="FIG-E-R2">
<name>RBS Address processed by BFR-R</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
.............. RecursiveUnit ............
. .
+-------+-----+-----+--------------------+-----+
|Tlen:32|R:011|L1:18|S:011000000110010001|E:001|
+-------+-----+-----+--------------------+-----+
8bit 3bit 8bit 18bit 3bit
. . .
. RecursiveUnit 1... .....
.
RecursiveUnit 2 ...
]]></artwork>
</figure>
<t>Because there are two recursive BP set in the BitString for R,
one for BFR S and one for BFR E, one Length field L1 is required in
the AddressingField, indicating the length of the
RecursiveUnit 1 for BFR S, followed by the remainder of
the RBS address being the RecursiveUnit 2 for BFR E.</t>
<figure anchor="FIG-E-R-BIFT">
<name>RBS BIFT on BFR R</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+---------+
|BP|Recursive|Adjacency|
+--+---------+---------+
| 1| 1| B |
+--+---------+---------+
| 2| 1| S |
+--+---------+---------+
| 3| 1| E |
+--+---------+---------+
]]></artwork>
</figure>
<t>BFR R accordingly creates one copy for BFR S using
RecursiveUnit 1, and only copy for BFR E using
RecursiveUnit 2, updating Padding accordingly for each
copy.</t>
</section>
<section anchor="bfr-s" numbered="true" toc="default">
<name>BFR S</name>
<t>BFR S receives from BFR B the packet and parses the
RBS address as shown in <xref target="FIG-E-S" format="default"/> using its RBS BIFT
of N=3 BP shown in <xref target="FIG-E-S-BIFT" format="default"/>.</t>
<figure anchor="FIG-E-S">
<name>RBS Address processed by BFR-S</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
.... RecursiveUnit ....
. .
+-------+-----+----+-----+------+-----+
|Tlen:18|S:011|L1:3|C:001|D:0001|Pad:6|
+-------+-----+----+-----+------+-----+
8bit 3bit 8bit 3bit 4bit 3bit
. . . .
.... ......
RecursiveUnit 1 . .
.
RecursiveUnit 2 .......
]]></artwork>
</figure>
<figure anchor="FIG-E-S-BIFT">
<name>RBS BIFT on BFR-S</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+---------+
|BP|Recursive|Adjacency|
+--+---------+---------+
| 1| 1| R |
+--+---------+---------+
| 2| 1| C |
+--+---------+---------+
| 3| 1| D |
+--+---------+---------+
]]></artwork>
</figure>
<t>BFR S accordingly sends one packet copy with RecursiveUnit 1
in the RBS address to BFR C and a second packet copy with
RecursiveUnit 2 to BFR D.</t>
</section>
<section anchor="bfr-c" numbered="true" toc="default">
<name>BFR C</name>
<t>BFR C receives from BFR S the packet and parses the
RBS address according to its N=3 BP entries BIFT (shown in
<xref target="FIG-E-C-BIFT" format="default"/>) as shown in <xref target="FIG-E-C" format="default"/>.</t>
<figure anchor="FIG-E-C">
<name>RBS Address processed by BFR-C</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-------+-----+-----+
|Tlen:3 |C:001|Pad:5|
+-------+-----+-----+
8bit 3bit 5bi
]]></artwork>
</figure>
<figure anchor="FIG-E-C-BIFT">
<name>RBS BIFT on BFR-C</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+-------------+
|BP|Recursive| Adjacency|
+--+---------+-------------+
| 1| 1| S |
+--+---------+-------------+
| 2| 1| D |
+--+---------+-------------+
| 3| 0| local_decap|
+--+---------+-------------+
]]></artwork>
</figure>
<t>BFR S accordingly creates one packet copy for BP 3
where the RBS address encapsulation is disposed of,
and the packet is ultimately forwarded to Client 2,
for example because of an IP multicast payload
for which the multicast flow overlay identifies
Client 2 as an interested receiver, as in BIER/BIER-TE.</t>
<t>To avoid having to use an IP flow overlay, the BIFT
could instead have one BP allocated for every non-RBS
destination, in this example BP 3 would then explicitly
be allocated for Client 2, and instead of disposing
of the RBS address encapsulation, BFR C would
impose or rewrite a unicast encapsulation to make the packet
become a unicast packet directed to Client 2. This option
is not further detailed in this version of the document.</t>
</section>
<section anchor="bfr-d" numbered="true" toc="default">
<name>BFR D</name>
<t>The procedures for processing of the packet on BFR D
are very much the same as on BFR C. <xref target="FIG-E-D" format="default"/> shows
the RBS address at BFR D, <xref target="FIG-E-D-BIFT" format="default"/> shows
the N=4 bit RBS BIFT of BFR D.</t>
<figure anchor="FIG-E-D">
<name>RBS Address processed by BFR-D</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-------+------+-----+
|Tlen:4 |D:0001|Pad:4|
+-------+------+-----+
8bit 4bit 4bit
]]></artwork>
</figure>
<figure anchor="FIG-E-D-BIFT">
<name>RBS BIFT on BFR-D</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+-------------+
|BP|Recursive| Adjacency|
+--+---------+-------------+
| 1| 1| S |
+--+---------+-------------+
| 2| 1| C |
+--+---------+-------------+
| 3| 1| E |
+--+---------+-------------+
| 4| 0| local_decap|
+--+---------+-------------+
]]></artwork>
</figure>
</section>
<section anchor="bfr-e" numbered="true" toc="default">
<name>BFR E</name>
<t>The procedures for processing of the packet on BFR E
are very much the same as on BFR C and D. <xref target="FIG-E-E" format="default"/> shows
the RBS address at BFR D, <xref target="FIG-E-E-BIFT" format="default"/> shows
the N=E bit RBS BIFT of BFR E.</t>
<figure anchor="FIG-E-E">
<name>RBS Address processed by BFR-E</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+-------+-----+-----+
|Tlen:3 |E:001|Pad:5|
+-------+-----+-----+
8bit 3bit 5bit
]]></artwork>
</figure>
<figure anchor="FIG-E-E-BIFT">
<name>RBS BIFT on BFR-E</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
+--+---------+-------------+
|BP|Recursive| Adjacency|
+--+---------+-------------+
| 1| 1| R |
+--+---------+-------------+
| 2| 1| D |
+--+---------+-------------+
| 3| 0| local_decap|
+--+---------+-------------+
]]></artwork>
</figure>
</section>
</section>
<section anchor="rbs-forwarding-pseudocode" numbered="true" toc="default">
<name>RBS forwarding Pseudocode</name>
<t>The following example RBS forwarding Pseudocode assumes
the reference encoding of bit-accurate length of BitStrings
and RecursiveUnits as well as 8-bit long TotalLen and AddressingField
Lengths. All packet field addressing and address/offset calculations
is therefore bit-accurate instead of byte accurate (which is what most
CPU memory access today is).</t>
<figure anchor="FIG-PSEUDOCODE">
<name>RBS address forwarding Pseudocode</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
void ForwardRBSPacket (Packet)
{
RBS = GetPacketMulticastAddr(Packet);
Total_len = RBS;
Rsv = Total_len + length(Total_Len);
BitStringA = Rsv + length(Rsv);
AddressingField = BitStringA + BIFT.entries;
// [1] calculate number of recursive bits set in BitString
CopyBitString(*BitStringA, *RecursiveBits, BIFT.entries);
And(*RecursiveBits,*BIFTRecursiveBits, BIFT.entries);
N = CountBits(*RecursiveBits, BIFT.entries);
// Start of first RecursiveUnit in RBS address
// After AddressingField array with 8-bit length fields
RecursiveUnit = AddressingField + (N - 1) * 8;
RemainLength = *Total_len - length(Rsv)
- BIFT.entries;
Index = GetFirstBitPosition(*BitStringA);
while (Index) {
PacketCopy = Copy(Packet);
if (BIFT.BP[Index].recursive) {
if(N == 1) {
RecursiveUnitLength = RemainLength;
} else {
RecursiveUnitLength = *AddressingField;
N--;
AddressingField += 8;
RemainLength -= RecursiveUnitLength;
RemainLength -= 8; // 8 bit of AddressingField
}
RewriteRBS(PacketCopy, RecursiveUnit, RecursiveUnitLength);
SendTo(PacketCopy, BIFT.BP[Index].adjacency);
RecursiveUnit += RecursiveUnitLength;
} else {
DisposeRBSheader(PacketCopy);
SendTo(PacketCopy, BIFT.BP[Index].adjacency);
}
Index = GetNextBitPosition(*BitStringA, Index);
}
]]></artwork>
</figure>
<t>Explanations for <xref target="FIG-PSEUDOCODE" format="default"/>.</t>
<t>RBS is the (bit accurate) address of the RBS address in packet
header memory. BitStringA is the address of the RBS address
BitString in memory. length(Total_Len) and length(Rsv) are the bit length of the two RBS
address fields, e.g.: 8 bit and 0 bit for the reference encoding.</t>
<t>The BFR local BIFT has a total number of BIFT.entries
addressable BP 1...BIFTentries. The BitString therefore
has BIFT.entries bits.</t>
<t>BIFT.RecursiveBits is a BitString pre-filled by the control
plane with all the BP with the recursive flag set. This is constructed
from the Recursive flag setting of the BP of the BIFT. The
code starting at [1] therefore counts the number of
recursive BP in the packets BitString.</t>
<t>Because the AddressingField does not have an entry for the
last (or only) RecursiveUnit, its length has to be calculated
by taking TotalLen into account.</t>
<t>RewriteRBS needs to replace RBS address with the
RecursiveUnit address, keeping only Rsv, recalculating
TotalLen and adding appropriate Padding.</t>
<t>For non-recursive BP, the Pseudocode assumes disposition of the
RBSheader. This is not strictly necessary but non-disposing
cases are outside of scope of this version of the document.</t>
</section>
<section anchor="operational-and-design-considerations-informational" numbered="true" toc="default">
<name>Operational and design considerations (informational)</name>
<section anchor="comparison" numbered="true" toc="default">
<name>Comparison with BIER-TE / BIER</name>
<t>This section discusses informationally, how and where
CGM2 can avoid different complexities of BIER/BIER-TE,
and where it introduces new complexities.</t>
<section anchor="eliminating-the-need-for-large-bift" numbered="true" toc="default">
<name>Eliminating the need for large BIFT</name>
<t>In a BIER domain with M BFER, every BFR requires M
BIFT entries. If the supported BSL is N and M > 2 ^ N,
then S = (M / 2 ^ N) set indices (SI) are required,
and S copies of the packet have to be sent by the BFIR
to reach all targeted BFER.</t>
<t>In CGM2, the number of BIFT entries does not need
to scale with the number of BFER or paths through
the network, but can be limited to only the number
of L2 adjacencies of the BFR. Therefore CGM2 requires
minimum state maintenance on each BFR, and multiple
SI are not required.</t>
</section>
<section anchor="reducing-number-of-duplicate-packet-copies-across-bfr" numbered="true" toc="default">
<name>Reducing number of duplicate packet copies across BFR</name>
<t>If the total size of an RBS encoded delivery tree is
larger than a supported maximum RBS header size, then
the CGM2 controller simply needs to divide the tree
into multiple subtrees, each only addressing a part
of the BFER (leaves) of the target tree and pruning
any unnecessary branches.</t>
<figure anchor="FIG-SMPLT">
<name>Simple Topology Example</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
B1
/ \
B2 B3
/ \ / \
/ \/ \
B4 B5 B6
/..| / \ |..\
B7..B99 B100..B200 B201...B300
]]></artwork>
</figure>
<t>Consider the simple topology in <xref target="FIG-SMPLT" format="default"/> and a multicast packet
that needs to reach all BFER B7...B300. Assume that
the desired maximum RBM header size is such that a
RBS address size of <= 256 bits is desired. The CGM2
controller could create an RBS address
B1=>B2=>B4=>(B7..B99), for a first packet, an
RBS address B1=>B3=>B5=>(B100..B200) for a second
packet and a third RBS address B1=>B3=>B6=>B201...B300.</t>
<t>The elimination of larger BIFT state in BFR
through multiple SI in BIER/BIER-TE does come at
the expense of replicating initial hops of a tree
in RBS addresses, such as in the example the encoding
of B1=>B3 in the example.</t>
<t>Consider that the assignment of BFIR-ids with BIER
in the above example is not carefully engineered. It is
then easily possible that the BFR-ids for B7..B99 are not
sequentially, but split over a larger BFIR-id space.
If the same is true for all BFER, then it is possible
that each of the three BFR B4,B5 and B6 has attached
BFER from three different SI and one may need to send
for example three multiple packets to B7 to
address all BFER B7..B99 or to B5 to address all
B100..B200 or B6 to address all B201...B300. These
unnecessary duplicate packets across B4, B5 or B6 are
because of the addressing principle in BIER and are not
necessary in CGM2, as long as the total length of an RBS
address does not require it.</t>
<t>For more analysis, see <xref target="analysis" format="default"/>.</t>
</section>
<section anchor="complexities" numbered="true" toc="default">
<name>BIER-TE forwarding plane complexities</name>
<t>BIER-TE introduces forwarding plane complexities to allow
reducing the BSL required. While all of these
could be supported / implemented with CGM2, this
document contends that they are not necessary, therefore
providing significant overall simplifications.</t>
<ul spacing="normal">
<li>BIER-TE supports multiple adjacencies in a single BIFT Index
to allow compressing multiple adjacencies into a single Index
for traffic that is known to always require replications
to all those adjacencies (such as when flooding TV traffic).</li>
<li>BIER-TE support ECMP adjacencies which
have to calculate which out of 2 or more possible adjacencies
a packet should be forwarded to.</li>
<li>BIER-TE supports special Do-Not-Clear (DNC) behavior of
adjacencies to permit reuse of such a bit for adjacencies
on multiple consecutive BFR. This behavior specifically
also raises the risk of looping packets.</li>
</ul>
</section>
<section anchor="bier-te-controller-complexities" numbered="true" toc="default">
<name>BIER-TE controller complexities</name>
<t>BIER-TE introduces BIER-TE controller plane mechanisms
that allow to reuse bits of the flat BIER-TE BitStrings
across multiple BFR solely to reduce the number of BP
required but without introducing additional complexities
for the BIER-TE forwarding plane.</t>
<ul spacing="normal">
<li>Shared BP for all Leaf BFR.</li>
<li>Shared BP for both Interfaces of p2p links.</li>
<li>Shared bits for multi-access subnets (LANs).</li>
</ul>
<t>These bit-sharing mechanisms are unnecessary
and inapplicable to CGM2 because there is no need to
share BP across the BIFT of multiple BFR.</t>
</section>
<section anchor="bier-te-specification-complexities" numbered="true" toc="default">
<name>BIER-TE specification complexities</name>
<t>The BIER-TE specification distinguishes between forward (link scope)
and routed (underlay routed) adjacencies to highlight, explain
and emphasize on the ability of BIER-TE to be deployed in an overlay fashion
especially also to reduce the necessary BSL, even
when all routers in the domain could or do support BIER-TE.</t>
<t>In CGM2, routed adjacencies are considered to be only
required in partial deployments to forward across non-CGM2
enabled routers. This specification does therefore not
highlight link scope vs. routed adjacencies as core
distinct features.</t>
</section>
<section anchor="forwarding-plane-complexity" numbered="true" toc="default">
<name>Forwarding plane complexity</name>
<t>CGM2 introduces some more processing calculation steps to extract
the BitString that needs to be examined by a BFR from
the RBS address. These additional steps are considered
to be non-problematic for todays programmable
forwarding planes such as P4.</t>
<t>Whereas BIER-TE clears bit on each hops processing,
CGM2 rewrites the address on every hop by extracting the recursive
unit for the next hop and make it become the packet copies
address. This rewrite shortens the RBS address. This hopefully
has only the same complexity as (tunnel) encapsulations/decapsulations
in existing forwarding planes.</t>
</section>
</section>
<section anchor="cgm2-rbs-controller-considerations" numbered="true" toc="default">
<name>CGM2 / RBS controller considerations</name>
<t>TBD. Any aspects not covered in <xref target="comparison" format="default"/>.</t>
</section>
<section anchor="analysis" numbered="true" toc="default">
<name>Analysis of performance gain with CGM2</name>
<t>TBD: Comparison of number of packets/header sizes required
in large real-world operator topology between BIER/BIER-TE and CGM2.
Analysis: Gain in dense topology.</t>
<section anchor="reference-topology" numbered="true" toc="default">
<name>Reference topology</name>
<t>Reference topology description:</t>
<ol spacing="normal" type="1"><li>Typical topology of Beijing Mobile in China.</li>
<li>All zones dual homing access to backbone.</li>
<li>Core layer: 4 nodes full mesh connected</li>
<li>Aggregation layer: 8 nodes are divided into two layers, with full
interconnection between the layers and dual homing access to
the core layer on the upper layer.</li>
<li>Aggregation rings: 8 rings, 6 nodes per ring</li>
<li>Access rings: 3600 nodes, 18 nodes per ring.</li>
</ol>
<figure anchor="FIG-REFTOPO">
<name>Reference Topology</name>
<artwork name="" type="" align="left" alt=""><![CDATA[
┌──────┐ ┌──────┐
│ ├──────────┤ │
/└──────┘\ /└──────┘\ Interconnected
/ / | \ \ / / | \ \ BackBone
┌──────┐/ / | \ \ / / | \ \┌──────┐
│ │ / | \ \ / / | \ │ │
└───┬──┘ / | \ \/ / | \ └─┬────┘
│ / | \ /\ / | \ │
┌──┴───┐ | / \ | ┌──┴───┐
│ │------------+ \/ +------------│ │
└──────┘\ | /\ | /└──────┘
\ | / \ | /
\ ┌──────┐/ \┌──────┐ /
\│ ├──────┤ │/
└───┬──┘ └───┬──┘
│ \ / │ Dual Return Access
│ \ / │
│ \ / │
│ / │
│ / \ │
┌─┴───┐/ \┌───┴─┐
│ ├─────┤ │
└─┬───┘\ /└───┬─┘
│ \ / │ Core Layer
│ / │
│ / \ │
┌─┴───┐/ \┌───┴─┐
/│ ├─────┤ │\
/ └──┬──┘\ /└──┬──┘ \
/ │\ \ / /│ \ Zone1
/ │ \ \ / │ \
/ │ \ / \ / │ \
/ +----│---+ +---│----+ \
/ / │ \ / │ \ \
/ / │ + │ \ \
/ / │ / \ │ \ \
┌───┐/ ┌┴──┐/ \┌──┴┐ \┌───┐
│ │\ /│ │ │ │\ /│ │
└─┬─┘ \ / └─┬─┘\ /└─┬─┘ \ / └─┬─┘
│ \ / │ \ / │ \ / │ Aggregation
│ \/ │ / │ \/ │ Layer
│ /\ │ / \ │ /\ │
┌─┴─┐ / \ ┌─┴─┐/ \┌─┴─┐ / \ ┌─┴─┐
│ │-- --│ │ │ │-- --│ │
└───┘ └───┘\ /└───┘\ └───┘
/ | \ \ / / | \
/ | \ \ / | \
/ | / \/ | \
/ +--|--+ \/+---|---+ \
/ / | /\ | \ \
┌───┐ ┌┴──┐/ \┌───┐ ┌───┐ ASBR
│ │ │ │ │ │ │ │
└─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘
│ │ │ │
│ │ │ │
┌─┴─┐ ┌─┴─┐ ┌─┴─┐ ┌─┴─┐
│ │ │ │ │ │ │ │
└─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘
│ │ │ │
│ │ 8Rings │ │
┌─┴─┐ ┌─┴─┐ ...┌─┴─┐ ┌─┴─┐
│ │---│ │ │ │---│ │
----└───┘ └───┘ └───┘\ └───┘
/ / \ \ | \ \ \ | \
/ / \ \ | \ \ +---|-+ \
/ / \ +-|---+\ \ | \ \
/ / \ | \\ \ | \ \
/ / \ | \\ \ | \ \
/ / \ | \\ \ | \ \
┌───┐ ┌───┐ ┌───┐ ┌───┐ ┌───┐ ┌───┐ CSBR
│ │ │ │ │ │ │ │ │ │ │ │
└─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘
│ │ Access │ │ │ │
│ │ Rings │ │ │ │
┌─┴─┐ ┌─┴─┐ ... ┌─┴─┐ ┌─┴─┐ ┌─┴─┐ ┌─┴─┐
│ │ │ │ │ │ │ │ │ │ │ │
└─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘ └─┬─┘
│ │ │ │ │ │