Danny Amor's Online Library/Danny Amor's Online Library - Volume 1.iso/html/faqs/faq/cell-relay-faq/part2

Text File | 1995-07-25 | 36.8 KB | 781 lines

Subject: comp.dcom.cell-relay FAQ: ATM, SMDS, and related technologies (part 2/2)
Newsgroups: comp.dcom.cell-relay,comp.answers,news.answers
From: [e-mail] (Carl Symborski)
Date: 10 Sep 1994 01:01:25 -0400

Archive-name: cell-relay-faq/part2
Last-modified: 1994/09/06


---This is part 2/2---

Of course, understand that the specified behavior may closely match the
way the traffic was going to behave anyway.  But by knowing precisely
how the traffic is going to behave, it is possible to allocate resources
inside the network such that guarantees about availability of bandwidth
and maximum delays can be given.


Brief Tutorial:
Assume some switches connected together which are carrying traffic.
The problem to actually deliver the grade of service that has been promised,
and that people are paying good money for. This requires some kind of resource
management strategy, since congestion will be by far the greatest factor
in data loss. You also need to charge enough to cover you costs and make a
profit, but in such a way that you attract customers. There are a number
of parameters and functions that need to be considered:

PARAMETERS
----------
There are lots of traffic parameters that have been proposed for resource
management. The more important ones are:
    mean bitrate
    peak bitrate
    variance of bitrate
    burst length
    burst frequency
    cell-loss rate
    cell-loss priority
    etc. etc.

These parameters exist in three forms:
    (a) actual
    (b) measured, or estimated
    (c) declared (by the customer)

FUNCTIONS
---------
(a) Acceptance Function
-----------------------
Each switch has the option of accepting a virtual circuit request based on
the declared traffic parameters as given by the customer. Acceptance is
given if the resulting traffic mix will not cause the switch to not
achieve its quality of service goals.

The acceptance process is gone through by every switch in a virtual
circuit. If a downstream switch refuses to accept a connection, an
alternate route might be tried.

(b) Policing Function
---------------------
Given that a switch at the edge of the network has accepted a virtual
circuit request, it has to make sure the customer equipment keeps its
promises. The policing function in some way estimates the the parameters
of the incoming traffic and takes some action if they measure traffic
exceeding agreed parameters. This action could be to drop the cells, mark
them as being low cell-loss priority, etc.

(c) Charging Function
---------------------
The function most ignored by traffic researchers, but perhaps the most
important for the success of any service! Basically, this function
computes a charge from the estimated and agreed traffic parameters.

(d) Traffic Shaping Function
----------------------------
Traffic shaping is something that happens in the customer premise equipment.
If the Policing function is the policeman, and the charging function is the
judge, then the traffic shaper is the lawyer. The traffic shaper uses
information about the policing and charging functions in order to change
the traffic characteristics of the customer's stream to get the lowest
charge or the smallest cell-loss, etc.

For example, an IP router attached to an ATM network might delay some
cells slightly in order to reduce the peak rate and rate variance without
affecting throughput. An MPEG codec that was operating in a situation
where delay wasn't a problem might operate in a CBR mode.



SUBJECT:  D4) * What is happening with signalling standards for ATM?

        The Signaling Sub-Working Group (SWG) of the ATM Forum's Technical
Committee has completed its implementation agreement on signaling at the
ATM UNI (summer 1993).  The protocol is based on Q93B with extensions
to support point-to-multipoint connections.  Agreements on addressing specify
the use of GOSIP-style NSAPs for the (SNPA) address of an ATM end-point
at the Private UNI, and the use of either or both GOSIP-style NSAPs and/or
E.164 addresses at the Public UNI.  The agreements have been documented
as part of the UNI 3.0 specification.

Additionally, the ANSI T1S1 as well as the ITU-T sudygroup XI are concerned
with ATM signalling.  In the latter half of 1993 a couple of things happened:
 1) The ITU finally agreed to modify its version of Q93B to more closely
    to align it with that specified in the ATM Forum's UNI 3.0 specification.
    The remaining variations included some typos which the ITU Study Group
    found in the Forum's specification.  Also, some problems were solved
    differently.  Aligned yes, but the changes could still cause
    incompatibilities with UNI 3.0.
 2) Given the above, the ATM Forum's signalling SWG decided to modify the
    Forum's specification to close the remaining gap and align it with the
    ITU.  The end result may be declared as errata to UNI 3.0 or defined
    as a UNI 3.1 specification

The biggest change is with SSCOP.  UNI 3.0 references the draft ITU-T SSCOP
documents (Q.SAAL).  However UNI 3.1 will reference the final ITU Q.21X0
specifications.  These two secifications are *not* interoperable so there
will be no backwards compatibility between UNI 3.0 and UNI 3.1.

Beyond that folks are looking forward to a UNI 4.0 which would include
things like ABR service class among other things.

The ATM Forum also has a Private-NNI SWG.  Their objective is to define an
interface between one Switching System (SS) and another, where each SS is a
group of one or more switches, such that the specification can be applied to
both the switch-to-switch case and the network-to-network cases.

Tentative schedule for the ATM Forum's work:

UNI 3.1     September 94
PNNI 1.0    March 95
B-ICI 2.0    March 95
UNI 4.0 draft    June 95


SUBJECT:  D5)   What is VPI and VCI?

        ATM is a connection orientated protocol and as such there is a
connection identifier in every cell header which explicitly associates a cell
with a given virtual channel on a physical link.  The connection identifier
consists of two sub-fields, the Virtual Channel Identifier (VCI) and the
Virtual Path Identifier (VPI).  Together they are used in multiplexing,
demultiplexing and switching a cell through the network.  VCIs and VPIs are
not addresses.  They are explicitly assigned at each segment (link between ATM
nodes) of a connection when a connection is established, and remain for the
duration of the connection.  Using the VCI/VPI the ATM layer can
asynchronously interleave (multiplex) cells from multiple connections.


SUBJECT:  D6)   Why both VPI *and* VCI?

        The Virtual Path concept originated with concerns over the cost of
controlling BISDN networks.  The idea was to group connections
sharing common paths through the network into identifiable units (the Paths).
Network management actions would then be applied to the smaller number of
groups of connections (paths) instead of a larger number of individual
connections (VCI).  Management here including call setup, routing, failure
management, bandwidth allocation etc.  For example, use of Virtual Paths in
an ATM network reduces the load on the control mechanisms because the function
needed to set up a path through the network are performed only once for all
subsequent Virtual Channels using that path.  Changing the trunk mapping
of a single Virtual Path can effect a route change for every Virtual Channel
using that path.

Now the basic operation of an ATM switch will be the same, no matter if it is
handling a virtual path or virtual circuit.  The switch must identify on
the basis of the incomming cell's VPI, VCI, or both, which output port to
forward a cell received on a given input port.  It must also determine what
the new values the VPI/VCI are on this output link, substituting these
new values in the cell.


SUBJECT:  D7)   How come an ATM cell is 53 bytes anyway?

        ATM cells are standardized at 53 bytes because it seemed like a
good idea at the time!  As it turns out, during the standardization process
a conflict arose within the CCITT as to the payload size within an ATM
cell.  The US wanted 64 byte payloads because it was felt optimal for
US networks.  The Europeans and Japanese wanted 32 payloads because it was
optimal for them.  In the end 48 bytes was chosen as a compromise.  So
48 bytes payload plus 5 bytes header is 53 bytes total.

The two positions were not chosen for similar applications however.
US proposed 64 bytes taking into consideration bandwidth utilization for
data networks and efficient memory transfer (length of payload should be
a power of 2 or at least a multiple of 4). 64 bytes fit both requirements.

Europe proposed 32 bytes taking voice applications into consideration. At
cell sizes >= 152, there is a talker echo problem. Cell sizes between 32-152
result in listener echo. Cell sizes <= 32 overcome both problems, under ideal
conditions.

CCITT chose 48 bytes as a compromise. As far as the header goes, 10% of
payload was perceived as an upper bound on the acceptable overhead, so 5 bytes
was chosen.


SUBJECT:  D8) * How does AAL5 work?

        Here is is a very simplified view of AAL5 and AALs in general.
AAL5 is a mechanism for segmentation and reassembly of packets.  That is,
it is a rulebook which sender and receiver agree upon for taking a long
packet and dividing it up into cells.  The sender's job is to segment the
packet and build the set of cells to be sent.  The receiver's job is to
verify that the packet has been received intact without errors and to
put it back together again.

AAL5 (like any other AAL) is composed of a common part (CPCS) and a service
specific part (SSCS). The common part is further composed of a convergence
sublayer (CS) and a segmentation and reassembly (SAR) sublayer.

         +--------------------+
         |                    | SSCS
         +--------------------+
         |        CS          |
         | ------------------ | CPCS
         |       SAR          |
         +--------------------+

SAR segments higher a layer PDU into 48 byte chunks that are fed into
the ATM layer to generate 53 byte cells (carried on the same VCI).  The
payload type in the last cell (i.e., wherever the AAL5 trailer is) is marked
to indicate that this is the last cell in a packet.  (The receiver may
assume that the next cell received on that VCI is the beginning of a
new packet.)

CS provides services such as padding and CRC checking. It takes an SSCS
PDU, adds padding if needed, and then adds an 8-byte trailer such that
the total length of the resultant PDU is a multiple of 48. The trailer
consist of a 2 bytes reserved, 2 bytes of packet length, and 4 bytes of CRC.

SSCS is service dependent and may provide services such as assured
data transmission based on retransmissions. One example is the SAAL
developed for signalling. This consists of the following:

         +--------------------+
         |       SSCF         |
         | ------------------ | SSCS
         |       SSCOP        |
         +--------------------+
         |        CS          |
         | ------------------ | CPCS
         |       SAR          |
         +--------------------+

SSCOP is a general purpose data transfer layer providing, among other
things, assured data transfer.

SSCF is a coordination function that maps SSCOP services into those
primitives needed specifically for signalling (by Q.2931). Different
SSCFs may be prescribed for different services using the same SSCOP.

The SSCS may be null as well (e.g. IP-over-ATM or LAN Emulation).

There are two problems that can happen during transit.  First, a
cell could be lost.  In that case, the receiver can detect the problem
either because the length does not correspond with the number of cells
received, or because the CRC does not match what is calculated.  Second,
a bit error can occur within the payload.  Since cells do not have any
explicit error correction/detection mechanism, this cannot be detected
except through the CRC mismatch.

Note that it is up to higher layer protocols to deal with lost and
corrupted packets.  This can be done by using a SSCS which supports
assured data transfer, as discussed above.


SUBJECT:  D9) * What are the diffferences between Q.93B, Q.931, and Q.2931?

        Essentially, Q.93B is an enhanced signalling protocol for call
control at the Broadband-ISDN user-network interface, using the ATM
transfer mode.  The most important difference is that unlike Q.931
which manages fixed bandwidth circuit switched channels, Q.93B has
to manage variable bandwidth virtual channels.  So, it has to deal
with new parameters such as ATM cell rate, AAL parameters (for
layer 2), broadband bearer capability, etc.  In addition, the ATM
Forum has defined new functionality such as point-to-multipoint
calls.  The ITU-T Recommendation will specify interworking
procedures for narrowband ISDN.

Note that as of Spring 1994, Q.93B has reached a state of maturity
susfficient to justify a new name, Q.2931 for its publised official
designation.



SUBJECT:  D10)   What is a DXI?

        The ATM DXI (Data Exchange Interface)is basically the functional
equivalent of the SMDS DXI.  Routers will handle frames and packets but not
typically fragment them into cells; DSUs will fragment frames into cells as
the information is mapped to the digital transmission facility.

The DXI, then, provides the standard interface between routers and DSUs
without requiring a bunch of proprietary agreements.  The SMDS DXI is
simple 'cause the router does the frame (SMDS level 3) and the DSU does
the cells (SMDS level 2).  The ATM DXI is a little more complicated
since it has to accomomodate AAL3/4 and/or AAL5 (possibly concurrently).



SUBJECT:  D11)   What is Goodput?

        When ATM is used to tranasport cells originating from higher-level
protocols (HLP), an important consideration is the impact of ATM cell loss
on that protocol or at least the segmentation processs.  ATM cell loss can
cause the effective throughput of some HLPs to be arbitrarily poor depending
on ATM switch buffer size, HLP congestion control mechanisms, and packet size.

This occurs because during congestion for example, and ATM switch buffer can
overflow which will cause cells to be dropped from multiple packets, runining
each such packet.  The preceding and the remaining cells from such packets,
which are ultimately discarded by the frame reassembly process in the receiver,
are nevertheless transmitted on an already congested link, thus wasting
valuable link bandwidth.

The traffic represented by these "bad" cells may be termed as BADPUT.
Correspondingly, the effective throughput, as determined by those cells which
are successfully recombined at the receiver, can be termed as GOODPUT.


SUBJECT:  D12)   What is LAN Emulation all about?

"LAN Emulation" is a work in progress in the ATM Forum.  There is not a
complete agreement on all aspects of LAN Emulation, but there is good agreement
on the requirements and general approach.  Here's the basics:

The organizations working on it say LAN Emulation is needed for two key reasons
1) Allow an ATM network to be used as a LAN backbone for hubs, bridges,
switching hubs (also sometimes called Ethernet switches or Token Ring switches)
and the bridging feature in routers.

2) Allow endstations connected to "legacy" LANs to communicate though a
LAN-to-ATM hub/bridge/switch with an ATM-attached device (a file server, for
example) without requiring the traffic to pass through a more complex device
such as a router.  Note that the LAN-attached device has a conventional,
unchanged protocol stack, complete with MAC address, etc.

LAN Emulation does not replace routers or routing, but provides a complementary
MAC-level service which matches the trend to MAC-layer switching in the hubs
and wire closets of large LANs.

The technical approach being discussed in the Forum among companies with
interest and expertise in this area include three elements:

1) Multicast/broadcast support
Since almost all LAN protocols depend on broadcast or multicast packet
delivery, an ATM LAN must provide the same service. Ideally, this would use
some sort of multipoint virtual circuit facility.

2) Mac address to ATM address resolution.
There are two basic variations being discussed:
a) do an ARP-like protocol to ask for a mapping from Mac address to ATM address
b) send packets to some sort of directory or cache server that sends the
destination ATM address back to the source as a sort of side effect of
delivering the packet.

3) Switched Virtual Circuit Management
It is generally desireable (for scalabilitiy, quality of service, etc) to
set up point-to-point virtual circuits between endpoints that want to
communicate with each other (client to file server, for example) once
the two atm addresses are known.  To make this work in the existing legacy LAN
environment, we don't have the freedom to push knowledge or management of
these virtual circuits up above the MAC level (no protocol changes, remember?)
so the logic to resovle for an ATM address and set up a virtual circuit on
demand must be in the LAN Emulation layer.  This would include recognising
when an SVC to another ATM endpoint already existed, so that the same circuit
could be used for other traffic.

4) Mac definition
The actual packet format would be some variant of the packets used on existing
networks.  For example, a packet leaving an Ethernet to go over a virtual
circuit to an ATM-attached file server would probably be carried directly
over AAL5, with some additional control information.


SUBJECT:  D13)   Information about the Classical IP over ATM approach.

        RFC1483 defines the encapsulation of IP datagrams directly in AAL5.
Classical IP and ARP over ATM, defined in RFC1577, is targetted towards making
IP run over ATM in the most efficient manner utilizing as many of the
facilities of ATM as possible.  It considers the application of ATM as a
direct replacement for the "wires" and local LAN segments connection IP
end-stations and routers operating in the "classical" LAN-based paradigm.
A comprehensive document, RFC1577 defines the ATMARP protocol for logical
IP subnets (LISs). Within an LIS, IP addresses map directly into ATM Forum UNI
3.0 addresses.  For communicating out a LIS, an IP router must be used -
following the classical IP routing mode.  Reference the RFCs for a full
description of this approach.


SUBJECT:  D14)   Classical IP and LAN/MAC Emulation approaches compaired.

        The IETF scheme defines an encapsulation and an address resolution
mechanism.  The encapsulation could be applied to a lot of LAN protocols
but the address resolution mechanism is specifically defined (only) for IP.
Further, the IETF has not (yet) defined a multicast capability.  So, those
protocols which require multicast definitely cannot adapt the IETF scheme for
their own use.

The purpose behind the ATM Forum's LAN-Emulation effort is to allow
existing applications (i.e., layer-3 and above protocol stacks) to
run *with no changes* over ATM.  Thus, the mapping for all protocols
is already defined.  In a PC environment, such applications tend to
run over an NDIS/ODI/etc. interface.  The LAN-Emulation effort aims
to be able to be implementable underneath the NDIS/ODI-type interface.

In contrast to LAN-Emulation, the IETF's scheme will allow IP to make
better use of ATM capabilities (e.g., the larger MTU sizes), and for
unicast traffic will be more efficient than having the additional
LAN-Emulation layer.  However, the Classical draft suggests that IP
multicast (e.g., the MBONE) will have to be tunnelled over ATM; I
suspect this will be less efficient than LAN-Emulation.

For better or worse, I think both are going to be used.  So, vendors
may have to do both.  The worse part is extra drivers (or extra
code in one driver that does both).  The better part is that all existing
LAN applications can use one (LAN Emulation), and over time (as their mapping
to ATM is fully defined) can transition to use the other (IETF Scheme).

I would summarize LAN-Emulation as follows:

The advantage of LAN-Emulation is that the applications don't know
they're running over ATM.  The disadvantage of LAN-Emulation is also
that the applications don't know they're running over ATM.


SUBJECT: D15) + Whats the difference between SONET and SDH?

    SONET and SDH are very close, but with just enough differences that
they don't really interoperate. Probably the major difference between them
is that SONET is based on the STS-1 at 51.84 Mb/s (for efficient carrying
of T3 signals), and SDH is based on the STM-1 at 155.52 Mb/s (for efficient
carrying of E4 signals).  As such, the way payloads are mapped into these
respective building blocks differ (which makes sense, given how the European
and North American PDHs differ).  Check the September 1993 issue of IEEE
Communications Magazine for an overview article on SONET/SDH.

The following table shows how the US STS and the European STM levels
compare:

US        Europe       Bit Rate (total)

STS-1      --            51.84 Mb/s
STS-3     STM-1         155.52 Mb/s
STS-12    STM-4         622.08 Mb/s
STS-24    STM-8        1244.16 Mb/s
STS-48    STM-16       2488.32 Mb/s
STS-192   STM-64       9953.28 Mb/s

From a formatting perspective, however, OC-3/STS-3 != STM-1 even though
the rate is the same.  SONET STS-3c (i.e., STS-3 concatenated) is the
same as SDH STM-1, followed by STS-9c = STM-3c, etc.

There are other minor differences in overhead bytes (different places,
slightly different functionality, etc), but these shouldn't provide
many problems.  By the way, most physical interface chips that support SONET
also include a STM operation mode.  Switch vendors which use these devices
could then potentially support STS-3 and STM-1 for example.


SUBJECT: D16) + What is ABR?

The ATM Forum Traffic Management (TM) subworking group is working on
the definition of a new ATM service type called ABR which stands for
Available Bit Rate.  Using ABR traffic is not characterized using
peak cell rate, burst tolerance, et.al., and bandwidth reseverations are not
made.  Instead traffic is allowed into the network throttled by a flow
control type mechanism.  The idea is to provide fair sharing of network
bandwidth resources.

Currently in the ATM Forum there are two approaches being discussed (fought)
to implement ABR.  One is a "Credit Based" scheme which keeps count of the
occupied buffers for each VC at each downstream hop(s).  At each hop, the
upstream switch (or terminal) stops transmitting on an ABR
circuit if its downstream buffer occupancy reaches a per-VC limit.
At the downstream end, the switch (or terminal) sends back a "credit"
after forwarding some number of cells on each circuit.

Unlike the hop-by-hop Credit scheme, the "Rate Based" approach is an
end-to-end congestion avoidance mechanism.  An end system would be allowed
to send cells into the network up until a "congestion indicated" cell is
received from the network.  This congestion cell could be generated by
any component along the route which the VC follows.  The end system would
respond by stopping its cell flow and then do some kind of V-J slow-start
similar to that done with TCP today.

The consumer hope is that the ATM Forum specify a SINGLE approach and not
fail by specifying both approaches as "options" for the market to sort out
(and end up with something like the VHS and BETA video tape format battle).
Then again, there are always folks like IBM who will move their own standards
forward, like the 25mhz ATM interface format.  So maby there is a limit to
what an industry form can do.


 -----------------------------------------------------------------------------
TOPIC:     E)   TOPIC: ATM VS. XYZ TECHNOLOGY
-----------------------------------------------------------------------------
SUBJECT:  E1)   How does ATM differ from SMDS?

    SMDS is the Switched Multi-megabit Data Service, a service offering
interface from Bellcore.  SMDS provides a datagram service, where a packet has
about a 40-octet header plus up to 9188 octets of data. The packets themselves
may or may not be transported within the network on top of a connection-
oriented ATM service.  SMDS uses E.164 (ISDN) addresses.  Therefore SMDS is
a connectionless packet switched *service*, not a cell-relay service.

HOWEVER, the SMDS Subscriber Network Interface is currently defined to use
IEEE 802.6 Distributed Queue Dual Bus (DQDB) access across the SMDS
user-network interface.  DQDB itself *is* a form of cell relay.  The lower
layers of SMDS fragment the packets into cells with a 5-octet header and
48-octet payload.  The payload itself has a 2-octet header, 44-octets of data,
plus a 2-octet trailer.  An SMDS cell therefore is nearly identical in form
to an AAL3/4 cell.

Note that while DQDB is used as the access protocol, either DQDB or AAL3/4
may be used for the switch-to-switch interface.

The best source of (readable) information on SMDS is probably the SMDS
Interest Group (SIG), 480 San Antonio Road, Suite 100, Mountain View,
California 94040, USA; Tel [phone]; Fax [phone].

The SIG is in many ways similar to the ATM Forum, and cooperates with
it. Also, there is a European branch known as ESIG which is concerned
with adapting the American SIG documents to fit European network
architectures and regulations. SIG work is mostly based on Bellcore
SMDS TAs and such like, while ESIG aligns with ITU and ETSI standards.

-----------------------------------------------------------------------------
TOPIC:     F)   TOPIC: FREELY AVAILABLE REFERENCE IMPLEMENTATIONS
-----------------------------------------------------------------------------
SUBJECT:  F1)   What and where is VINCE?

         Vince has now on record as the first "publicaly available" software
source code in the ATM technology area.  This work was carried out by the
Research Networks section of the Center for Computational Science at the
Naval Research Laboratory, with support from the Advanced Research Projects
Agency and NAVSEA.  In the Grand Internet Tradition, these fine folks have
contributed their efforts to the community in support of further research.

VINCE RELEASE 0.6 ALPHA

Vince, the Vendor Independent Network Control Entity, is
publicly available (in source code form) as an
alpha release. Its primary function is to perform ATM
signalling and VC management tasks. It currently includes
a hardware module that allows it to run on Fore ASX-100(tm)
switches.  Other hardware modules are under development.

Vince consists of a core which provides basic ATM network
semantics, and modules to perform specific functions. Modules
included in this release are:

  spans  - module which intereroperates signalling and routing
           with Fore Systems ASX switch and various host interfaces.
           SPANS is (tm) Fore Systems, Inc.

  q93b   - an implementation of signalling as specified in the ATM
           Forum UNI 3.0 document.  The vince core includes sscop
           and aal5 in its protocol library.

  sim    - a software ATM switch or host that can be used to test
           signalling and routing implementations without ATM
           hardware.

  sroute - an address independent VC routing module.

The Vince distribution also contains a driver that uses spans for
signalling and supports the Fore Systems SBA-100 under SunOS(tm).

Work has already begun on a kernel version of Vince, which will
allow ATM Forum UNI signalling for hosts.  Also in development
are SNMP/ILMI support, interdomain routing, and support for other
switches.

The intent is to provide a redistributable framework which
allows for code sharing among ATM protocol developers.

Vince and its architecture document are available for
anonymous ftp at hsdndev.harvard.edu:pub/mankin

A mailing list for Vince developers and users can be joined
by sending mail to [e-mail].


-----------------------------------------------------------------------------
TOPIC:     G)   TOPIC: FLAMES AND RECURRING HOLY WARS
-----------------------------------------------------------------------------

         As with any News and/or email list, topics will be raised which
elicit a broad range of viewpoints.  Often these are quite polarized and yield
a chain of replies and counter topics which can span weeks and even months.
Typically without resolution or group consensus.  This section lists some
memorable (lengthy?) topic threads.

PLEASE NOTE that the idea here is not to re-kindle old flames, and not to
somehow pronounce some conclusion.  Rather, recorded here are are a few
pieces of the dialogue containing information which might be of some use.


SUBJECT:  G1)   Are big buffers in ATM switches needed to support TCP/IP?

         A recurring theme in 1993 concerned the suitability of ATM to
transport TCP/IP based traffic.  The arguments generally centered around the
possible need for ATM WAN switches to support very large buffers such that
TCP's reactive congestion control mechanism will work.  Points of contention
include: are big buffers needed, if so then where, and what exactly is the
TCP congestion control mechanism.

Undoubtedly, many of these discussions have been fueled by some 1993 studies
which reported that TCP works poorly over ATM because of the cell-discard
phenomenon coupled with TCP's congestion control scheme.

The longest thread on this subject started in the October 1993 timeframe and
ended in December under the subject of "Fairness on ATM Networks".
Generally folks expressed opinions in one of the following postures:

1) Big buffers are not needed at all....

  A few argued that if ATM VC's are provisioned and treated as fixed leased
  lines then ATM will be able to support TCP/IP just fine.  This means that
  you would need to subscribe to the maximum possible burst rate which would
  be very inefficient use of bandwidth since TCP is usually very bursty.

2) Put big buffers in routers and not ATM switches....

  If you are using wide-area links over ATM, then use a router smart enough
  not to violate the Call-Acceptance contract.  The call acceptance function
  should be such that it doesn't let you negotiate a contract that causes
  congestion.  Congestion should only occur when there is a fault in the
  network.  A router is quite capable of smoothing out bursts.  That is what
  they do right now when they operate over leased lines.  The advantage of
  an ATM connection replacing a leased line is that the connection parameters
  can be able to be renegotiated on the fly, so if your IP network (as
  opposed to the ATM network) is experiencing congestion, then it can request
  more bandwidth.

  Supporting this thinking is the notion that for most data networks using ATM
  as their wide-area medium, a router would likely be the access point with
  many TCP connections being concentrated on a given ATM connection.

3) Still others suggest that ATM switches should implement priorities and
   that there should be different buffer sizes allocated per priority.
   The different priorities and associated buffer sizes would support
   traffic separation, trading off cell loss for delay. So for example,
   "voice" traffic could have small buffer sizes and "data" traffic could
   have big buffer sizes.  The switches would then provide the buffering
   necessary to support TCP's reactive congestion control algorithms.

   Some folks argued that this would be "expensive" to implement.  Regardless,
   many new switches being anounced in 1993/4 claim to have such priorities
   and buffer size capabilities.

Finally many folks were not clear on the differing TCP reactive congestion
control mechanisms. A quick summary follows:

In the original algorithm,  the TCP goes into slow-start when a packet loss
is detected.  In slow-start, the window is set to one packet and increased
by one for every acknowledgement received until the window size is half what it
was before the packet is dropped. You get a series of larger and larger
bursts but the largest causes half the number of packets to be buffered as
there were before the packet drop occurred.  Hence there is no burst until the
window size is half what it was before the packet is dropped and is then
increased at a much lower rate, 1/(window size) for each acknowledgement.
This window control algorithm ensures that the only bursts generated are
probably small enough to be no problem.

In the Reno algorithm, the window is halved so that packets start being sent
in response to acknowledgements again after half the old window's of
acknowledgements have been received.  Hence there is no "burst" of packets
generated.  The only packess generated are in response to acknowledgements,
and only after half an old window of acknowledgements are received.

For more information check out Van Jacoboson's algorithms published in
ACM SIGCOMM 1988.


SUBJECT:  G2)   Can AAL5 be used for connection-less protocols?

         This thread started with questions about wether AAL5 supports
connection oriented or connection-less protocols.  Check the November
and December 1993 archives for the subject: "AAL Type 5 question".

First some background
---------------------
Officially, AAL 5 provides support for adaption of higher layer connection-
oriented protocols to the connection-oriented ATM protocol.
There was, however, a debate going on, claiming, that AAL 5 could also be
used to adapt higher layer connectionless protocols to the connection-oriented
ATM protocol.

The whole debate is grounded in a systematical approach of the ITU-T, which
states, that all AALs should be classified into four different classes, to
minimise the number of AALs required for supporting any imaginable service.

The classification of the ITU-T is as follows:

+------------------------+-----------+-----------+-----------+-----------+
|                        |  Class A  |  Class B  |  Class C  |  Class D  |
+------------------------+-----------+-----------+-----------------------+
|Timing relation between |        Required       |   Not Required        |
|source and destination  |                       |                       |
+------------------------+-----------+-----------+-----------------------+
| Bit Rate               |  Constant |          Variable                 |
+------------------------+-----------+-----------------------+-----------+
| Connection Mode        |     Connection-Oriented           |Connection-|
|                        |                                   | less      |
+------------------------+-----------------------------------+-----------+

AAL 5 is currently agreed to be in Class C. Some parties at the
standardisation bodies claim that it could be as well in Class D.

At the moment the following mapping between AALs and classes applies:
Class A: AAL 1
Class B: AAL 2
Class C: AAL 3/4, AAL 5
Class D: AAL 3/4

The reason for AAL3/4 in classes C and D is the follwing:
The ITU-T started to define AAL3 for Class C and AAL 4 for Class D. They
turned out to be identical after long debates.

Reality Check
-------------
The real issue is how to run a connection-less service over ATM which is
inherently connection-oriented.  AALs themselfs merely transport higher
layer packets across an ATM virtual circuit.  Connection-less services
are actually provided by higher layer protocols such as CLNAP.  Given
that, there is nothing to prevent folks from using AAL5 to implement
a connection-less communication mode.  This is exactly what the IETF is
doing with IP over ATM, and what the ATM Forum is also doing with
LAN Emulation.

The reality is that these folks expect that AAL5 will be largely used for
connection-less upper layer protocols such as CLNP and IP.  So some
find it strange to have AAL5 classified as an AAL for connection-
oriented services only.

However, from an ITU-T service Class perspective, you must stick strictly to
the view that to call an AAL "Class D" it must support each and every
posssible connection-less protocol.  The current agreement in the ITU-T
is that AAL5 can not claim this and so is officially considered a
"Class C" AAL.


-----------------------------------------------------------------------------
CONTRIBUTORS

This FAQ is a collective work which has been compiled by and is maintained
by Carl Symborski.  The information contained herein has been gathered from
a variety of sources.  Most derived from a consensus of postings on the
cell-relay group.  The following individuals have provided direct
contributions to this FAQ, either by posting to the cell-relay list or
by private EMail coorespondance.  If you would like to claim responsibility
for a particular item, please let me know.

Reto Beeler, [e-mail]
Kingston Duffie, [e-mail]
A. Gavras, [e-mail]
Rajeev Gupta, [e-mail]
Mark Jeffrey, [e-mail]
Gary C. Kessler, [e-mail]
Mark Laubach, [e-mail]
Matthew Maa, [e-mail]
Bert Manfredi, [e-mail]
George Marshall, [e-mail]
Keith McCloghrie, [e-mail]
Chris O'Neill, [e-mail]
Craig Partridge, [e-mail]
Vijay Rangarajan [e-mail]
Allen Robel, [e-mail]
Lee D. Rothstein, [e-mail]
Jukka Saukkonen, [e-mail]
Carl Symborski, [e-mail]
Chris Wilcox, [e-mail]
Mark Williams, [e-mail]
Mark, [e-mail]
------ END OF FAQ -------