Comparison of Fibre Channel and Gigabit Ethernet Networking
Technologies
 
Preface
 
The topic of this term paper is related to the two high speed
technologies, the Fibre Channel and the Gigabit Ethernet, that emerge
recently as a solution to solve the congestion of the overload network
backbone.   Most of the information is from the existing trade
publications, and (draft) specifications.   The topics related to these
technologies have been shown in many articles and magazines.  I would
like to take this opportunity to re-exam some of the similarities and
differences of these technologies, specially in their protocol layers
and topologies.
 
1.0  Background
 
Data traffic continues to increase across the entire spectrum of
computing network.   More and more business and institutions are unable
to deliver their data using the current data network bandwidth.  The
deficiencies in the current transmission rates result in the
communication bottleneck.  While computers have become faster and
capable of handling large amounts of data, the data transfer between
computers contributes traffic congestion on the backbone.  The high
speed and high data performance network architectures hence are needed
to solve the network problems, and be able to handling the future
network traffic.
 
Many high speed network architectures have been proposed and some have
been implemented successfully in the data storage and network
environments.   One of these approved technology is Fibre Channel.  It
is an American National Standards Institute (ANSI)  standard with one
Gigabits per second ( 1 Gbps) transmission rate.  Another emerging high
speed network technology, Gigabit Ethernet, is under debate in the IEEE
802.3z standards committee.  Although the Gigabit Ethernet specification
is yet to be finalized, many big network vendors already jump to the
bandwagon supporting this future cashcow.
 
The objectives of this paper are to review the communication layer
stacks and topologies of these two new breeds in the area of high
performance networks, and provide a comparison of their protocols and
their topologies.  

2.0 Key Issues
 
2.1 Protocol Stacks
 
2.1.1 Fibre Channel Communication Layers
 
The Fibre Channel consists of a set of protocols in which its base
protocol, Fibre Channel^Òs Physical and signaling Interface (FC-PH),
became an ANSI X3.230 standard in 1994.   The FC-PH includes the lowest
three protocol layers known as FC-0, FC-1 and FC-2, and is analogous to
the physical through transport layers in the OSI (Open System
Interconnection) model.  The other two upper layers provide common
services for higher-layer protocol and seamless integration of existing
standards.
 
Physical protocol, FC-0, describes the electrical and optical
infrastructure characteristics of the various physical media interface
for a variety of speed.  Each link is described using speed, media,
transfer, and distance.  The speed is specified in megabytes per second 
(1 megabyte (MB) = 8 x 220 bits).  The cabling alternatives are fairly
broad, with support for fiber optic, coaxial, and shielded twister-pair
cables.   The available data rates are 12.25 MBps, 25 MBps, 50 MBps, and
100 MBps that can support distance up to 10 km.   These effective data
rates are correspond to the line speeds from 132.8 Mbaud to 1062.5 Mbaud
that including 20% coding overhead.  ( I try to figure out  how the
1.0625 GHz link frequency  relates to the 100 MBps data rate, The
calculated number I come up with is not matched those number in most of
references.  If 1.0625 GHz link frequency is specified, then the maximum
data rate should be 95.37 MBps.  If 100 MBps data rate is specified,
then the link frequency should be 1.0486 Gbps.  Since the 1.0625 Gbps
link frequency and 100 MBps data rate are widely used in most of the
references, I remain using these numbers for any further discussion). 
 
Transmission protocol, FC-1, defines the synchronization and the
encoding/decoding schemes.  It is designed to improve the transmission
characteristics of the information sent.  Fibre Channel uses a
dc-balance 8B/10B code scheme from IBM ESCON transmission code, which
decodes each 8 bits of data into a 10-bit transmission character.  The
transmission characters ensure that enough transitions are presented in
the serial bit stream to make clock recovery possible.   This coding
increases the possibility of error detection and simplifies byte
synchronization.   Bits received on the serial channel are collected 10
bits at a time and decoded into 8-bit codes.  The 10-bit coding scheme
supports all 256 8-bit combinations and some control characters. 
Un-recognized codes at the receiver are treated as code violation
errors.  
 
Signaling protocol, FC-2, defines the rules for framing data to be
transferred between ports, the different mechanisms for using Fibre
Channel^Òs circuit and packet switched service classes, and the means of
managing the sequence of data transfer.
 
Framing function
 
Fibre Channel uses variable-sized frames up to 2,148 bytes including 36
bytes of overhead that provides Start of Frame (SOF), Frame Header,
CRC32, and End of Frame (EOF).  The Frame Header itself contains the
source and destination port addressing, service type, Sequence Count,
Sequence Identifier and Exchange Identifier.  A Sequence Count and
Sequence Identifier are used to manage the series of frames that are
fragmented from the single, higher-level protocol message.  Exchange
Identifier is used to identify one or more nonconcurrent sequences for a
single operation. 
  
Classes of service
 
FC-2 defines three classes of service.  Class 1 is a connection-oriented
service, where two nodes must establish a logical connection prior to
any data transfer.   This class ensures the frame orders and provides
full bandwidth for end-to-end nodes. Class 2 is connectionless service
with acknowledgments and flow control.  Different nodes using the class
2 service can share the channel bandwidth.   Class 3 is a datagram
service.  The FC standard also defines an optional service called
intermix.  The Intermix service combines class 1 and class 2,3 services.
Any left-over bandwidth from class 1 will be shared by the other
classes. 
  
 
Flow Control
 
Fibre Channel uses credit-based flow control scheme to control the flow
frames on the links and to avoid traffic congestion within the fabric. 
The fabric is a switching mechanism that is responsible to route the
packets from the source node to the destination node.   Different flow
control schemes are used for different service classes.  End-to-end flow
control is used for two communicating nodes.  The buffer-to-buffer flow
control is used for link between two hops (node to fabric, or fabric to
fabric).  
 
Common Service Protocol Layer, FC-3, defines the mechanisms for
striping, hunt groups and multicasts.  The striping is a technique used
to increase transmission bandwidth by employing multiple N_ports in
parallel; hunting group allows more than one port to respond to the same
alias address; and the multicasting is a single transmission sent to
multiple selected N_ports.
 
Upper-Level Protocol Mapping Layer, FC-4, provides interfaces to several
upper-layer protocols.  It allows Fibre Channel carry data from other
networking protocols and application.   The following networks and
channel protocols are either specified or proposed as FC-4s:  SCSI, IPI,
HIPPI, IP, AAL5, FC-LE, SBCCS, SNMP, and IEEE802.2 LLC specification.  
 
2.1.2 Draft Gigabit Ethernet Communication Layers
 
While Fibre Channel protocol stack contains layers that are analogous to
the OSI model from the physical layer to the transport layer.  Gigabit
Ethernet, likes its predecessors, only defines the two lower layers in
the OSI network model: Physical (PHY) and Data link layer that contains
the Media Access Control (MAC) sublayer and the Logical Link Control
(LLC) sublayer.
  
Physical Layer (PHY) 
 
In its rush to real product, Gigabit Ethernet work group proposes to use
the proven Fibre Channel physical signaling layers, FC-0 and FC-1, with
a minor changes as a Gigabit Ethernet ^Ñs Physical layer  for optic
fiber, 1000BASE-X.  The standard committee also evaluates the twisted
pair cable proposal, 1000BASE-T, and its performance to be used in
Gigabit Ethernet PHY layer.   The proposed Gigabit Ethernet^Òs
transmission code for optic fiber will be 8B/10B as in FC-1
specification.  This encoding ensures sufficient signal transition to
make clock recovery at the receiver end.  The standard committee still
has not determined what transmission code technique will be deployed for
the twisted pair cable in 1 Gbps data rate.  The limitation on distance
is one of the debating issue for the category 5 unshielded twisted-pair
(UTP) cable.  Unlike Fibre Channel, since the Gigabit Ethernet operates
in a full 1 Gbps data rate, the link frequency needs to be increased to
1.25 GHz to compensate the 20% data coding overhead, not 1.0625 GHz as
defined in the Fibre Channel FC-1 specification.  
 
Gigabit Media Independent Interface (GMII) 
 
The Gigabit Media Independent Interface (GMII)  specification provides a
simple, inexpensive and easy-to-implement interconnection between
various PHY layers to MAC sublayer.  It supports all three Ethernet
generations: 10 Mbps, 100 Mbps and 1 Gbps data rates.
 
Reconciliation Sublayer (RS)
 
The Reconciliation Sublayer provides a mapping between the signals at
GMII and MAC^Òs physical layer signaling.
 
Media Access Control (MAC) Sublayer 

The proposed Gigabit Ethernet specification supports both CSMA/CD
half-duplex and full-duplex operating modes.  The CSMA/CD uses a
back-off algorithm to prevent more than one device from sending
information at a time.  In order to retain the compatibility with its
predecessor, a Gigabit Ethernet maintains the same frame structure and
the minimum and maximum frame sizes.   Several new features are also
proposed to use the CSMA/CD to enable efficient operation over a
collision domain at 1 Gbps speed in half-duplex mode operation.  These
features are carrier extension and packet bursting.  The proposed
standard has changed some MAC parameters from the 10/100 Mbps
predecessors.  The transmission slot time is raised from 512 bits times
to 4096 bit times to ensure proper operation of the CSMA/CD protocol,
and the inter-frame gap is shorten to 0.096 us.  Other MAC parameters
remain the same as the 10/100 Mbps MAC.

In Half-Duplex operating mode at 1 Gbps speed, the minimum frame size is
not worked properly with the CSMA/CD protocol.  The IEEE 802.3z standard
committee proposes appending a sequence of extension bits to those
frames less than slot-time bits in length, The proposed standard allows
continuously transmitting series of frames up to a specified limit in
case of frame bursting.  In this case, only the first minimum frame
needed to be extended with non-data bits, and all consecutive frames are
transmitted without the carrier extension.  The maximum continuous
transmission from a single station is limited to no more than the time
required to transmit two maximum-size frames.
 
In full-duplex operating mode, there is no need to detect bus contention
with other traffic on the medium.  The station can transmit frames after
waiting at least 0.096us frame gap delay regardless of the presence of
receive activity. 
 
A link-level flow control has been debated within the IEEE 802.3z Work
group and several approaches have been suggested.  The proposal of
credit-base flow control, asymmetric flow control and  the ^ÓX on/ X off^Ô
flow control were discussed.   The ^Ó Xon/Xoff  flow control allows
switch buffers to drain before additional traffic can be sent to the
port.  

Logic Link Control (LLC) Sublayer 
 
The Gigabit Ethernet standard also proposes to deploy the IEEE 802.2
Logical Link Control sublayer on top of MAC sublayer to provide the
services needed for the network layer to exchange packets with remote
network layer entities.
 
 
2.2 Topologies

While Gigabit Ethernet proposes to retain the traditional Ethernet
topologies in network arena, the Fibre Channel defines various
topologies to be used in both the channel and network environments.   

2.2.1  Fibre Channel Various Topologies
 
Fibre Channel is a closed system that requires each N_port identifies
itself by login function during initialization.  Once identified, the
N_ports only handle communications with the fabric and don^Òt care how
the frame getting route to the destination.  The routing function is a
responsibility of Fibre Channel^Òs fabric.   Various fabric topologies
include point-to-point, Crosspoint-switched, and arbitrated loop.

Point-to-point
This is a simplest topology to use bi-directional point-to-point links
to interconnect the N_ports of a pair of nodes.  each N_port can
transmit information to another N_port at all time without being
blocked.  The advantage of this nonblocking approach is to provide
instantaneous access to the destination ports.  However, this
nonblocking approach usually underutilized the communication link^Òs
bandwidth.
  
Crosspoint-switched
This topology based on a switched fabric provides a choice of multiple
routing paths between pair of fabric ports (F_port) to be configured as
a nonblocking communication.  This frame-switching approach allows data
transmission bandwidth to be dynamically allocated on the link-by-link
basis.   Depending on the routing within the fabric, individual frames
between the same pair of N_port may take alternative paths; however, the
fabric may be implemented to guarantee in-order delivery of frames to
the destination N_port.  The advantage of this topology is the efficient
sharing of the available bandwidth among all N_ports during bursting
across networks.  However this switching approach is complex and
expensive to implement.  It also contributes to data contention.

Arbitrated loop
 
The Arbitrated Loop topology is the newest addition to the Fibre Channel
topology group that attaching multiple ports in a loop without hubs and
switches.  This topology is defined in the FC-AL standard (X3.272).  The
Arbitrated loop interconnects loop ports (L_port) at the nodes or the
fabric via unidirectional links and allows any port to arbitrate for use
on a loop.   Once an L_port wins the arbitration, a second L_port may be
opened to complete a single bi-directional point-to-point circuit.  The
loop is self-configuring and may operate with or without a fabric
present.  A fairness algorithm is used to provide equal access of all
ports in the loop.  This loop topology provides a low cost solution for
sharing communication channel bandwidth among all L_ports, but it also
carries the possibility of blocking since only one pair of L_ports is
allowed to communicate at one time.  In addition, communication among
the L_ports could be terminated if any link in a loop goes down.  
 
2.2.2 Gigabit Ethernet topologies

Bus Topology

Because the draft is not finalized, The myth of what topologies will be
deployed continues to be a debating topics.   The original Ethernet
specification defines a shared media Bus topology that allows multiple
nodes to share a same wire.  The shared-media bus topology may also be
deployed in Gigabit Ethernet along with multi-port repeaters.  The new
switched and routed topologies could be applied for the full-duplex
operating mode for switch-to-switch and switch-to-end station
connections to solve the current networking bottleneck problems.  The
shared media bus schemes may not be a good approach as a number of nodes
on the network rise, the node bandwidth will be reduced accordingly.  

Repeater

The repeater set extends the Ethernet physical system topology by
repeating signal from one segment to another^Òs.  It receives and decodes
the data from any segment,  and re-transmits the data to all other
segments attached to it.  Only one repeater set is allowed for a single
collision domain in 1 Gbps baseband network.   The repeater set is not
considered to be the Ethernet station, so it does not apply to the count
of the overall limit of 1024 stations on a network. Multiple collision
domains can be joined by multi-port bridges and/or routers to form 
network topology which can be deployed either homogenous 1 Gbps networks
or heterogeneous 10/100/1000 Mbps mixed CSMA/CD networks.  

3.0  Analysis

One of the bright area of the Fibre Channel is the recognition of part
of its FC-PH (Fibre Channel ^Ñs Physical and signaling layers) protocols
that have been and will be dominantly used as a low level protocols for
most of the high speed networks. 
 
When using the technology for a clustering system, Fibre Channel is
benefit from the complete layers specification of FC-PH which is
analogous to the physical through transport layers of the OSI model.  
Many high speed applications could be designed to run directly on top of
FC-PH with a minimum latency.  

Essentially, Fibre Channel is not only an alternative network
architecture that shared the high speed race with Gigabit Ethernet and
other such as ATM (Asynchronous Transfer Mode), Fast Ethernet, 100VG
AnyLAN, but also competes with Ultra-SCSI ( Small Computer System
Interface), IBM SSA (Serial System Architecture) on the high-speed
hardware for mass storage devices.  It could make a substantial advance
in enterprise storage and high speed disk subsystems for the future.
 
Fibre Channel ^Ñs strength is its various topologies that help to build
many different high performance switches for clustering of high
performance workstations.  An n-port fabric could allow up to n/2
distinct pairs of nodes to communicate at the full bandwidth
simultaneously.  This bandwidth could also be double if the
interconnected nodes utilize the full-duplex capability of Fibre Channel
links.  
 
Fibre Channel Arbitrated Loop (FC-AL) is one of the cost-effective Fibre
Channel rollouts.  FC-AL makes it possible to achieve increased
performance, large capacities and campus-wide support while adding to
the robust security, high availability and scalability options in
today^Òs advanced disk array subsystem.
 
Fibre Channel ^Ñs most threatening drawback, perhaps, is its inability to
function as an effective WAN technology and its popularity in the
networking arena. Most administrators are not as familiar with the Fibre
Channel technology as they are with the more popularized technologies
such as ATM and Ethernet.  With the wide spread use of TCP/IP protocols
in current LAN and WAN networks,  Fibre Channel has a hard time
convincing vendors to implement the Fibre Channel network with TCP/IP. 
Even if Fiber Channel ^Ñs FC-4 mapping layer easily intermingles with the
TCP/IP protocols, the high software development cost and lack of support
from the big networking vendors may cost Fibre Channel a chance to
compete with Gigabit Ethernet or ATM in the popular LAN and  WAN
markets. 
 
Meanwhile, Gigabit Ethernet is benefit from the software reuse with the
vast compatible installed bases of its low speed predecessors in the
local area networks. Although Gigabit Ethernet standard is not yet
released, it already becomes a well-known technology in the networking
arena. By using the same high-level protocols, networking engineers can
avoid protocol conversion to run high speed Ethernet traffic.  
 
With a support for full-duplex operation, Gigabit Ethernet will be an
ideal backbone interconnect technology of the 10/100Base-T switches, as
a connection to high performance servers and as an upgrade path for
future high-end desktop computers requiring more bandwidth than
100Base-T can offer.  Gigabit Ethernet would reduce traffic congestion
on 100 Mbps Ethernet networks, and would be able to compete with ATM in
real-time and multimedia applications with its gigabit data rate. 
However, unlike ATM, Gigabit Ethernet is not as scaleable for multimedia
application and doesn^Òt has the same quality of service as ATM. 
Furthermore, there still some problems needs to be solved by the IEEE
802.3z Work Group before the standard is finalized.  One of these
problems is the proposal of using same minimum frame size with CSMA/CD
for half-duplex gigabit Ethernet.    The Ethernet CSMA/CD can^Òt keep up
with data bursting across a network at a gigabit data rate, and the
cable length must be significant short (less than 25 m) if supporting
the same minimum frame size required by the standard Ethernet.  With
proposed carrier extension, the network bandwidth will be wasted if too
many small frame  size packages are transmitted. 
 
4.0  Conclusion

A Gigabit Ethernet, if it is successfully implemented, could serve well
as a backbone for Ethernet based network, because of its vast Ethernet
installed bases and its directly inter-operability with the TCP/IP.  It
would also provide enough bandwidth to be used for data intensive
applications.   Gigabit Ethernet seems to be a logical evolution from
its early 10 /100 Mbps Ethernet.  Specially, its gigabit speed
full-duplex link is probably the easiest and most desirable way to solve
the future network needs.  
 
On the other hand, with the gigabit speed, Fibre Channel could be the
driving force to revolutionize the data access technology in the area of
clustering and data intensive applications.  It would be well fit in
with more specialized applications like high speed mass storage, and
clustering workstations.  Accessing data from the storage disks that
connect directly to the network through the Fibre Channel will be much
faster than most of the file servers can deliver the data through the
current networks.  The Fibre Channel has been deployed primarily by
storage device vendors as a migration path for SCSI.
 
Only future can tell if either Gigabit Ethernet or Fibre Channel will be
a winner in a high speed race.   Fibre Channel offers more raw speed and
guarantees synchronous delivery except lack of high-level error
detection during transfer.  Its open architecture and high performance
throughput should make Fibre Channel the connection for high-end systems
such as video/graphics and network servers. Likewise, Gigabit Ethernet
has technical feasibility, economic feasibility, and a very broad market
potential because the basic for the technology already exists; however,
it has not yet been standardized.
 
No matter which technology wins, the prospects of networks operating at
gigabit speed hold important implications for network architectures and
the applications they support.  High-speed race will never end, when the
gigabit speed would become heavily use in the desktops and local LANs,
the bottleneck and traffic congestion will once again becoming a problem
in the network backbones.  New ultra-high-speed technologies will then
be invented.  Does it sound familiar?  Stay tuned for the next ^ÓTerabit
speed^Ô network in the future^Å
 
5.0  References
 
 ^ÓNew-age networking^Ô Computer Shopper July 1996 v16
 ^ÓFinding answer with fibre^Ô by O^ÒMara, Bruce.  Computing Canada Nov 7,
1996
 ^ÓFinally, fibre!^Ô LAN Magazine August 1996 V11
 ^ÓFibre Channel fits in: technology stirs vendors^Ò interest^Ô  Computer
Reseller News August 26, 1996.
 ^ÓFibre channel Fusion: Low Latency, High Speed^Ô  by E. Frymoyer,
Hewlett-Packard Co, Feb, 95
 ^ÓFibre channel a technical Description^Ô  Fibre Xpress network, Systran
Corp.
 ^ÓChanging channels by Goralski and Kesslet, LAN Magazine may 1996 v11
 ^ÓFibre Channel and Related standards^Ô by Sachs, IEEE Communications
magazine,  Aug 1996
 ^ÓGigabit Ethernet may ease congestion^Ô InfoWorld Jan 8, 1996 v18
 ^ÓGigabit Ethernet^Ô Gigabit Ethernet Alliance, August 1996
 ^ÓGigabit networking topologies converge^Ô by S. Haim, Electronic
Engineering Times Sept 12, 1996
 ^ÓGigabit Ethernet: Technology, System, and Network Applications by
Eijk, Electronic Design, Apr 1,1997.
 ^ÓIEEE Draft P802.3z/D2^Ô by Lan MAN Standards committee of the IEEE
Computer Society. February 19, 1997