Fiber Optic Essentials

138 Fiber Optic Essentials

Government supported networks like these are often used as examples in

the technical literature to illustrate network traffic management models.

6.3 Latency

Dense wavelength division multiplexing devices also function as channel

extenders, allowing many datacom protocols to reach previously impos-

sible distances (50–100 km or more). Combined with optical amplifier

1

E

W

2

E

W

3

E

W

ATM IP TDM

Terminal site

1

E

W

2

E

W

3

E

W

ATM IP TDM

Terminal site

(a)

3

E

W

2

E

W

ATMIP

ATMIP

1

E

W

ATM IP

OADM site OADM site

OADM site

1

E

W

2

E

W

3

E

W

ATM IP TDM

Terminal Site

(b)

Figure 6.3 WDM network topologies (a) point-to-point (b) hubbed ring.

6. Wavelength Multiplexing 139

1

E

W

IP

OADM

2

E

W

ATM

OADM

1

E

W

2

E

W

ATM IP

Terminal

1

E

W

2

E

W

ATM

IP

Terminal

(c)

1

E

W

2

E

W

OADM or

Terminal site

2

E

W

OADM or

Terminal site D

1

E

W

OADM or

Terminal site A

1

E

W

OADM or

Terminal site C

2

E

W

OADM or

Terminal site B

(d)

1

E

W

2

E

W

ATM IP

OADM site

Terminal site

2

E

W

IP

Terminal site

1

E

W

ATM

(e)

Figure 6.3 (continued) (c) dual hubbed ring (d) meshed ring (e) linear optical

add/drop multiplexer (OADM).

Abilene International Network Peers

Pacific Wave

AARNET,

CA*net, GEMNet,

KOREN/KREONET2,

SingAREN, TANet2

New York

SINET,

SURFnet

/IEEAF-TYC

StarLight

ASNet, CA*NET, CERN/DataTAG, GÉANT*/EuroLink,

HARNET, JGN2 (APAN), KOREN/KREONET2, GLORIAD**,

SURFnet, TANet2, CERNET

MANLAN

CA*Net, GÉANT*,

HEANET, SINET,

SURFnet

/IEEAF-TYC

Qatar FN

Los Angeles

APAN/TransPAC

UNINET

Pacific Wave

Seattle

Sunnyvale

Houston

San Diego

(via CALREN2)

CUDI

EI Paso

(via UTEP, UT)

CUDI

Miami (via AMPATH)

ANSP, RETINA,

RNP2, REACCIUN-2

Denver

Chicago

Kansas City

Atlanta

Washington

GÉANT*

Indianapolis

*vis GEANT: ACOnet, ARNES, BELNET, CARNet, CERN, CESNET, CYNET, EENet, Forskningsneltet, Funet,

G-WIN, GARR, GRNET, HEAnet, HUNGARNET, IUCC, JANET, LANET, LITNET, Univ, Malta, POL34, RBnet,

RCTS2, RedIRIS, Renater, RESTENA, REUNA2, Rhnet, RoEduNet, SANET, SUNET, SURFnet, SWITCH,

ULAKBYM, UNINETT

*via APAN/TransPAC: WIDE/JGN, IMna1, CERNet

/CSTnet /NSFCNET, KOREN/KREONET2, PREGINET,

SingAREN, TANET2, ThaiSARN, WIDE (v6)

**vis GLORIAD: CSTNET, RBnet

Figure 6.4 Peer nodes for the initial implementation of the next generation optical Internet.

6. Wavelength Multiplexing 141

technology, this has led some industry analysts to proclaim “the death

of distance,” that is connection distances should no longer pose a seri-

ous limitation in optical network design. However, in many real world

applications, it is not sufficient to simply extend a physical connection;

performance of the attached datacom equipment must also be consid-

ered. Latency, or propagation delay due to extended distances, remains

a formidable problem for optical data communication. The effects of

latency are often protocol specific or device specific. For example, using

DWDM technology it is possible to extend an ESCON channel to well

over 50 km. However, many ESCON control units and DASD are syn-

chronous, and exhibit timing problems at distances beyond about 43 km.

Some types of asynchronous DASD overcome this limitation; however,

performance of the ESCON protocol also degrades with distance. Due

to factors such as the buffer size on an ESCON channel interface card

and the relatively large number of acknowledgments (ACKs) or hand-

shakes required to complete a data block transfer (up to six or more),

ESCON begins to exhibit performance droop at around 9 km on a typical

channel, which grows progressively worse at longer distances. For exam-

ple, at 23 km a typical ESCON channel has degraded from a maximum

throughput of 17.5 MB/s to about 10 MB/s; if the application is a SAN

trying to back up a petabyte database, the time required to complete a full

backup operation increases significantly. This problem can be addressed

to some degree by using more channels (possibly driving the need for a

multiplexer to avoid fiber exhaust) or larger data block sizes, and is also

somewhat application dependent. Other protocols, such as FICON, can

be designed to perform much better at extended distances.

A further consideration is the performance of the attached computer

equipment. For example, consider the effect of a 10 km duplex channel

with 100 ms round trip latency. A fast PC running at 500 MHz clock

rate would expend 50,000 clock cycles waiting for the attached device

to respond, while a mainframe executing 1000 MIPS could expend over

100,000 instruction cycles in the same amount of time. The effect this

may have on the end user depends on factors such as the application

software. As another example, there is tremendous activity in burst mode

routing and control traffic for next generation Internet (NGI) applications.

In many cases, it is desirable for data to be transported from one point

in the network to one or more other points in the least possible time. For

some applications the time sensitivity is so important that minimum delay

is the overriding factor for all protocol and equipment design decisions.

A number of schemes have been proposed to meet this requirement,