Fibre Channel, SCSI, and You
Lab Reports by Dean Porter
This article appeared first in Windows NT Magazine in September 1997
Reprinted with permission from Windows NT Magazine.
A preview of Fibre Channel performance
Today's environment demands very fast transfer of large volumes of information. No wonder the Fibre
Channel (FC) and Fibre Channel Arbitrated Loop (FC-AL) storage interfaces have generated so much
interest. If you compare SCSI's current high-end 40MBps data transfer rate with FC's high-end 106MBps,
you might think you can realize a 265 percent improvement in performance. But is FC that much faster?
The Windows NT Magazine Lab compared the two interfaces.
ANSI developed the FC Standard in 1988 as a practical, inexpensive, and expandable method of using
fiber optic cabling to transfer data among desktop computers, workstations, mainframes,
supercomputers, storage devices, and display devices. ANSI changed the standard to support copper
cabling; today, some kinds of FC use two-pair copper wire to connect the outer four pins of a nine-pin
type connector, as you see in the photo. So despite the name, most current implementations of FC don't
use fiber optic cabling.
Copper wire connections work for up to 30 meters; beyond that distance, you must use fiber optic
cabling and connectors. With fiber optic cables, you plug in an optic converter on both ends of the FC
connection, and you can reach up to 10km. This distance is clearly superior to the current limitation of
25 meters for the differential SCSI technology. (For more information about SCSI, see Sean Daily, "SCSI
and IDE: Defining the Differences," June 1997.)
FC's design is similar to the Open Systems Interconnect (OSI) network layers. FC supports several data
communication protocols, including Fiber Distributed Data Interface (FDDI), High-Performance Parallel
Interface (HiPPI), Intelligent Peripheral Interface (IPI)-3, Internet Protocol (IP), SCSI-3, Ethernet, Token
Ring, and asynchronous transfer mode (ATM).
You can configure FC for data rates of 13.3MBps, 26.6MBps, 53.2MBps, and 106.4MBps; and it can
achieve these transfer rates in both directions simultaneously (duplex). Thus, FC can transfer data at
more than 200MBps, if usage is balanced in both directions quite an improvement over the simplex and
half-duplex interfaces such as SCSI. Work is under way for FC specifications of 400MBps (again, in both
directions simultaneously; this technology could attain a data rate of more than 800MBps). Also note
that an Ultra2 SCSI version is under development that will offer speeds up to 80MBps.
TABLE1:
Performance of FC and SCSI: SQL Benchmarks
Order entry: Single read/write (transactions per second)
Step 1 Step 2 Step 3 Step 4
Fiber 14 16 18 19.99
SCSI 13.97 15.97 17.86 19.77
Order entry: Mixed read/write (transactions per second)
Step 1 Step 2 Step 3 Step 4
Fibre 14 16 17.99 19.94
SCSI 13.99 16 18 20
TABLE2:
Performance of FC and SCSI: File Transfer Benchmarks
Simulated Fibre Channel SCSI
Step Users KBps KBps
1 20 1410 1152
2 40 2516 2278
3 60 2964 2866
4 80 3076 3022
5 100 3252 3266
FC Configurations
You can configure FC ports three ways: in an FC-AL, in point-to-point links, or in a switch. The most
common configuration (and the one you can buy now for Windows NT) is FC-AL, which ANSI developed
to connect peripherals. In FC-AL, you usually connect the output of one FC device to the input of another
FC device, and you connect the last device back to the first device, creating a loop. An FC-AL natively
uses the SCSI-3 (SCSI FCP) protocols and can address 127 FC devices or nodes within the limitations of
the 30-meter copper cable or the 10km optic cable. (A SCSI connection can handle only seven devices,
excluding the computer.)
FC-AL is a simple closed serial loop. An FC-AL device has two connectors (one in and one out) an
arrangement that makes connecting devices a breeze. You run a cable from the card in the computer to
the first FC device and then connect each FC device in the chain to the next device. On the last device in
the chain, you plug a loop-back connector into the out connector, which runs the data from the
connector's send side to its receive side to complete the loop. Each connection uses four wires: two for
transmitting and two for receiving. Using an electrical differential technique, FC (like differential SCSI)
uses a balanced negative and positive wire pair to improve data integrity and to let you spread the
network over greater distances.
FC and NT
Several vendors support FC in the NT market. Adaptec and Emulex have FC cards with drivers for NT,
and Raidtec has a complete FC hard drive setup: Raidtec hard drive enclosure, an Emulex LightPulse PCI
FC Host Adapter card, and Seagate FC-AL hard drives. Some manufacturers are producing hard drives for
use in FC arrays. Unfortunately, no hardware-based FC RAID solutions are currently available. If you
want RAID, you must use the slower, software-based RAID that is included with NT Server.
The Lab compared Raidtec's FC offering with its SCSI-3 RAID system, using an Amdahl 200MHz quad-
processor Pentium Pro system with 512MB of RAM. Because hardware-based RAID isn't available for FC,
we tested both systems using a four-drive stripe set, which writes the data evenly across all the drives
without fault tolerance. To get a general idea of the difference in transfer speed, we timed copying NT's
Service Pack 3 from one directory to another directory on the same disk. The FC system was about 5
seconds faster.
For a more sophisticated test, we used BlueCurve's Dynameasure benchmarking tool to test the
networks' performance. (For more information about this tool and the benchmarks we used, see Carlos
Bernal, "Dynameasure Enterprise 1.5 by BlueCurve," page 81.) Because we were working with alpha and
beta FC products, we kept our tests simple and used the SQL and File transfer benchmarks to look only
at whether FC offers a performance advantage.
For the SQL benchmark, we used the Order Entry/Single Read/Write and Order Entry/Mixed Read/Write
tests with a 96MB dataset. For the File benchmark, we used the Copy All Bi-directional benchmark, which
copies a random mix of files; the file dataset size was 35.6MB. We used 20 client systems in a five-step
test that simulated 20 to 100 users.
FC and SCSI-3 performed about the same in the SQL test, as you see in Table 1. In the file transfer test,
we found FC to be 10 percent to 20 percent faster than SCSI-3, as you see in Table 2. This result is
much different from the 265 percent that you might expect. Of course, performance is subject to change
with the final releases of the hardware components and software drivers.
So, why wasn't FC faster? Perhaps the hard drives couldn't spit out data fast enough, or the FC card in
the server couldn't pump data into memory fast enough, or the CPUs couldn't take the data out of
memory and dump it back to the FC card fast enough, or a combination of these factors. This situation is
analogous to the difference between a 10Mbps and a 100Mbps Ethernet: The performance doesn't
increase 10 times when you move from 10Mbps to a 100Mbps network, only about 2.5 times. The speed
of the wire (the transfer rate) is only one of many factors that come into play when you interconnect
equipment.
Today, FC can connect more devices and give you an extra bit of performance over SCSI-3, although you
can approach FC's speed and connectivity by using multiple SCSI controllers. FC costs more now because
it is new technology, but its costs are already coming down. If you need the fault tolerance of RAID, you
can buy hardware-based SCSI RAID controllers, which are much faster than software-based RAID, thus
making a multichannel SCSI solution a better solution than FC for many applications. But tomorrow will
bring hardware-based FC RAID solutions and FC-to-SCSI converters. Then FC will be more attractive.
ABOUT THE AUTHOR
Dean Porter is a lab technician and product reviewer for Windows NT Magazine. You can reach him at
[e-mail].
Additional Information
You can find more information about FC at the following locations:
Fibre Channel Association (FCA): http://www.amdahl.com/ext/CARP/FCA
Fibre Channel Systems Initiative (FCSI): http://www.amdahl.com/ext/CARP/FCA/FCSI.html
Fibre Channel Consortium: http://www.iol.unh.edu/consortiums/fc
Fibre Channel Loop Community: http://www.symbios.com/fclc
Ancor Communications: http://www.ancor.com/fcinfo.htm
American National Standards Institute (ANSI): http://www.ansi.org/docs/home.html
You can find more information about SCSI technologies at these locations:
ANSI X3T10 Subcommittee: http://www.symbios.com/x3t10 3t10
ANCOT: http://www.ancot.com
Contact Info
Adaptec " 408-945-8600 or 800-442-7274
Web: http://www.adaptec.com
Email: [e-mail]
Amdahl " 408-945-8600 or 800-442-7274
Web: http://www.amdahl.com
Emulex " 714-662-5600 or 800-854-7112
Web: http://www.emulex.com
Email: [e-mail]
Raidtec " 770-664-6066
Web: http://www.raidtec.com
Email: [e-mail]
Seagate Technology " 408-438-8111
Web: http://www.seagate.com
About Windows NT Magazine
Windows NT Magazine is the leading publication for corporate IS teams deploying Windows NT and
related applications and technologies. Each month you'll find timely, how-to articles, tips and techniques
to help you get the most out of your NT system. Keep up with the rapid pace of Windows NT technology
changes and better equip yourself to make those all-important technology decisions for your company.
Subscribe today!
Over 100,000 paid subscribers receive Windows NT Magazine every month in over 110 countries. Don't
delay, for subscription information/to order, point your browser to:
http://www.winntmag.com/subscribe.html
Call 970-203-2782 or 800-793-5712, or email [e-mail].
We at Microsoft Corporation hope that the information in this work is valuable to you. Your use of the
information contained in this work, however, is at your sole risk. All information in this work is provided
"as -is", without any warranty, whether express or implied, of its accuracy, completeness, fitness for a
particular purpose, title or non-infringement, and none of the third-party products or information
mentioned in the work are authored, recommended, supported or guaranteed by Microsoft Corporation.
Microsoft Corporation shall not be liable for any damages you may sustain by using this information,
whether direct, indirect, special, incidental or consequential, even if it has been advised of the possibility
of such damages.
TechNet Home Feedback
© 1999 Microsoft Corporation. All rights reserved. Terms of Use.