Focus - Fibre Channel Vs. Ethernet--SUMMARY:----Computer Storage
( a bit different than the normal Mkt Lab Board-of-
Fair but these SAN's and Network Attached Storage
companies like BRCD, NTAP, EMC are just too darn
interesting...... -g-)
March 13, 2000
*Recently, there has been quite a stir as to what will be THE networking
technology of the future. We believe Fibre Channel's disruptive technology
is the catalyst driving the debate and, therein lies our focus.
*In this report, we set out to separate fact from fiction and establish a
timeline.
*We believe Brocade (BRCD, $333, 2H) is one of the clear leaders in the
storage networking industry today and will continue to be well positioned
into the future, regardless of the outcome of this debate.
*The "safer bets" for investors are companies without any networking
preference; particularly: EMC (EMC, $130.06, 1M), VERITAS (VRTS, $168.69,
2H), and Network Appliance (NTAP, $236.88, 2H).
--OPINION:------------------------------------------------------------------
Meet George Jetson
The driving focus on the "Fibre Channel vs. Ethernet" debate stems from
the belief that the world should have one heterogeneous network that can
interconnect all types of voice and data traffic. In fact, this is the
goal of many networking companies' current initiatives and has been for
many years (note the competition between Cisco and Lucent over who has
the ultimate networking technology). However, we believe that making it
a reality may be as far out as the reality of "The Jetsons" cartoon.
Therefore, we believe investors should be focussed on when and how it
might happen in order to put things into perspective.
When
We have heard many people implying that this concept of "one network"
is just around the corner and unfolding today. While we agree it is
being architected today, we believe the availability of a technology that
can meet all types of networking demand is way out into the future (more
than 5 or 10 years).
It has been our experience that technology shifts of this magnitude take
more time - not less - than originally anticipated. Right now, people
are talking about Gigabit Ethernet moving to 10 Gbps and 100 Gbps; we
note that Ethernet is currently just beginning to ramp transmission
speeds of 1 Gbps. Also, Fibre Channel is not standing still. 2 Gbps
Fibre Channel should begin rolling out later this year and 10 Gbps Fibre
Channel spec is already in the works.
In the end, we believe that it will take at least two or three years
before Ethernet is ready to go head-to-head with Fibre Channel.
Thereafter, we believe the two technologies will overlap.
How
The how is a bit tougher to explain. We will break this explanation into
1) understanding the difference between various networking medias and
protocols, 2) what we believe will unfold in the next three years, and 3)
our longer-term vision. We believe it is important to recognize that
only points 1 and 2 are where investors should be focussed. We believe
point 3, while fun to talk about, is too futuristic and, therefore,
should not consume investors' focus today.
We Note
Who would have guessed that Cisco's technology, a company that very few
people had heard of in the late eighties and early nineties, could have
usurped Big Blue's efforts to network the Internet with their
technology? Point: nobody really knows what technologies will survive in
the distant future (we believe a more appropriate window to focus on is
two or three years out).
Our Points Of View
*What about SCSI? SCSI will last for a long, long time (five plus years).
*By the end of this year SAN vs. NAS arguments will cease to exist in
favor of discussions of one uniform "Data Centric" architecture
(architecture, not network) which will incorporate multiple networking
platforms, including Fibre Channel, Ethernet and ATM.
*Ethernet TCP/IP networking companies will both compete and partner with
Fibre Channel networking companies throughout 2000 and well into the
future.
*Ethernet TCP/IP will develop into a more robust technology over the next
three years, and only then, will it be able to truly compete with Fibre
Channel to network storage.
*By the time Ethernet TCP/IP is developed to function similarly to Fibre
Channel, it will no longer have a cost advantage. In other words, you
get whet you pay for; there are no free lunches.
*Fibre Channel will continue to be the only viable storage networking
solution for enterprise class data for at least another two or three
years.
*Fibre Channel will continue to gain momentum through the end of 2001, no
matter how many resources are stacked against it. Note: it takes a long
time between a technology's inception and when its ready for prime time.
*Storage Networking companies will continue to recognize record growth in
the next two years.
*Storage Networking companies will continue to migrate to incorporate
Ethernet, ATM and any other networking technology which presents itself
as a viable storage networking alternative for the future.
*Brocade is one of the clear leaders in the storage networking industry
today and will continue to be well positioned into the future, regardless
of the outcome of this debate.
*The "safer bets" for investors are companies without any networking
preference; particularly: EMC, VERITAS, and Network Appliance.
Who Wins?
We certainly see many new technological advancements developing from both
sides of the debate. For example, we believe Cisco's recent torrid pace
of acquisitions is a clear indication it recognizes that networking is
constantly changing. We also believe almost every one of the Fibre
Channel companies has morphed quite a bit since inception. For example,
Ancor who was focussed on Fibre Channel LANs in the early nineties, has
moved into Fibre Channel storage networking, and is now helping define
Infiniband with Intel. Note: we used Ancor as an example because of its
long history and dramatic migration, we could have used any other Fibre
Channel company.
We believe the ultimate winners will be companies from both Ethernet and
Fibre Channel networking ancestries who have displayed 1) technology
leadership, 2) visionary management leadership, 3) the ability to react
quickly to changing technologies and changing customer demands, and 4) a
willingness to migrate their core competencies to suit customer needs two
to three years out by developing fresh technologies in new directions.
We believe when the dust settles in 10 years, we will have networking
companies with Ethernet roots and ones with Fibre Channel roots. In our
view, the key for investors is not to pick a technology, rather quality
companies which have displayed the ability and willingness to adapt.
Why Does Storage Need Fibre Channel Today?
Small Computer Systems Interface (SCSI) has been the storage connectivity
solution for decades. SCSI differs from IP connectivity solutions in
that it enables "block level" data transmissions from servers to
storage (we discuss why this is important in the section titled: Ethernet
Is Cumbersome). With data doubling every year, SCSI has begun to fall
short of meeting computing needs. Therefore, Fibre Channel has emerged
to: 1) assist in circumventing the I/O bottleneck, 2) move data more
quickly, and 3) enabler server and storage connectivity over longer
distances.
Fibre Channel Helps Resolve I/O Constraints
The emergence of Fibre Channel is a result of the ever increasing demand
for more data, more quickly, i.e the demand for higher I/O bandwidth (I/O
refers to Input/Output and represents the rate at which a computer can
receive and send information). Also called "bandwidth" solutions,
connectivity solutions refer to those technologies which aid in the
transfer of data between its storage site and the client or server
computer. The connectivity solution can significantly enhance total
system performance by increasing data transfer rates between PCs,
servers, peripherals and networks. Connectivity is the element that
links the user to the data.
Note: when connectivity solutions are switched or routed we refer to them
as networking solutions. Traditional SCSI can connect a maximum of 16
nodes while Fibre Channel can connect 126. Fibre Channel can be switched
which has enabled companies such as Brocade to interconnect over 3000
nodes.
Connectivity, especially in the high-end of the storage systems market,
is undergoing a rapid transition to higher bandwidth technologies. A
number of trends are driving the need to increase effective I/O bandwidth
between clients, servers, peripherals and networks. These factors have
created a rapid increase in the transfer of data between the desktop,
servers, peripherals and networks, resulting in substantial I/O and
network bottlenecks. Some of these are as follows:
X The introduction of increasingly powerful processors require more
rapidly accessed and intelligently managed data.
X Advanced operating systems, especially advances in open systems such as
UNIX and Windows NT, allow for faster I/O and multitasking.
X The growth of data-intensive software applications, such as graphics
and video, require increases in bandwidth.
X The proliferation of client/server networks, the Internet, email and
corporate intranets all drive growth in the number of servers clients,
all communicating over the same network.
X The growth in high performance peripherals, such as high capacity hard
disk drives, scanners, CD-ROMs, digital photography and voice recognition
technologies adds to the flow of information over the network.
The existing storage connectivity standard, SCSI, has begun to fall short
of the rapidly advancing system I/O demands. Due to the inherent
limitations of the SCSI standard, a new storage connectivity standard,
Fibre Channel, has emerging to take its place. This new standard has
gained strong momentum in the past year as the heir apparent to SCSI and
the enabler of Storage Area Networks (SANs); thereby, bringing the
additional benefits of networking to the storage world.
Note: Fibre Channel is based on SCSI protocol, think of it as supped up
SCSI.
Fibre Channel Moves Data More Quickly
Fibre Channel represents the combination of connectivity and networking.
While SCSI currently moves data at 40 and 80 Megabytes per second, Fibre
Channel currently moves data at one Gigabit per second (referred to as 1
Gbps, also referred to as 100 Megabytes per second or 100 MB/sec or 100
Mbps; its all the same thing).
Another benefit of Fibre Channel is that it is bi-directional, which
means that it can send and receive data at the same time. This feature
effectively doubles Fibre Channel's band width to 2 Gbps. SCSI
functions in one direction, either read or write. Fibre Channel
overcomes many of the limitations of SCSI.
Does this mean that SCSI is dead? No. In our experience technological
shifts of this magnitude occur over long periods of time and, in many
cases, both technologies co-exist for prolonged periods. We would refer
to this migration as evolutionary, not revolutionary. We predict that
both connectivity solutions will co-exist for another five to ten years,
albeit at a diminishing rate.
Fibre Channel Traverses Greater Distances
When storage was directly connected to servers, a short SCSI cable did
the job. With the expansion of data and the desire to centrally locate
and manage storage, SCSI's short distance limitations have lent Fibre
Channel another leg-up. While SCSI reaches distances of 12 meters, Fibre
Channel can span up to 10 kilometers.
Fibre Channel vs. Ethernet
Backing up in time for a moment, Fibre Channel actually came to the
market in the late eighties to be a data communications networking
technology; in other words, instead of using Ethernet one would use Fibre
Channel. In fact, Ancor Communications, one of the current leaders in
Fibre Channel networking, was founded in the late eighties as a LAN
company.
The benefit of Fibre Channel was that it could transmit larger packets of
data effectively (referred to as "block level" data transfers), while
Ethernet was more focussed on small packets. In the end, Ethernet won
because of its backing by the government and its broad-based support from
multiple organizations. Note, there was no real need at the time to send
block level data over the LAN, that was being handled behind the server
by SCSI which at the time did not have a demand to be networked.
In the mid 1990's, when SCSI began to look like it was running out of
gas, companies such as EMC and Brocade began to realized the benefits
that Fibre Channel could bring by networking storage.
Fibre Channel Components
Disk drive suppliers and component manufacturers also realized the
benefits that Fibre Channel components could offer. Today, Fibre Channel
technology is not only used for connectivity between servers and storage,
it is also becoming a preferred component technology. The benefit of
Fibre Channel components are 1) their ability to provide thinner internal
connectivity within systems, thereby enabling smaller systems, 2) that
they produce less heat, and 3) that they enable faster transmissions.
For example, Network Appliance was able to triple its Filers' capacities
to 1.4 terabytes, simply by using Fibre Channel drives instead of SCSI
drives.
Ethernet's TCP/IP Started Out Servicing Government
TCP/IP was originally design by the US Department of Defense and largely
used in governmental agencies and universities. However, it was the use
of TCP/IP to develop the Internet that gave it the ultimate upper hand
over competitors such as Novell's SPX/IPX and IBM's SNA protocols.
Ethernet Is Cumbersome
Ethernet uses a TCP/IP stack (Transmission Control Protocol/Internet
Protocol). The TCP part is for data transmission (what is being sent)
and the IP part is for communication (where to go and how to get there).
Ethernet works well for small bits of information, but generates too much
overhead and is too unreliable to effectively transmit larger blocks of
data.
The overhead in TCP/IP is partly a result in its smaller framing size (1k
frames), relative to Fibre Channel (2k frames which can be grouped
without interrupting the host). Further, TCP/IP often routes each frame,
or packet, differently; whereas Fibre Channel logically groups packets.
The result is that TCP/IP often drops and/or loses packets of data when
the network becomes congested. Lost data must then be retransmitted.
TCP/IP also has a more cumbersome instruction set (over 5000 instructions
per stack versus a few hundred in Fibre Channel), which requires the
utilization of more CPU cycles than Fibre Channel. While TCP/IP does the
job for file level data on the LAN, it does not currently meet the
performance requirements of block level storage transmissions.
Okay... let's look at this in a way that is easier to understand. Think
of a wooden log that you need to transport between point A and point B.
The log represents the data. Ethernet takes the log at point A, puts it
through a wood chipper, sends the various pieces to point B through
multiple paths, loses pieces along the way and reassembles the log at
point B. On the other hand, Fibre Channel takes the log at point A,
breaks it into a couple of pieces, establishes a dedicated connection to
point B, links the pieces together end-to-end, transfers the pieces to
point B over the dedicated connection and reassembles the log at point B.
There is not any debate that Fibre Channel is a more robust block level
data transfer technology! Also note that Gigabit Ethernet, although it
raised its data transfer rate 10x by moving to 1 Gbps, only increased its
throughput by approximately 3x as a result of its high overhead (one
could argue the increase is 2x or 4x, depending on the method of
measurement). Therefore, we believe the argument that Ethernet is
growing to higher capacities is largely irrelevant. In our view, in
order for Ethernet to match Fibre Channel it must reconfigure its
protocol.
We believe two solutions have been in the works to make Ethernet TCP/IP
more robust. The first is stacking encapsulated SCSI over TCP with IP.
The second approach is to beef up the current TCP/IP protocol to function
more like today's Fibre Channel. We believe either approach would take
many years to develop.
The Only Available Solution Ready For Primetime Today Is Fibre Channel
In the end, IT (Information Technology) professionals need a solution
today! The promise of Ethernet is, in our estimation, at least two or
three years out. Even then the revised protocol must gain the support of
broader industry groups to work in a unified direction to create hardware
and software support. Also, the resulting Ethernet stack will not be
compatible with today's Ethernet, so significant additional customer
investments must be made to effect the migration. We believe there are
going to be a heck of a lot of Fibre Channel networks deployed between
then and now and we have yet to met an IT professional who is in favor of
coldly abandoning a technology which they have spent years installing.
Therefore, we have to conclude that Fibre Channel will be around for some
time.
While we have heard some very good arguments that IP is a preferred
protocol as a result of its broad acceptance and use on the Internet, we
believe it is many years away from being able to service block level
storage data traffic reliably.
What Has EMC Said And Done On The Subject?
We believe EMC helped start this whole debate back in December of 1995
when it bought McDATA Corp., which was one of the first Fibre Channel
networking companies. Since then EMC has generated over $2 billion of
Fibre Channel revenues. Although these revenues are not all Fibre
Channel networking per se, we believe it is an interesting statistic in
light of today's debate.
EMC's Recent Announcement Of SRDF Over IP Is What Has Ignited Today's
Debate.
One of EMC's most recognized developments has been its ability to
remotely mirror multiple Symmetrix storage subsystems in multiple
locations real-time. The benefit of this is the ability to recover from
a Disaster in which all systems in one location fail; this is why its
referred to as Disaster Recovery (DR). The lack of being able to recover
data can cost companies millions of dollars in past, current and future
business.
Historically, remote mirroring has gone over leased T1 lines in order to
interconnect multiple sites; remember, IP is not robust enough to handle
block level storage traffic effectively. Recently, EMC announced that
its Symmetrix Remote Data Facility (SRDF) is available over IP. So now
IT managers can take advantage of the Internet and cut costs be using
SRDF over less expensive IP instead of having to lease expensive T1
lines? No!
Yes SRDF can go over IP, but not for Disaster Recovery as a result of the
high latencies associated with IP. EMC's SRDF over IP only supports
asynchronous applications. In other words, you can only use SRDF over IP
for static applications such as Internet content. We believe this
actually opens up new opportunities for EMC which complement SRDF's DR
functions.
EMC Has Entered The NAS Market To Offer Direct File Access Over IP.
A second application EMC is using IP for is to attach storage directly
onto Ethernet LANs in order to serve up file data directly to clients
(note: file data, not block level data), this is usually referred to as
Network Attached Storage or NAS. EMC's NAS solution connects a
Symmetrix directly to the LAN through its Celerra File Server. EMC can
also connect that same Symmetrix to a Fibre Channel based storage network
behind the server to keep block level data transfers separate. Notice
the convergence of storage networking behind the server (SANs) and
storage access directly on the LAN (NAS).
EMC Is Agnostic To The Debate Between Ethernet And Fibre Channel.
We believe a third application EMC is anticipating is Ethernet
connectivity between servers and storage. At the same time, it is our
understanding that EMC sees storage networking traffic remaining separate
from file data traffic. In other words, even if Ethernet TCP/IP were
just as robust as Fibre Channel, storage and datacom networks would
continue to be separate.
If the networks do, in fact, remain separate, we believe this will
benefit Fibre Channel storage networking which is continually gaining
momentum. We believe Fibre Channel and Ethernet networking technologies
will begin to overlap in the three to five year time frame.
EMC has told us that they do not foresee any precipitous change in the
storage networking landscape in the next two years. In other words,
Fibre Channel will continue to be the most robust block level data
storage networking technology for the next two years. Furthermore, they
do not have any preference as to which technology customers ultimately
favor.
Brocade's Move To Partner Up With ONI
We believe Brocade's recent ONI (Optical Networks) partnership
represents a significant shift in the dynamics of networking. The
agreement promotes Fibre Channel to fiber-optic connectivity across
Metropolitan Area Networks (MANs). We believe partnerships between voice
networking, datacom networking and storage networking companies has been
imminent. We believe Brocade has been strategizing with most major
networking companies in an effort to create synergies between various
networking platforms. We believe Brocade will continue its efforts to
leverage its core competencies across other networking platforms,
including Ethernet and ATM. In the end, we believe Brocade is a <b<storage
networking company, not a Fibre Channel networking company.
It is important to note, the most robust storage networking platform
available today is Fibre Channel. Therefore, companies such as Brocade
use Fibre Channel. Many Fibre Channel networking companies are working
to develop networking capabilities across a wide variety of networking
platforms. Evidence of this can be seen in Gadzoox (ZOOX, $57.75, 3H)
recent acquisition of SmartSAN. SmartSAN has been developing Fibre
Channel to ATM and Ethernet connectivity. Note: we believe any competing
storage networking platform is at least two or three your away and will
take much longer to gain broad based adoption.
We expect many announcements similar to Brocade's ONI partnership
throughout this year and well into the future.
|
