Here is a repost of the material posted on Yahoo:ANCR by John 316 today:
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
Brocade Communications Systems, Inc. (BRCD - 179 5/16) April 3, 2000
Ashok Kumar, CFA, 650.328.5437,
The Fibre Channel Vs. Ethernet Debate
How Many Angels Can Dance On The Head Of A Pin?
The introduction of serial SCSI over IP networks, as opposed to Fibre Channel networks is generating a heated debate, with the two communities of interest regularly firing volleys at each other. Let's explore the pros and cons:
ú Fibre Channel is a serial extension of SCSI. As such, it is designed to retrieve disk files block by block. Various extensions have been identified including, to everyone's confusion, IP over FC. (This may have been Pearl Harbor|)
ú Gigabit Ethernet is the new-generation Ethernet. It is just beginning to enter wide use. It too has extensions, including, to no-one's surprise, SCSI over IP.
ú The two systems share a great deal of structure. They use the same specification for the wiring/optical fibre, but diverge widely in the sequence of bits each message contains. Both systems go through switches that inherently do the same things. Both systems are expected to see rapidly expanding use over the next five years.
So, why are they different? There are some factors that derive from the different missions of the two connection schemes. For instance, the Fibre Channel community is convinced they offer a deterministic dataflow, while Ethernet by design allows packets to be lost. It's true this is an advantage with small configurations, but in fact it is a disadvantage for Fibre Channel in large configurations, since Fibre Channel has more difficulty handling lost packets.
Fibre Channel is faster than Ethernet.. It carries more payload per second, and has inherently lower latency. However, much of this comes from the extra hardware in the NICs and switches. Even so, Fibre Channel does win out, because the software protocol stack is smaller.
Page 2
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
On the other hand, there are many more players addressing the IP space. This will drive Gigabit cost, switch performance and NIC performance. We won't see a replay of the dumb NIC' scenario that so hurt 3COM (COMS). Alteon (ALTN), for example, is now delivering NICs with two RISC processors integrated into their chipset. IP switches are generally more sophisticated than FC switches, with most having capabilities into the higher layers of the OSI model.
Given the convergence of hardware sophistication, are we moving to a level pinhead? Sadly the answer is "NO|" The two communities are not willing to merge, and enough equipment is designed and deployed to make any likely technical convergence moot. Fibre Channel and Gigabit Ethernet are here to stay|
There is a very real possibility that the most important question for this debate is not "Which is better - Fibre Channel or Gigabit?" Rather, the real question is "What is storage?" The drop in the price of compute power, and even more notably, the drop in the price of storage are also dramatically influencing the economics of interfacing disks to networks. We see strong support in the IT community for easily installed Network-Attached Storage (NAS). This is a form of "Disk over IP" and one may wonder if the cost/performance/management advantages of this approach are resonating better with IT professionals than any claim that "Fibre Channel is faster".
Page 3
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
What This All Means
Despite the price and acceptance advantages of Gigabit, Fibre Channel-based storage will show gains in acceptance over the next five years. Simply put, SANs are better than the "old" ways of connecting storage on clustered systems. Nonetheless, SAN revenue growth has been lowered by the stress of standards wars and the demands on a diminishing IT staff to learn a new protocol/topology. This has opened the door for NAS technology to take significant storage market share. In the next few years, NAS will "grow up," fronting storage in the Terabyte range, and will generally be attractive to developers because its increasingly smart content moves the product out of commodity pricing.
Gigabit demand will of course explode as high-density server appliances with high-throughput channels correct server performance content delivery bottlenecks. This is one of the safer bets in this world| Overall, we believe NAS and high-density servers, both with Gigabit, will show high revenue CAGR.
The Future
Here we go again| It's 2002, and the debate is about the protocols to be used on 10Gigabit links. Have we learnt anything? There are three (yes, we are ashamed to say, three|) standards that will each solve all the world's problems. We have Fibre Channel and IP joined by InfiniBand. Physics has been searching for the GUT (Grand Unification Theory, or the Theory of Everything) for 20 years. The computer industry needs a GUP (Grand Unified Protocol)|
Investment Opinion
With the FC connectivity market peaking earlier and at a smaller size, what are the ramifications to Brocade's valuation? Did anybody say Timberrrrr|
Price to Sales CY00 CY01
Ancor (ANCR-#) 29.9x 18.7x
Brocade (BRCD) 79.4x 48.7x
Emulex (EMLX-#) 25.1x 16.4x
QLogic (QLGC-#) 38.3x 27.4x
Page 4
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
Myth (M) Vs. Reality©
M: Ethernet allows for packet loss.
R: Ethernet packet loss is overblown as a reason to choose FC. Ethernet packet loss occurs due to CSMA-CD (ie. shared media) links and as a congestion relief mechanism in very busy switches. Very busy Ethernets are inefficient for handling large block transfers because packets lost to congestion cause many packets to be retransmitted. As a defense mechanism, congested Ethernet drops packets - this is well known and, today, reasonably well handled. The FC community simply doesn't have that much experience in building very large, very busy networks. Further, because the design decisions were very different than Ethernet, that experience is only partly applicable.
M: The TCP stack can consume 50%-100% of the CPU cycles versus 5% CPU overhead for FC.
R: Gigabit Ethernet (GbE) can consume these CPU levels, slower Ethernet will not. Of course, GbE is the most interesting, but not the ONLY option available. FC is a relatively efficient protocol, but some of that efficiency is gained at the expense of more complex hardware.
M: FC is optimized to transfer large blocks of data with minimum processing overhead. Ethernet packet size is 1518 bytes vs. FC frame size of 2112 bytes with provision to link 65,536 frames - or a transfer size of 128MB.
R: A true statement, but slightly misleading. It is possible that transaction systems could process more I/Os per second if smaller frames are used (512 byte or 1K). Storage I/O varies a great deal depending on the application. Large frames are not necessarily always the best. Small blocks of data might actually be better for transaction processing. There is no reason storage GbE could not be implemented to have a larger frame size. Storage GbE is the idea that GbE vendors would create specialized equipment to pursue this opportunity.
M: FC provides dynamic flow control by monitoring congestion on fabric.
R: FC has a flow control mechanism between end-points (Class 1, 2 ACK based) and between ends of a link (RRDY). However, neither mechanism really does much about monitoring congestion in the fabric and there are some pathologic conditions which can arise within switches - especially in multi- switch fabrics - that are not well handled. GbE flow control might actually work fairly well. Maybe not quite as good, but will it be enough of a difference for the market to care? It will likely be a techno-babble discussion that few want to hear.
Continued
First Call Corporation, a Thomson Financial company.
All rights reserved. 888.558.2500
News Retrieval in PIPER2491B.GWS - - Apr 03 08:46
IND/F US/F NA/F MMN/F SEM/F BRCD ANCR EMLX QLGC JNIC ZOOX VIXL
08:28 FPJ Brocade (BRCD): The Fibre Channel Vs. Ethernet Debate; Buy; 2 o
Page 5
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
Brocade Communications Systems, Inc.
April 3, 20000 - Part 2 of 2
M: 1FC simplifies management by integrating critical services like named service into the network rather than the device itself.
R: This argument is quite a stretch. Ethernet networking equipment is incredibly sophisticated, providing all kinds of services and multiple levels. It could also be argued that Fibre Channel's lack of standards in exchanging this service information today is a fairly serious shortcoming.
M: FC error detection and recovery embedded is in hardware/firmware (layer 2). IP relies on software-based error recovery schemes (TCP) which includes overhead and reduces performance
R: This is true, but there is work underway to provide layer 4 (TCP) error control in hardware. Integrating the TCP/IP stack in hardware and integrating into the subsystem controller is a non-trivial task. But, we believe this technology will be available by summer.
M: Ethernet is at least 2/3 years away from developing a block level storage networking alternative.
R: First of all a block server (disk subsystem) is such a dumb thing.Why perpetuate an old model? (Except that it works.) While there will be some latency in deployment, the complexity is overblown. It is not necessary for disk drives to develop SCSI/ethernet interfaces to make these new extensions work. In fact what is likely to occur is that subsystem companies (EMC for example) will develop ethernet based controllers that decode SCSI over ethernet protocol and then write to internal SCSI, ATA or Fibre Channel drives. Disk drives are cheap internal components that get attached by internal plumbing inside subsystems
M: Cisco (CSCO) SCSI/IP is primarily applicable for low performance applications in the SOHO segment
R: Right, as if CSCO is targeting the SOHO market. One of the dangers in SCSI/IP is that it can work in lower performance business applications at far lower cost. What about all the systems out there that can use 10MB transmissions over Fast Ethernet at far less cost than FC? But that is a different discussion, the fact is - customers will likely install specialized SCSI/IP networks tuned for storage. It won't be the best initially, but it will be optimized as the Ethernet community figures out the requirements.
M: EMC (EMC) Symmetrix Remote Data Facility (SRDF)-over-IP is only for asynchronous applications. It cannot be used for disaster recovery due to high latency (only for static apps).
Page 6
*** U.S. Bancorp Piper Jaffray *** U.S.
R: Yes, so? Almost any remote mirror will be asynch for FC too. Contention that SRDF/IP cannot be used for disaster recovery is pure hogwash. It shows a lack of understanding of disaster recovery. Why mirror at all, if not for disaster recovery?
M: Brocade recently released certain elements of the routing protocol - Fabric Shortest Path First (FSPF) to NCITS. This should promote interoperability among switches.
R: This is astute wool pulling. Based on OSPF, FSPF is not hard to reverse engineer. Ancor (ANCR - #) already did it. As long as it's headed for de- facto standardization, why not do the right PR thing and give it to the world? A very good move by Brocade. But the question needs to be asked - "does FSPF provide optimized routing for FC traffic?" There are not enough people that understand this well enough to have an open dialogue. Brocade is talking out of both sides of their mouth with this one. They say Ethernet is not good enough for storage, but then they adopt an Ethernet routing standard for their switches.
M: About half of Brocade's deployments are at test sites.
R: Probably some high percentage like this, but there is no data to support it however. One has to ask why we are not seeing more customer satisfaction stories in the news. Could it be because there actually are not that many satisfied FC customers? That is a scary thought
M: Due to closed implementations and a black box approach, it is questionable if SAN deployment can ever transition from islands to integrating legacy servers and storage systems.
R: Yes, but only after customers have the skill set to accomplish the task. There is a huge dearth of FC talent. Ethernet talent, on the other hand, is everywhere. This is a huge advantage of SCSI/IP over FC. The FC industry faces a huge uphill battle to educate the market.
M: Odds are that the FC connectivity market opportunity of $3-3.5B by 2003 is over blown?
R: We believe that the FC opportunity is overstated by at least 30% as a large percentage of the projection is based on assumptions that are proving to be not so. Our sense is that the IT community would like there to be a convergence but the parties to that convergence are as entrenched as ever. Short term, the trend towards NAS poses more of the threat toward SAN FC model than CSCO SCSI over IP. NAS with performance and scalability plus scalable, high performance SQL appliances could make things very interesting.
M: Fibre Channel networking companies will become a relic.
R: Fibre Channel networking companies will be able to change spots and develop ethernet storage switches, if they start early enough. But will they get past their dogma?
Page 7
*** U.S. Bancorp Piper Jaffray *** U.S. Bancorp Piper Jaffray ***
Brocade Communications Systems, Inc. (BRCD - 179 5/16) April 3, 2000
Ashok Kumar, CFA, 650.328.5437, akumar@pjc.com / Paul Mansky, 612.342.6474,
pmansky@pjc.com
Some or all of the following hedges may pertain: (#)U.S. Bancorp Piper Jaffray Inc. makes a market in the company's securities. (~)A U.S. Bancorp Piper Jaffray Inc. officer, director, or other employee is a director and/or officer of the company. (@)Within the past three years, U.S. Bancorp Piper Jaffray Inc. was managing underwriter of an offering of, or dealer manager of a tender offer for, the company's securities or securities of an affiliate. Additional information is available upon request. Not FDIC Insured No Bank Guarantee May Lose Value This material is based on data obtained from sources we deem to be reliable; it is not guaranteed as to accuracy and does not purport to be complete. This information is not intended to be used as the primary basis of investment decisions. Because of individual client requirements, it should not be construed as advice designed to meet the particular investment needs of any investor. It is not a representation by us or an offer or the solicitation of an offer to sell or buy any security. Further, a security described in this publication may not be eligible for solicitation in the states in which the client resides.
U.S. Bancorp Piper Jaffray individual officers, employees, or members of their families may own the securities mentioned and may purchase or sell those securities in the open market or otherwise. In the United Kingdom, this report may only be distributed or passed on to persons of the kind described in Article 11(3) of the Financial Services Act 1986 (Investment Advertisements) (Exemptions) Order 1996 (as amended by the Financial Services Act 1986 (Investment Advertisements) (exemptions) Order 1997). Securities products and services offered through U.S. Bancorp Piper Jaffray Inc., member of SIPC and NYSE, Inc., a subsidiary of U.S. Bancorp.
¸2000 U.S. Bancorp Piper Jaffray Inc., 222 South Ninth Street,
Minneapolis,
Minnesota 55402-3804 piperjaffray.com
First Call Corporation, a Thomson Financial company.
All rights reserved. 888.558.2500
Nondeposit investment products are not insured by the FDIC, are not deposits or other obligations of or guaranteed by U.S. Bank National Association or its affiliates, and involve investment risks, including possible loss of the principal amount invested.
Past performance does not guarantee future results. We consider our sources reliable. Accuracy and completeness are not guaranteed. Information is subject to change. Transactional details should not be relied on for tax purposes and do not supersede normal trade confirmations or statements. Messaging outside U.S. jurisdictions from U.S. Bancorp Piper Jaffray to non-institutional parties is not intended for solicitation purposes.
Electronic mail sent through the Internet is not secure. We will not accept time-sensitive, action-oriented messages, transaction orders, fund transfer instructions or check stop payments electronically.
If you are not the intended recipient, notify the Sender. This information is intended only for the person named above and for the purposes indicated. Do not distribute this message without written consent of the author. Non-business opinions may not reflect opinions of U.S. Bancorp Piper Jaffray and its affiliates.
U.S. Bancorp Piper Jaffray and its affiliates reserve the right to monitor all e-mail.
Securities products and services are offered through U.S. Bancorp Piper Jaffray Inc., member SIPC and NYSE, Inc., a subsidiary of U.S. Bancorp. |
