Sie sind auf Seite 1von 29

Fibre Channel over Ethernet:

An Introduction
Session ID: SSA01
Peter Mescher
STG Solution Central Team, RTP, NC, USA

2008 IBM Corporation

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Disclaimers
NOTES:
All statements regarding IBM's future direction and intent are subject to change or withdrawal without notice, and represent goals and
objectives only.
Information about non-IBM products is obtained from the manufacturers of those products or their published announcements. IBM has not
tested those products and cannot confirm the performance, compatibility, or any other claims related to non-IBM products. Questions on the
capabilities of non-IBM products should be addressed to the suppliers of those products.
Prices are suggested US list prices and are subject to change without notice. Starting price may not include a hard drive, operating system
or other features. Contact your IBM representative or Business Partner for the most current pricing in your geography.
Any proposed use of claims in this presentation outside of the United States must be reviewed by local IBM country counsel prior to such
use.
The information could include technical inaccuracies or typographical errors. Changes are periodically made to the information herein; these
changes will be incorporated in new editions of the publication. IBM may make improvements and/or changes in the product(s) and/or the
program(s) described in this publication at any time without notice.
Any references in this information to non-IBM Web sites are provided for convenience only and do not in any manner serve as an
endorsement of those Web sites. The materials at those Web sites are not part of the materials for this IBM product and use of those Web
sites is at your own risk.
IBM makes no representation or warranty regarding third-party products or services including those designated as ServerProven,
ClusterProven or BladeCenter Interoperability Program products. Support for these third-party (non-IBM) products is provided by non-IBM
Manufacturers.
IBM may have patents or pending patent applications covering subject matter in this document. The furnishing of this document does not
give you any license to these patents. Send license inquires, in writing, to IBM Director of Licensing, IBM Corporation, New Castle Drive,
Armonk, NY 10504-1785 USA.

17-Aug-08

2008 IBM Corporation

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Agenda
 Overview of Fibre Channel over Ethernet (FCoE) as a
product
 Protocol Summary

17-Aug-08

2008 IBM Corporation

This presentation is largely divided into two halves, the first half is a short
introduction as to what FCoE is and how it is likely to play in the market.
The second half is an introduction as to how FCoE works. It will be of interest to
anybody that will have to actually implement FCoE in the field.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

What Is Fibre Channel over Ethernet?


 A direct mapping of Fibre Channel onto Ethernet
 Aims to reduce cost through sharing of Ethernet
infrastructure
 New standard specified by ANSI T11 Committee
within FC-BB-5
 Backed by virtually entire storage networking industry,
including IBM, Brocade, Cisco, QLogic, HP, and Intel

17-Aug-08

2008 IBM Corporation

In short, all FCoE primarily consists of is the encapsulation of Fibre Channel frames
inside of Ethernet frames. Thats it. There is some additional bits of the protocol
that deal with device discovery and initialization, but on the whole, it is a very simple
standard.
The primary goal of FCoE is to reduce cost by sharing the Ethernet infrastructure
virtually all servers currently have.
While the primary driving force behind FCoE is the current Storage and SAN
vendors, the standards effort is currently being handled by the ANSI T11 committee;
the same standards body that manages the main Fibre Channel standards. Indeed,
FCoE is buried inside a larger T11 FC standard, FC-BB-5.
FCoE has received wide backing in the storage industry, including all of the major
SAN disk vendors, the four major SAN infrastructure vendors (Brocade, Cisco,
QLogic, and Emulex), and Intel.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

General Characteristics
 Simple No complicated flow control or frame loss
detection
Entire standard 24 pages

 Viable alternative to native Fibre Channel


 Generally intended for, but not limited to, 10Gb
Ethernet

17-Aug-08

2008 IBM Corporation

The entire FCoE standard (as of the June draft) takes up only 24 pages. For those
that have not read standards documents before, this is extremely short. FCoE is
completely lacking in session protocols (FCP/SCSI handles that), flow control (there
isnt any), or detection for frame loss (FC takes care of that function.)
It will likely be a viable alternative to native Fibre Channel, at least in some
environments. (More on that later in the presentation.)
The first mainstream products will likely use 10Gb Ethernet, although it will certainly
function over 1Gb links.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Basic FCoE Installation


SAN A

LAN

SAN B

FCF/
Ethernet
Switches

Servers
Fibre Channel

Ethernet

17-Aug-08

2008 IBM Corporation

The most simple sensible FCoE installation is shown above. It consists of a pair of
servers, each connected to a pair of Ethernet/FC splitters. (The part of these
switches that handle the FCoE functions are called Fibre Channel Forwarders in
the standard.) The FCF/Switch units each connect into the LAN and a single SAN
Fabric.
We assume the servers use two separate ports for redundancy.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Why use FCoE?


 Large hosts generally use two networks right now:
Ethernet and Fibre Channel
 FCoE eliminates need for additional Host Adapters
Reduces Cost
Reduces slot and bus space usage
Very helpful for Blades

17-Aug-08

2008 IBM Corporation

Why should we use Fibre Channel over Ethernet? Currently, SAN-attach servers
have two networks: Ethernet and Fibre Channel. FCoE eliminates the need for
separate adapters for networking and storage. It will also eliminate the need for
multiple edge switches connected to a single host; this reduces total networking
cost.
One big advantage of this is in Blade servers, which typically have very limited
expansion options. Also this can be useful for 1U pizza box servers or dense
servers used for clustering.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE vs. FCIP/iFCP


 Not really equivalent
 FCoE Direct replacement for FC links
 FCIP/iFCP Generally used for Distance Extension
FCoE unsuitable for long distances or complex
networks

 No FCIP or iFCP storage devices or HBAs

17-Aug-08

2008 IBM Corporation

How is FCoE different from FCIP or iFCP, the two existing Fibre Channel over
Networking standards? The simple answer is that they are not really equivalent.
FCoE is a direct replacement for local Fibre Channel links. The most common
usage for FCIP and iFCP is distance extension, where the use of dark fiber is
cost-prohibitive or unnecessary.
FCIP and iFCP are currently limited to gateway devices, and it is not expected that
these standards will extend into hosts or storage devices. (Although iFCP was
designed for this use, it never became widespread.) FCoE will be deployed on
hosts first, and possibly storage devices much later.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE vs. iSCSI


iSCSI

FCoE

 Local-area, lossless links no


routing allowed

 Allows many hops, lossy


connections, high latency

 Simple encapsulation of Fibre


Channel

 Substantial complexity on top


of TCP very different from
FCP

 Low overhead 1% over


regular FC

 Overhead varies typically


higher.

 Fibre Channel wellunderstood by storage


administrators

 TCP/IP well-understood by
just about everybody else

 Substantial frame loss


catastrophic

 Frame loss merely painful


17-Aug-08

2008 IBM Corporation

The big question in the industry about FCoE is its suitability vs. iSCSI. Each has its
strengths and weaknesses.
FCoE is designed, at its core, to be run over Ethernet networks of the same quality
and size as their FC equivalents. iSCSI is meant to run over any TCP/IP network.
Take a Fibre Channel Frame, and insert it into an Ethernet frame; as stated before,
that is FCoE in a nutshell. That will make up the majority of FCoE traffic; indeed, an
early draft of the standard consisted of only that. iSCSI, on the other hand, is a
complete re-implementation of the FCP (SCSI over Fibre Channel standard), on top
of the well-known TCP/IP stack. Whether or not this makes for a better protocol
overall is a matter of great debate.
FCoE is a very low-overhead protocol; it only adds 1% on top of the already lowoverhead Fibre Channel. iSCSI overhead varies but it is generally greater than FC
overhead due to the TCP/IP stack.
Fibre Channel is familiar to virtually every enterprise storage administrator, and has
well-known behaviors in high-performance environments. TCP/IP, on the other
hand, is well-understood by a huge population, and is far more widespread in the
field than Fibre Channel. iSCSI is not necessarily well-understood, but from a user
perspective, there isnt much to it. (The protocol itself is quite complex, but users
are largely insulated from it.)
In a poor network, significant frame loss is catastrophic to FCoE, just as it is with
Fibre Channel. iSCSI has all of the error recovery capabilities of TCP/IP. While
frame loss will really slow it down, operations do not slow to a crawl, as they do with
Fibre Channel.

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE vs. iSCSI long-term


 No overwhelming advantages for one vs. other
 FCoE appropriate for existing SAN installs
More SCSI-like
No complex, gateway for non-iSCSI-native systems

 iSCSI serious choice for new solutions


Takes advantage of substantial existing IP infrastructure
and knowledge

 Lossless Network needed for FCoE helpful to both


 iSCSI likely will win, in the end
10

17-Aug-08

2008 IBM Corporation

What is the long-term market outlook for the two protocols?


In the short term, there is no clear winner. The Fibre Channel networks FCoE is
designed for arent going away any time soon; FCoE (and FC itself) is more SCSIlike, and has a history ultimately going back decades through to the original IBM
OEMI Channel design. In addition, if you wish to use iSCSI, but have storage units
without iSCSI adaptors, you must use an expensive and slow gateway device. The
design of a FCF, on the other hand, is quite simple and easy to implement in
hardware, and to operate at line speed.
On the other hand, for new installations, iSCSI is a serious contender. It takes
advantage of the extremely well-understood and widespread TCP/IP protocol stack,
and virtually every decent-sized IT department has a skilled TCP/IP administrator.
This is not true for Fibre Channel.
One commonly cited disadvantage for FCoE is the need for a lossless Fibre
Channel network (as in, one implementing flow control.) While this is a drawback, it
is only a minor one, as a lossless fabric would be of great benefit to iSCSI as well; it
would vastly reduce the performance hit caused by lost frames in a congested
network.
In the end, it is this authors opinion that iSCSI will eliminate both Fibre Channel and
FCoE. FCoE will serve as a transition to pushing storage over Ethernet networks,
but once that transition is complete, there is no need for Fibre Channel at all any
more, and it will likely give way to iSCSI. Any theoretical performance advantages
of Fibre Channel/FCoE/FCP over iSCSI will likely not be outweighed by the added
administration overhead of the Fibre Channel network.

10

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Software vs. Hardware Implementations


 FCoE is available as a software driver for most
operating systems
Primary use is proof-of-concept and interoperability test

 Virtually all production deployments expected to be


hardware based
New Converged Network Adapters (CNA)
In reality, FC HBAs with an Ethernet MAC layer

11

17-Aug-08

2008 IBM Corporation

Currently FCoE is available as a software driver for most operating systems. This
driver has been used in protocol development and interoperability testing. The
biggest issue is that the SCSI layers necessary for this driver have to be written
from scratch, as the SCSI function has traditionally been done in hardware. This
could be a significant source of performance and stability issues until the driver
becomes more mature.
It is expected that field deployments will virtually all be hardware based, at least
initially. These installations will use new Converged Network Adapters that
combine an ordinary Ethernet NIC with the hardware necessary to support Fibre
Channel.
In reality, these will essentially be FC HBAs with an Ethernet MAC layer added in.
Because of this, they are expected to be slightly more expensive than FC HBAs
(due to low initial volumes), and they will certainly be much more expensive than
ordinary Ethernet NICs. The nearest TCP/IP equivalent to these Converged
Network Adapters would be iSCSI-offload HBAs, which are not common in the
marketplace, although several were developed. (Software iSCSI initiators generally
provide acceptable performance on modern CPUs.)

11

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Disadvantages
 No flow control (not even buffer credits)
Relies on one or more optional Ethernet extensions
802.3-2005 PAUSE mechanism
802.1Qau Congestion Management

Required to deal with potential FC congestion, even if


Ethernet uncongested

 Assumes high-quality network


Lossless Ethernet network required

12

17-Aug-08

2008 IBM Corporation

Flow control has traditionally been a weak point for stability in Fibre Channel
networks. FCoE is even worse off; it has absolutely no flow control of any kind, not
even the rudimentary system of buffer credits used in Fibre Channel.
While the FCoE protocol itself does not require any sort of flow control, any realworld deployment certainly will. This can be done through any one of several
methods, none of which is currently very widespread. These include the 802.32005 PAUSE mechanism or the 802.1Qau Congestion Management protocol.
Even if the Ethernet network itself is uncongested, the FCoE system must be
prepared to deal with congestion on the Fibre Channel network itself or congestion
within the host. The alternative to implementing flow control would be for the
Ethernet switch to drop congested frames, which would cause severe performance
issues in the FCoE end devices.

12

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Possible Industry Deployment




FCoE intended for 10Gb Ethernet links


1Gb functional, not recommended

1. Blades with Ethernet/FCoE splitter on in-Chassis


switch
2. Edge switches / Director blades
3. Entire Directors / Core Ethernet switches (same
thing?)
4. Storage

13

17-Aug-08

2008 IBM Corporation

FCoE is really intended to be used on 10Gb Ethernet networks. While it will work
on 1Gb networks, this opens up the possibility of contention for the network
resources between TCP/IP and FCoE traffic. This is not expected to be an issue
with 10Gb links.
Initially, the first mainstream products are expected to be blade servers (with CNAs)
connected to an Ethernet switch that will split off the traditional Ethernet and FCoE
traffic. (Technically, this is called a Type C Fibre Channel Forwarder in the FCoE
standard.)
Next will be edge switches or Director blades with FCoE-capable ports. (These are
called ENodes.)
Then, we could expect to see what could best be called an Ethernet Director that
will instead have just a few Fibre Channel ports to talk to storage devices.
Lastly, we might eventually see FCoE storage devices. This could result in FCoE
SANs with absolutely no Fibre Channel ports; there would just be some software
stubs to handle Fibre Channel services. This, however, would require a
substantial overhaul of Ethernet switch architecture.
Beyond this stage, iSCSI will likely become the most common protocol, with FCoE
still used in Mainframe environments, which are probably not going to implement
iSCSI any time soon.

13

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Summary of FCoE as a Product


 FCoE, in summary:
A new standard designed to converge Fibre Channel
SANs and Ethernet networks
Simple in concept and design
Will be initially used with smaller servers
Fundamentally different from iFCP and FC/IP
In competition with iSCSI

 Questions? (Information on protocol details to follow)


14

17-Aug-08

2008 IBM Corporation

In Summary:
Fibre Channel over Ethernet is a new standard that is likely to become quite
common in the storage marketplace. Its primary design point is the convergence of
Fibre Channel with Ethernet networks. It is a simple concept, and can be easily
implemented. In the beginning, it will likely be used only on smaller servers
(primarily for cost reasons.)
FCoE is extremely different from other IP-based storage schemes, including the
other Fibre Channel-related protocols of FCIP and iFCP. Long term, the expected
competition will be with iSCSI, which will possibly displace it some years from now.

14

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Terms
 FCoE Mapper (FCM) Provides translation between
Ethernet and Fibre Channel
Present in NICs and switches

 FCoE Forwarder (FCF) Ethernet / Fibre Channel


switch
Also provides FCoE discovery and login services
Does not need actual Fibre Channel ports

 FCoE ENode (ENode) Ethernet port w/ FCoE


abilities (software or hardware)

15

17-Aug-08

2008 IBM Corporation

An FCoE Mapper (FCM) is the fundamental product for FCoE. It provides the
encapsulation and de-encapsulation of the Fibre Channel frames onto Ethernet
links.
The Fibre Channel Forwarder (FCF) is the object that will move the frames back
and forth from the Fibre Channel fabric. It should be noted here that it is possible
for the Fibre Channel Fabric to exist only internally; the only requirement for the
FCF is that it provide access to Fibre Channel services and other Fibre Channel
devices. The FCF does not need to contain any actual Fibre Channel ports. A
single hardware unit could contain multiple FCFs; this could be used to provide
VSAN-like services.
FCoE ENode (ENode) Simply an Ethernet port in an end device that can handle
the FCoE protocol. Each ENode can handle one or more Virtual F_Ports.

15

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Terms (cont.)
 FC-MAP Fabric ID
Usually 0E:FC:0z (z = a Fabric ID number, which
defaults to 0.)
Used for first three bytes of Fabric-Assigned MAC
Addresses

16

17-Aug-08

2008 IBM Corporation

The FC-MAP is just another term for the Fabric ID. This allows multiple FCoE
virtual fabrics using Fabric Provide MAC Addresses to run in a single Ethernet
network. This is implemented as a three-byte value that is usually 0E:FC:0z. The
z is the fabric ID. If not used, it defaults to zero. Note that the FC-MAP can be
any three-byte value; the one listed above is just a suggestion.
The FC-MAP is also used for the first three octets of Fabric Assigned MAC
addresses. Because of this, it is important that the FC-MAP not overlap with any
valid IEEE OUI values. (The default values do not overlap.)

16

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Standard
 Listed in ANSI T11, FC-BB-5, Chapter 7
 Two protocols in one standard:
FCoE The encapsulation standard
FCoE Initialization Protocol (FIP) Protocol to provide
device discovery and login services
Use different EtherTypes
Facilitates FIP snooping by non FCF switches

17

17-Aug-08

2008 IBM Corporation

The FCoE Standard is, as mentioned earlier, controlled by the ANSI T11 committee;
the same group in charge of Fibre Channel. FCoE was inserted into recent drafts of
the FC-BB-5 standard; this refers to the 5th edition of the FC standard dedicated to
describing how to carry Fibre Channel frames over non-Fibre Channel networks,
such as IP (FCIP is in this standard), and WAN networks, such as SONET. A
mostly-complete draft of the standard was introduced in June, with only minor
details left to be worked out.
FCoE is made up of two protocols: FCoE itself, which specifies how to perform the
FC/Ethernet encapsulation, and FCoE Initialization Protocol, which handles device
discovery and fabric login. Each of the two protocols has a distinct EtherType.
(0x8906 for FCoE and 0x8914 for FIP) This decision was made so that FCoEaware switches that did not have FCF functionality could automatically configure
themselves to pass FCoE traffic without compromising their Access Control Lists.
Having the FIP frames use a different EtherType makes those frames easier to spot
and snoop.
As a side note, the standard document itself goes into great detail on the
architecture model for FCoE; this is very interesting if you are designing a FCoE
switch, NIC, or software stack. The only thing you, as an end-user, need to know is
that each FCoE-capable Ethernet port can have multiple Virtual F-Ports or Virtual EPorts, as long as your software and hardware supports it.

17

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE and the Protocol Stack

18

17-Aug-08

2008 IBM Corporation

This diagram comes directly from the FCoE Standard. Shaded in grey is where
FCoE is used.
On the left, we have a diagram of a Fibre Channel Forwarder (FCF), which is the
unit that translates frames between Fibre Channel and FCoE. On the right, it shows
the place of FCoE within the protocol stack.
For readers familiar with the OSI 7-layer model, FCoE inserts an additional layer 2
into each frame, on top of the existing FC-2 level, which provides layer-2 functions
in ordinary Fibre Channel.

18

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Frame Format

 Other than normal Ethernet headers, no per-frame


overhead
 Mini-Jumbo Ethernet frames strongly suggested
Not a difficult requirement; virtually all do

19

17-Aug-08

2008 IBM Corporation

As can be seen in the diagram above, FCoE frames are nothing more than ordinary
Ethernet frames with a Fibre Channel frame inserted into it, in its entirety. (The
Ethernet frame can contain an 802.1Q tag, not shown here.)
Since a maximum-size FC frame with a 2112 byte payload is far larger than a
standard 1500-byte possible Ethernet payload, your Ethernet network should be
able to handle so-called Mini-Jumbo frames. Given that even consumer-level
Gigabit Ethernet switches available for a few Euro implement Jumbo frames, all but
the oldest Gigabit Ethernet switches should be able this requirement.

19

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Frame Forwarding


 Relies on Ethernet Spanning Tree or other forwarding
scheme
 No built-in routing such as FSPF
 Load balancing generally poor

20

17-Aug-08

2008 IBM Corporation

FCoE relies on whatever internal scheme the Ethernet network chooses to use to
direct frames; it has no built-in capabilities for doing so. In most networks, this will
be some variant of the Spanning Tree protocol. However, there is work underway
by various switch vendors to use more sophisticated methods.
This does not mean that FSPF traffic is prohibited over FCoE networks, rather it
means that FSPF is in charge of routing within the actual FC fabric; but
FCoE/Ethernet/Spanning Tree is responsible for directing frames that travel over
the virtual ISLs between fabric islands.

20

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

More-Complex FCoE Topology


Note NonRedundancy!
SAN A

SAN B

FCF

21

Ethernet

17-Aug-08

2008 IBM Corporation

This is one of the suggested topologies for FCoE installations (it actually came right
out of the FCoE ANSI committee.) Its design will be quite foreign to most SAN
architects, who almost always use nothing but completely redundant installations,
with no points of connection between the two fabrics. In networking, the idea of
using two completely isolated networks is extremely rare.
Yes, this does open up users to a complete network outage. However, Ethernet
networks are somewhat simpler in complexity (i.e. usually nothing like FSPF, zoning
propagation, etc.), so complete network outages are somewhat more rare. This is
something storage administrators will have to keep in mind when laying out FCoE
networks.

21

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Network Without Much FC

SAN
Ethernet
Ethernet
Ethernet

FCF

22

Ethernet

17-Aug-08

2008 IBM Corporation

This is another diagram from the ANSI committee. As you can see, none of the
Fibre Channel Forwarders use Fibre Channel for their ISLs. Instead, they each use
an Ethernet connection to communicate with each other. The FCF on the left does
not even have any Fibre Channel devices connected to it at all. This is a perfectly
valid configuration.

22

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Addressing
 Two options
Server Provided MAC Address (SPMA)
Typical Ethernet standard globally unique address

Fabric Provided MAC Address (FPMA)


Format XX:XX:XX:YY:YY:YY
XX:XX:XX = FC-MAP
YY:YY:YY = FCID

 No relation to WWPN of adapter

23

17-Aug-08

2008 IBM Corporation

There are two options for the MAC addresses assigned to FCoE ports: ServerProvided (generally referred to as burned-in) and Fabric-Provided.
Server Provided MAC Addresses are merely normal six-byte IEEE-assigned
addresses. The first three bytes generally refers to the vendor, and the last three
bytes are a unique combination assigned by the vendor.
Fabric Provided MAC Addresses consist of two parts: The first is the FC-MAP, the
second is the FCID for the Virtual F_Port. This is communicated during the FIP
process which will be described shortly.

23

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Addressing (cont.)
 Multiple MACs per adapter expected
One for regular Ethernet traffic (i.e. TCP/IP)
ENode MAC Address

One for each FCoE address


VN_Port MAC Address

 NPIV supported

24

17-Aug-08

2008 IBM Corporation

It is expected that most NICs will end up with multiple MAC addresses: One for
regular Ethernet traffic (and FIP login), and then one for each FCoE virtual port.
NPIV is supported, just as it is with Fibre Channel, and it uses the same mechanism
for doing so as FC. (Perform a FDISC for each additional address.)

24

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Discovery


Utilizes special Fabric Initialization Protocol (FIP)


frames
Separate from regular FCoE frames
Used for FLOGI/FDISC, LOGO, ELP and FCoE
discovery
Source MAC probably burned-in MAC Address

25

17-Aug-08

2008 IBM Corporation

FCoE discovery uses special Fabric Initialization Protocol frames. These are not
regular FCoE frames with Fibre Channel frames embedded inside. They are used
for FCoE discovery and the basic FC login commands FLOGI/FDISC, LOGO and
ELP.
The source MAC for these commands is probably the burned-in MAC Address.

25

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Discovery and FC Login Process


1. Send multi-cast Ethernet frame to All-FCF-MACs
standard address (or specific FCF) with Discovery
Solicitation payload.
Contains:

26

MAC Address used to receive response


The hosts name
Maximum Receive Size
If an FCF itself, include FC-MAP (if using FPMA)

17-Aug-08

2008 IBM Corporation

The first step for a new ENode is to discover the FCF ports the ENode has access
to; it accomplishes this by sending a Discovery Solicitation FIP multi-cast frame to
a special All-FCF-MACs address. Alternatively, if it has a specific FCF preprogrammed, it can send the Solicitation just to that address.
This contains the MAC address the port would like to receive the response on, a
name for the host (i.e. the hostname) and the maximum frame size the port can
process. If the port that is initializing is an FCF, it includes its FC-MAP address, if it
uses FPMA.

26

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Discovery and FC Login Process (cont.)


2. FCF responds with Discovery Advertisement
Advertisement contains:
Priority 1-byte value used to determine which FCFMAC to use, if multiple respond
MAC Address of the FCF-MAC
If using FPMA, the FC-MAP of the FCF
Name of the Switch
Name of the Fabric

27

17-Aug-08

2008 IBM Corporation

All FCF MACs that receive the Solicitation respond with an Advertisement. (If a
single FCF has multiple MACs, each should respond.) This Advertisement
contains:
A Priority value to help the port decide which FCF-MAC to use if it receives multiple
Advertisements.
The MAC Address of the FCF-MAC. (Just in case, for some reason, it is different
from the source MAC of the Ethernet frame.)
If the FCF-MAC uses FPMA, it will send the FC-MAP of the FCF
A name for the switch.
A name for the fabric.

27

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

FCoE Discovery and FC Login Process (cont.)


3. ENode sends FIP FLOGI
Just like FC FLOGI, only in a FIP frame
Returned FCID also used to build FPMA MAC
Address
For FCFs, FIP ELP sent instead

For NPIV, FIP FDISC is used


Works just like FC counterpart

4. The FCF responds with a FIP LS_ACC (or


ELP_SW_ACC)

28

17-Aug-08

2008 IBM Corporation

After the ENode has received the Discovery Advertisements, it selects a FCF-MAC
to use for that VN_Port; it then sends a FIP FLOGI. Just as with a FC FLOGI, the
purpose is to communicate fabric parameters, and to obtain a FCID. The frame
contains only two things: the ENode MAC address, and an ordinary FLOGI payload.
The returned FCID is used to build an FPMA, if used. The FCID is also, of course,
used in the FC payload of all of the FCoE frames.
If this is a FCF establishing a connection
With NPIV, a logged in port would send a FIP FDISC, just as it would send a FC
FDISC in a native FC network.
Lastly, the FCF responds to the FLOGI (or ELP) with the appropriate FIP ACC (or
RJT) response. Again, this works just as it does in native Fibre Channel.

28

New Enterprise Forum: Information Infrastructure - IBM System Storage and Storage Networking Technical Symposium

Summary


FCoE has two parts


FCoE frames simple Ethernet frame plus Fibre
Channel frame.
FIP protocol handles device discovery, fabric login
functions, and address assignment

29

17-Aug-08

2008 IBM Corporation

In summary, the simple FCoE stack has two parts: FCoE frames and the FIP
protocol.
FCoE frames are nothing more than Ethernet frames with an entire Fibre Channel
frame encapsulated inside
The FIP protocol handles device discovery (through Discovery Solicitations and
Advertisements), fabric login (through FIP FLOGIs, FDISCs, and ELPs) and
address assignment (through the use of FC-MAP and the FCID assigned to the
link.)

29

Das könnte Ihnen auch gefallen