Cisco Cisco MDS 9000 4 44-Port 8-Gbps Host-Optimized FC Switching Module Libro bianco
Silverton
Consulting, Inc.
StorInt™ Briefing
Fibre channel’s next generation has arrived and now supports 8-gigabit per
second (Gb/s) transfer rates over one fibre channel (8GFC) link, effectively
doubling available bandwidth to 840 megabytes per second (MB/s). However,
8GFC technology is priced at a significant premium over 4GFC and depending on
the configuration, deployment of 8GFC may be up to 2X the cost of using 4GFC.
In contrast, one appealing factor of 8GFC links is its downward compatibility
with 4GFC and 2GFC links thus allowing a gradual move to this new technology.
Increased bandwidth capability, the large price differential, and the downward
compatibility should be the critical factors in determining an imminent proposed
move to 8GFC. Aside from these, other less objective factors must also influence
the ultimate 4GFC versus 8GFC decision. Indeed, since 4GFC technology was
introduced in 2004, major enhancements in storage area networking, server,
application, and storage subsystem performance have been made, making the
switch to 8GFC even more complicated.
8GFC and storage area network switching
Storage area networking, particularly switch technology, has seen numerous
changes in recent years. Specifically, switches have necessarily become more
complex to match increasing fabric functionality requirements. In addition, layers
of switches and the number of ports have been proliferating to support more
complex fabrics. Today, core and core to edge switching uses 4GFC, 10GFC or
inter-chassis copper channels for inter-switch links (ISLs). One area particularly
suited to 8GFC technology are the ISLs such as between the core networking
and/or edge storage switch components thereby reducing port counts and cabling
by at least a factor of two.
Additionally, 10GFC ISLs could conceivably be economically replaced with a
more general purpose, 8GFC solution. SAN extensions over metro area networks
use 10GFC but contrary to the ISL use above, 8GFC metro area networks with FC
clocking cost significantly more than 10GFC with Ethernet clocking and thus
continues to justify the status quo.
changes in recent years. Specifically, switches have necessarily become more
complex to match increasing fabric functionality requirements. In addition, layers
of switches and the number of ports have been proliferating to support more
complex fabrics. Today, core and core to edge switching uses 4GFC, 10GFC or
inter-chassis copper channels for inter-switch links (ISLs). One area particularly
suited to 8GFC technology are the ISLs such as between the core networking
and/or edge storage switch components thereby reducing port counts and cabling
by at least a factor of two.
Additionally, 10GFC ISLs could conceivably be economically replaced with a
more general purpose, 8GFC solution. SAN extensions over metro area networks
use 10GFC but contrary to the ISL use above, 8GFC metro area networks with FC
clocking cost significantly more than 10GFC with Ethernet clocking and thus
continues to justify the status quo.
8GFC and servers
Server configurations have experienced several performance-enhancing changes
in recent years. One significant change has been the emergence of blade servers
to manage multiple commodity servers running similar, high-intensity workloads.
With blade servers, separate infrastructure hardware is now shared across a
number of server blades within a single cabinet, thereby significantly reducing
necessary equipment and corresponding power requirements. Even so, the
addition of 8GFC to these blade servers may be of questionable benefit depending
on bandwidth requirements to meet the overall I/O demand.
in recent years. One significant change has been the emergence of blade servers
to manage multiple commodity servers running similar, high-intensity workloads.
With blade servers, separate infrastructure hardware is now shared across a
number of server blades within a single cabinet, thereby significantly reducing
necessary equipment and corresponding power requirements. Even so, the
addition of 8GFC to these blade servers may be of questionable benefit depending
on bandwidth requirements to meet the overall I/O demand.