White Papers

BP1028 EqualLogic iSCSI for Fibre Channel Professionals 6
Some Ethernet switches have stacking abilities, where a dedicated (usually proprietary) interface is
connected between two or more switches, allowing them to be managed as a single switch and also
to forward traffic as if it was a single switch. Non-stackable switches can be connected by configuring
standard ports or higher-speed uplink ports as a Link Aggregation Group (LAG) – although sometimes
these are also referred to as ISL’s. Some switches also offer virtual-chassis features to allow linking
several switches together to operate as one. Of course high-end chassis based switches use a high-
speed switching backplane to connect line-cards or blade modules together, and allow expansion by
adding more cards or modules to the chassis.
When links are created between Ethernet switches, whether it be a LAG or a single port, careful
attention must be given to Spanning-Tree configuration. When multiple links are detected, Spanning-
Tree Protocol (STP) will block or disable one or more links to avoid a loop (where a switch has more
than one path to another switch). This design can sometimes cause optimal links to be blocked and
even reduce the bandwidth available between switches. Assigning cost to the links allows the network
administrator to determine which link or links are blocked by default. Configuring a LAG allows two or
more links to be treated as a single link by STP so that the aggregate bandwidth is achieved.
The FC protocol is also confined for local use. That is, it is not capable of routing or travelling over a
WAN as is. For FC networks to travel longer (than the maximum of 10 Km) distances, either specialized
extenders must be used, or the FC packets must go through a routing or bridging device. FC over IP
(FCIP) is another example of how FC is encapsulated in TCP/IP so that it can be used over longer
distances.
Since iSCSI natively travels over Ethernet and TCP/IP networks, it does not require specialized
hardware; however careful consideration must be placed on architecting any long-distance SAN
connection. For example, connecting two sites together for iSCSI (or FC over IP) replication may
require significantly more bandwidth than had been used previously by other non-storage WAN
activity. Also, network connections that are used for long-distance replication should be selected with
latency as low as possible and that higher latency WAN traffic is isolated from the SAN through
methods like VLANs, tunnels, firewalls or routing filters.