Directors are designed to connect hundreds of servers and storage arrays. But even with the industry's highest full-speed 8 and 4 Gbit/sec port density, many customers need to connect thousands of devices. While any Fibre Channel switch, director or backbone can be connected using inter-switch links (ISLs), not all products can "trunk" the individual ISLs into a single logical link. Trunking provides better reliability, throughput and availability of data travelling between directors. Much like a RAID array delivers better performance by striping data across different disks, trunking delivers better performance by "striping" data over different ISLs. Trunking should occur automatically and require little to no configuration — SAN administrators should be able to "set it and forget it" and know that their trunking is delivering maximum performance. If high-bandwidth connections are required, these should be the largest possible links in order to simplify cabling.
| Function | Brocade | Cisco
| Inter-Chassis Links
| Yes | DCX: 4 x 128 Gb ICLs (528 Gbit/sec) No
| Frame-based trunking
| Yes | DCX and 48K - up to 8 x 8 Gbit/sec ISLs (64Gb) No
| Exchange-based trunking
| up to 64K frames per sequence up to 1K sequences per exchange Yes | DPS, Standard Yes | Port Channels, licensed FSPF re-routing | & load balancing Yes | 48000 - standard M6140, Mi10000 - licensed Yes | standard |
|---|
How does Brocade compare?
Brocade continues to be the only company with frame-based trunking. The Brocade 48000 can provide multiple 32 Gbit/sec frame-level trunks built out of 8 x 4 Gbit/sec links, and the Brocade DCX can provide multiple 64 Gbit/sec frame-level trunks out of 8 x 8 Gbit/sec links. Another enhancement is "masterless trunking." Masterless trunking has been available since 2004 when Brocade Fabric OS 4.4 was released. Brocade trunks are self-forming and self-healing requiring no configuration. Brocade's common hardware and software product line means trunking works between different generation switches and directors. The Brocade DCX can trunk to Brocade 48000s, and both the DCX and 48000 can negotiate trunks with older 2 Gbit/sec switches and directors.
Two Brocade DCXs can be connected using Inter-Chassis Links, or ICLs. These ICLs are located on the Core Blades and are each 128 Gbit/sec wide. They can be trunked into a single 512 Gbit/sec connection, allowing more ports to be dedicated to hosts and storage instead of ISLs. The ICLs greatly simplify cable management -- they are equivalent to 128 x 4 Gbit/sec ISLs, or 64 x 8 Gbit/sec ISLs on each DCX. Assuming all ports are at 8 Gbit/sec, these ICLs are worst-case 6:1 oversubscribed, and at 4Gb they are just 3:1 oversubscribed.
In addition to frame-based trunking, Dynamic Path Selection (DPS) allows for exchange-level balancing between trunk groups. When connecting Brocade DCXs and/or 48000s together, 8-way trunking and DPS between trunks automatically takes place without any need for user configuration, providing up to 256 Gbit/sec balanced trunks for the 48000 and 512 Gbit/sec balanced trunks in the DCX. These features provide immediate benefits to network performance even in the absence of 8 Gbit/sec devices, and DPS in particular can operate even when connecting to installed 2 Gbit/sec switches. DPS also allows ISLs of considerably different lengths to be balanced, an advantage in metro ring topologies. DPS is a free, standard feature in Brocade Fabric OS.
As of FOS 6.0, Brocade also supports "F_port Trunking" to switches embedded in bladed server chassis running in Brocade Access Gateway mode. This simplifies bladed server deployment while still allowing wide logical connections that allow for greater availability.
Cisco cannot build frame-based trunks. Cisco PortChannels are based on exchange-based balancing*, similar to Brocade DPS, and can balance up to 16 connections. However, a Fibre Channel exchange can be considerably larger than a Fibre Channel frame. (An exchange is analogous to a chapter in a book, while a frame is equivalent to a word.)* Brocade's unique frame-based trunks are more granular and better balanced than any exchange-based balancing technology. DPS also comes free with Brocade Fabric OS.
Without ICLs, Cisco is forced to ISL using ports that could otherwise be used for connecting hosts and storage. If Cisco were to try to build a PortChannel with ISLs between 9513s as large as a Brocade ICL, only 4 free ports would remain on each 9513.
While Cisco's NPV mode is somewhat similar to Brocade Access Gateway, it is not able to trunk ISLs.