From the Keylabs report:
Overall Impressions (theirs not mine)
The Brocade switches outperformed the Ancor and Vixel switches in every area. Throughout the testing, the Brocade Silkworm 2400 switch was occasionally matched in performance and functionality, but was never exceeded by the other switches. This included the software tools used to configure and administer the switches. Perhaps the most valuable resource in any data center is the administrator's time. KeyLabs gives its highest praise to the Brocade switch for consistently allowing quick and easy configuration of its switches, every time. The flexibility, feature set, and design of the switch management tools demonstrated a remarkable commitment to addressing the administrator's needs in setting up and maintaining a SAN fabric. Whether a command-line power-user or a novice user needing a full GUI, the administrator was given a full range of access to the switch's features. To a remarkable degree, every facet of the switch seemed designed for convenience and flexibility in deploying and maintaining it within an enterprise-wide Fibre Channel fabric. This switch is far-and-away the most enterprise-ready and administrator-friendly of the three switches. The Ancor switch has dynamic port configuration, making set up fairly easy. However, the switch has a port connectivity limitation?when an inter-switch link (ISL) is attached to a port on the switch, two of the ports on the switch are dynamically assigned the T-port type. This assignment of T-port types by pairs could be accommodated in most cases, yet it always had to be considered. Some switch features were dependent on having the correct firmware loaded. Having to change firmware between multistage and single-stage to obtain the desired feature also had to be considered during switch setup and configuration. Utilizing full speed of the port in a full-duplex traffic pattern appeared to be dependent on the port used to connect the switches--not all T-ports would utilize the full bandwidth. The Vixel switch constantly required manual port configuration. Having to set up every port every time for every change was at best inconvenient. And having to take the switch off-line to either flash load or update memory for every port change seemed inconsistent with the fundamental purpose of a switch?to facilitate flexible and continuous connectivity across the fabric. However, once configured and running, the switch performance was generally solid.
Availability
The Ancor switch exhibited limited fail over capability. It could not be configured to recognize available data paths through inter-switch links (ISLs) more than one hop away without restricting every port on the switch to a single port type by setting the switch to "cross-connect" mode. The Ancor documentation defines its fail over scheme to be redundant links between two switches, with only one link active at a time. The Brocade switch was extremely flexible as to how it could be configured for data path redundancy. Throughout the testing, KeyLabs experienced no restrictions as to hops or loops in inter-switch connectivity; ports on any switch could be chosen arbitrarily. Unlike the Vixel and Ancor switches that could utilize only one inter-switch link (ISL) between switches, the Brocade switch used every available data path and would immediately fail over to any active link when a single ISL was failed. It also dynamically recognized a restored link or an added link, so that subsequent ISL failures would immediately fail over to any active link. The Vixel switch was quite flexible as to how it could be configured, even if the set up process was interruptive and somewhat convoluted. Its fail over capabilities included the ability to recognize and use data paths through a second switch. However, under heavy load the Vixel switch took a significant amount of time to fail over, if it failed over at all. After fail over of a single ISL, the switch was sometimes not able to dynamically recognize that the failed ISL had been restored so that it could handle the failure of a second link.
Scalability The Ancor switch exhibited only limited scalability. All port configurations were dynamic, so there was minimal loss of connectivity when configuring changes in the fabric. However, two areas of functionality limited this switch:
A. port connectivity speed ? there was often no increase in throughput of an ISL when going from half-duplex to full-duplex traffic patterns. KeyLabs discovered that this behavior was port dependent. Some ports did allow full-duplex traffic patterns, others did not. This behavior was not explained or documented in the Ancor product literature.
B. inter-switch throughput ? the switch utilized only one ISL between any two switches. Adding redundant ISLs did not increase aggregate throughput between the switches. In a full-duplex traffic pattern, aggregate throughput for the switch was consistently half of the other switches unless specific ports were chosen.
This switch also performed rather poorly over long fiber connections. Over a 10 kilometer ISL, the switch throughput was only 16-17 K-frames per second when running under full load.
The Brocade switch was by far the most scalable switch, exceeding the other switches in capability and functionality. Every port is independently available for connectivity, and each is capable of achieving maximum full-duplex connectivity no matter what the port or the port type. Even with a 10-kilometer fiber cable between switches, the full-duplex throughput on the port was 95% of theoretical maximum (throughput at 95MB/sec). The switch automatically configured itself to utilize all active ISLs to increase aggregate throughput between switches. Each ISL could move frames at maximum, full-duplex speed. All switch configurations were dynamic, being done on-the-fly without loss of connectivity. Scaling a Fibre Channel network using this switch would be convenient and effective.
The Vixel switch was quite flexible as to how its ports could be configured, but the lack of dynamic port configuration places a very real limitation on its scalability. It simply could not detect any device or connection without first running the management utility and then resetting the switch to flash load or memory load the new configuration. The switch did support all of the topologies defined in this comparison. The throughput of the switch was excellent, as long the data stream did not go through an ISL. In every case where the connectivity included a full duplex traffic pattern across an ISL, the overall throughput on the port stayed at half the capacity of the fiber cable. In addition, the switch could not utilize more than one active connection between switches. Adding redundant links between switches did not increase aggregate throughput, but merely facilitated fail over. KeyLabs would not recommend that the switch be used where a heavy load was required over a long fiber cable; the switch throughput on a 10Km cable at full load was extremely poor?8-9 K-frames per second.
Migration Support
On the Ancor switches, KeyLabs could not find any way to connect a private-loop device. The switches never allowed private-loop connectivity through the switch. The switch did allow IP over Fibre Channel. The Brocade switches exhibited no problems in allowing fabric-aware devices and private-loop devices to coexist on the same switch or within the fabric. The switch facilitated dynamic connectivity among all devices whether fabric aware or private loop. The switch also supports IP over Fibre Channel. Throughout this entire test, any port on any switch could be used for any connection at the administrator's convenience. The Vixel switches could connect either private-loop devices or fabric-aware devices, but not both simultaneously. For this test suite, the switches could not be configured to allow connectivity among both device types on the same switch or switch set. The switches did not support IP over Fibre Channel.
Manageability
Hardware zoning on the Ancor switches was inconvenient because it required single-stage firmware. This meant having to give up the capabilities of the multistage firmware in order to have hardware zoning, so hardware zoning was limited to a single switch. Broadcast zoning was supported on these switches and was easy to implement. The Brocade switches exhibited much more flexible hardware zoning capabilities than the other switches, supporting arbitrary hardware zones across the entire fabric. Hardware zones, including overlapping hardware zones, could be set up across the fabric by switch/port assignments. One very nice feature that was exclusive to the Brocade switch was its capacity to create a software zone based on World-Wide Name (WWN). This allowed individual devices within a single loop to be separately included in different zones. The most convenient feature about zones based on WWN was that every device in the zone was included by its name, so wherever it was moved within the fabric, the zone remained intact. This meant that the device remained within the defined zone even when it logged in to th e fabric from a different port from where it was last connected. The WWN zone did not need to be reconfigured when any of its devices moved from one port to another within the fabric. The broadcast zoning on the Brocade switch was flexible, allowing the broadcast zone to include ports throughout the fabric. The Vixel switches allowed hardware zones only within a single switch. No cross-fabric hardware zones were possible. The switches did not support either software zoning or broadcast zoning.
Usability
The Ancor switch offered an adequate set of management tools for all switch features. The switch statistics were solid. However, the experience of the KeyLabs administrators in running the utilities was that convenient usability features were lacking. The Web-based GUI had somewhat different functionality from the Windows-based utility, in that the Web-based utility offered views not offered in the Windows-based utility. Initial switch setup and firmware upgrades were not convenient. There was no way to easily tell which switch was matched to which management window open on the screen. The world-wide name was too convoluted to be useful for quick identification on the screen. Changes made to any switch did not appear in the fields on the screen until the screen was manually refreshed. For these management tools, the switch features were well covered, but some conveniences had been overlooked.
In KeyLabs' opinion, the management tools for the Brocade switch were built with the administrator's convenience in mind. The Brocade switch had the broadest and most accessible set of management tools of the three switches tested. The switch supported multiple management modes (Graphical User Interface and Command Line) that could be used from almost any platform. The switch allowed administrative connectivity through serial connection, Ethernet, and IP over Fibre Channel. In addition, these tools were more flexible and "administrator-friendly" than the management tools available for the other switches. For an administrator who needs quick and easy enterprise-wide zoning or management tools that can be configured without interrupting connectivity or who requires complete switch statistics and a system log of events, the Brocade switch was the only one to offer all of these in its management utilities.
The Vixel switch also had adequate means for managing its switches. Task-oriented help was lacking in both the GUI interface and in the documentation. Error messaging was inadequate for letting the administrator know when an incorrect entry had been made. Unique ID numbers and other addresses could be entered incorrectly and accepted without error. To KeyLabs, the most glaring usability issue was that every change to every port required going into the management tool, setting the switch manually, and resetting the switch in order for the change to take effect. This not only took an inordinate amount of the administrator's time; it interrupted connectivity as well.
Trademark Notice
All trademarks, service marks and trade names used in this Report are the trademarks, service marks and trade names of their respective owners, and no endorsement of, sponsorship of, affiliation with, or involvement in, any of the testing, this Report or KeyLabs should be implied.
|
