8Gb Fibre Channel is now ratified by the FCIA: so what?
"If 4 Gb/s and 10 Gb/s Fibre Channel will be available this year, why bother with 8Gb/s?"
To answer this, it must first be clarified that when the FCIA ratified 8Gb/s Fibre Channel (aka "8GFC") for copper storage device interconnects they were talking about accepting 8Gb/s inside the disk enclosure and between storage arrays (disk or tape). In other words, the 8Gb/s Fibre Channel speeds between the disk drives on a Fibre Channel Arbitrated Loop (FC--AL) all within a storage disk array. 8GFC was added to the roadmap so that suppliers and users could map their future to a smooth migration from 4GFC. 8GFC has an initial availability roadmap timeframe of 2007/2008.
In a disk enclosure, the present speed is 2Gb/s Fibre Channel (2GFC). 4Gb/s Fibre Channel (4GFC) is on the horizon for later this year. A migration to 8Gb/s Fibre Channel is a logical step as it uses the same 8b/10b encoding (where 8-bit data bytes are encoded using 10-bit transmission characters) found in 2GFC and 4GFC. In addition, 8GFC is capable of auto-negotiating with lower speed 2GFC and 4GFC, stepping down to these speeds for operation. The backward compatibility of 8GFC is compelling because it assures users that their 2GFC and 4GFC investments will be protected and preserved going forward. 8GFC disks will be able to plug into 2GFC and 4GFC enclosures, thereby eliminating the need to replace everything around the disk, allowing users to gradually progress to the faster technology in the most economic way possible.
Unlike the SAN where 10Gb/s Fibre Channel (10GFC) is an option, disk manufacturers made a conscious decision not to have 10Gb/s on their roadmaps. The efforts of the working group that came to the 8GFC consensus looked hard at 10GFC as an option, but 10GFC uses an alternate encoding scheme and does not allow loop attachment--undesirable features for the extremely cost-sensitive disk drive market. The working group's marketing requirements for disk migration beyond 4GFC demanded full backward compatibility to 4GFC and 2GFC, and even went as far as requiring the use of the same connector as currently deployed on Fibre Channel disk drives. Multiple lane approaches, each running at slower speeds like 10G XAUI, ganged together to provide high overall speed, were not considered due to their inherent cost and complexity burden to the disk drive.
There were other obstacles encountered when considering 10GFC for disk drives. Since 10GFC uses a different encoding scheme than 2GFC and 4GFC, dual encoders would be required in order to maintain backward compatibility, adding cost and complexity to the disk drive. 10GFC would also require a different non-backward compatible, more expensive connector. In summary, 10GFC backward compatibility to 2GFC and 4GFC cost too much to make it backward compatible. After an exhaustive nine-month effort, the group settled on 8GFC as the most effective path to meet the cost-sensitive, backward compatible market requirements of Fibre Channel disk drives.
When designing RAID-based disk storage using 8GFC disks, the traditional Peripheral Component Interconnect (PCI) bus--used as a mezzanine bus in the RAID controller for attaching the disk drives--cannot inherently support 8GFC. In order to support 8GFC at the full data rate of 8Gb/s, PCI-X 2.0 or PCI Express will be needed.
If we look historically at why 2Gb/s was adopted so swiftly we come to understand why 4GFC will become a popular requirement for OEMs later this year; it is a matter of cost and ease of migration. Advances in manufacturing of silicon will allow manufacturers to offer 4Gb/s with twice the performance of 2Gb/s Fibre Channel at about the same cost--as was the case when 2GFC was offered in a 1GFC world. Additionally, just as 2GFC was fully backward compatible with 1GFC, 4GFC is backward compatible with 1GFC and 2GFC, making it unnecessary to do any major technology migrations. These same arguments apply to promote 8GFC.
Using 8GFC in disk enclosures also leads to better utilization of the 10GFC "fat pipes" used for Inter-Switch Links (ISL). The 8Gb/s of data bursting out of the RAID controller in the disk array will better utilize the 10 Gb/s fat pipes in the core of the SAN. It should be noted that 8GFC results in 8 * 1.063 Gb/s = 8.504 Gb/s of usable bandwidth.
8GFC will prevent bottle-necks in the point-to-point and loop topologies over copper inside the disk array before they can occur.
In addition, 8GFC will be used to interconnect disks with high-speed tape arrays before they connect to a SAN switch and become part of the fabric. The issue to be overcome is the fact that 8GFC copper has a distance limitation of 0.6 m. Loop switches can help extend the distance.
In the longer term, 8GFC could migrate from the back-plane and storage interconnect applications and into the SAN fabric, as did 4GFC. All this could place greater demand on today's 10Gb/s ISLs between switches, causing those connections to migrate to perhaps 40Gb/s and beyond. In 1965, Gordon Moore made his famous prediction regarding exponential increases in the number of transistors per IC chip every few years, so too can we expect to see fatter pipes at the edge forcing still fatter pipes in the core of the storage network. This prediction is not so far-fetched or futuristic if you consider that between 2001 and 2003 the adoption of 2GFC went from 10% to 90%.
When a technology offers twice the performance with no obvious cost or other penalties, it is a difficult proposition to pass up.
Ravi Prakash is outbound marketing manager at QLogic Corporation (Aliso Viejo, CA)
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Fibre Channel Industry Association|
|Publication:||Computer Technology Review|
|Date:||May 1, 2004|
|Previous Article:||iSCSI deployment in business IP storage network.|
|Next Article:||FCIA & FCIA-J span the Pacific.|