Video applications rely on Fibre Channel RAID
When the film industry started digitizing special effects, Fibre Channel became popular, primarily because of its 100MBps (200MBps in full duplex mode) transmission speeds. In fact, today, the combined use of Fibre Channel for real-time speeds and RAID arrays for data reliability has become almost a necessity.
By Dennis Pederson
In the entertainment market and others, Fibre Channel`s feature set is particularly advantageous. A primary advantage is its ability to be used as both a channel and a networking protocol. In comparison to other interfaces, Fibre Channel also has some significant benefits, including:
- Copper or optical physical connections.
- Cable lengths up to 30 meters with copper, or 10 kilometers with fiber optic.
- Up to 126 nodes on a single loop.
- Hot-swapping of drives and dual porting.
- Multiple topologies, such as point to point, arbitrated loop, and fabric.
- Multiple protocols on the same cable, including HIPPI, TCP/IP, FDDI, ATM, and SCSI.
Fibre Channel can be used for channel and network applications because it supports different topologies. Point to point, for example, is dedicated to transferring data between two devices, such as a disk subsystem and a computer.
Fibre Channel can also be configured in fabric networks. Unlike other networking topologies where devices must compete for network bandwidth, the fabric topology provides virtually unlimited bandwidth. Fabrics include switching functions, which connect any two arbitrary devices on the fabric. The total available bandwidth is limited only by the switching capacity of the fabric.
The arbitrated loop topology supports both the channel and the network. Up to 126 devices can coexist on a Fibre Channel loop, and any device can move data. This feature essentially eliminates the restrictions of point-to-point configurations without requiring network switches. Most implementations are traditional channel applications, where Fibre Channel is used to interconnect disks and RAID subsystems to workstations and servers.
Fibre Channel can handle almost any visual computing storage requirements. However, for applications in film and High Definition Television (HDTV) production--for which 100MBps bandwidth is still not enough--multiple Fibre Channel implementations combined with high-performance RAID can be a cost-effective solution.
The bandwidth requirements placed on a visual computing system are determined by the amount of digitized visual data required to represent a film or video image. The playback of uncompressed National Television Standards Committee (NTSC) format video (640 x 480 resolution) with 24 bits of color per pixel requires a data rate of more than 27MBps. Uncompressed high-resolution playback of film (2048 x 1536) or HDTV (1920 x 1080) requires more than 280MBps and 186MBps, respectively.
The data storage volume required for a video that is compressed to 3MBps is more than 22MB for one minute of playback. Using a 9.1GB disk, approximately 400 minutes are available, enough for about three movies. If the application requires more titles streaming from the server, more disks and Fibre Channel bandwidth is needed. In addition, storage redundancy is needed to prevent data loss. RAID architectures address this requirement.
Depending on the video out quality requirements, compression ratios of 4:1 or more can reduce the data rate for a single stream of video for these high resolutions. However, applications such as video-on-demand, commercial insertion, and broadcast station automation create the need to deliver hundreds of streams and very high data rates.
Fibre Channel`s bandwidth is also important for video server applications used to deliver digitized video data to a communication delivery system in real-time. Upon request, or on a pre-programmed schedule, the video server must retrieve video from on-line storage and forward it through system buffers for video output and transmission. This process must be performed in real-time to ensure the video is delivered at a continuous rate of 30 frames per second.
The objectives for video storage are straightforward. The storage must be cost effective, which means that the subsystem needs to deliver the highest stream-to-disk-spindle ratio possible. Additionally, there must be adequate capacity for the total content to be available on-line. And storage must be fault tolerant.
There are six different RAID levels or architectures, though RAID0,1,3, and 5 are most common. RAID3 and 5 are most often used in visual computing applications. RAID0, or JBOD (just a bunch of disks), is also used in visual computing applications, but it offers no data protection if a disk fails.
RAID1, or disk mirroring, is the simplest RAID implementation. For each data drive, there is a redundant, or mirror, drive. RAID1 is highly reliable and easy to implement. However, because every disk is duplicated, twice the total number of drives is needed, so it is the most expensive solution.
In comparison, RAID3 and 5 stripe data across all the drives in the array. They typically offer the best cost/performance/reliability ratio for visual computing. They have different levels of performance, but they have similar costs and share at least two common performance features. Any single drive in a RAID configuration may fail and all the data stored in the array remains accessible. An added drive is also used to provide redundancy, typically counting for about 10% of the total array cost.
RAID3 provides the highest sustained transfer rate for large, sequential files in visual computing applications, and there is no performance degradation after a drive has been removed from the array.
The added drive in RAID3 is dedicated to redundancy and cannot contribute to data transfer, but because of parallel striping techniques, the sustained transfer rate approaches the media limits of the drives.
In other words, in an 8 + 1 configuration, a sustained transfer rate is approximately equal to 8 times the sustained transfer rate of an individual drive.
RAID5 performs well when the request sizes are small and random. On the downside, the performance of some RAID5 implementations degrade by as much as 50% after a drive fails. When a RAID5 drive fails, the remaining drives are accessed every time the failed drive makes a request, preventing the functional drives from servicing their own requests.
RAID3 and RAID5 use the added drive differently to provide redundancy. RAID5 can use the added drive for data transfer. The drive distributes the data and parity information evenly across all drives. Theoretically, RAID5 can support "n" drives worth of streams; however, any request distribution pattern that doesn`t keep all the drives busy degrades performance.
RAID5 is better suited for small file sizes that occur in random, and they provide higher performance when all drives are operating. RAID3, however, is typically better for the large, sequential files in visual computing. The key difference in these two RAID architectures is the performance after drive failure: RAID3 doesn`t experience any performance degradation, while RAID5`s performance may drop as much as 50%.
Fault tolerance is one more measure of ensuring data protection. Redundant power supplies and cooling fans--two of the most likely components to fail--ensure reliability. "Hot swap" is the ability to remove and replace a failed component without shutting down the subsystem.
In summary, Fibre Channel provides higher bandwidth to meet the application needs in the entertainment market. It also increases the importance of RAID architectures and fault-tolerant storage characteristics to protect the content to be shared and stored.
Dennis Pederson is marketing manager, entertainment, at Ciprico, in Plymouth, MN.
By 2002, 45% of all embedded RAID controllers will ship with Fibre Channel.