Universal Serial Bus (USB) is starting to live up to its name. It’s a key component to legacy-free PCs that do away with the parallel, serial, and PS/2 ports found on most of today’s PCs. One USB connector gives a PC access to multiple devices with significantly improved performance, device interaction, and reliability compared to older interfaces. It can reduce the number of connections on most PCs to three: USB, video, and Ethernet.
The path to this point has not been easy. The first USB standard ran at a pokey 12 Mbits/s. It delivered hot-swap capability, power to devices, and a reliable, bi-directional communications link between host and peripheral. USB 1.0 was great compared to legacy interfaces but insufficient to meet the demands of multiple high-speed peripherals like DVD drives. At 480 Mbits/s, USB 2.0 meets those demands.
USB 2.0 has also arrived as driver quality and interface intellectual property (IP) has improved. Standard-class drivers are now part of the package developers can choose from when dealing with USB. Human interface, storage, and other standard-class drivers are sufficient for most PC peripherals and for most embedded applications. This significantly improves interoperability, a problem with USB 1.0.
Today, it’s rare to find a new device that’s not USB-based, except for wireless or 1394-based peripherals. Printers, scanners, mice, and keyboards are being delivered in USB-only versions. This greatly simplifies use and reduces tech support requirements because USB devices integrate better with PC device drivers.
The USB 2.0 On-The-Go (OTG) standard appears to be making steady inroads in the embedded space, permitting intelligent peripherals to co-exist with PCs and peripherals. OTG allows a USB device to act as either a host or client. A USB OTG digital camera can connect to a PC or drive a USB peripheral. But OTG is still in its infancy. For example, the ability of a USB OTG digital camera to connect to a USB printer directly will vary depending upon both sides of the connection. In many instances, the results will be disappointing. This will improve over time, but for now the best results will be between a set of well-defined devices.
USB is showing up in more multimedia devices like DV camcorders, which were once the exclusive realm of the IEEE-1394 interface (also known as FireWire and iLink). The original specification, 1394a, has been upgraded in speed and functionality to 1394b, with a minimum speed of 800 Mbits/s an impressive maximum of 3.2 Gbits/s. FireWire could potentially be the link between HDTV and the PC, in addition to providing more robust connections to multimedia peripherals compared to USB 2.0. With significantly improved networking capabilities, 1394b competes with Ethernet in limited environments, such as the desktop or the home. It uses standard Cat 5 cable to operate up to 100 m, but can use other media, like optical fiber, over greater distances. Actual delivery of products has yet to reach more than a trickle, however.
USB and 1394 are on the outside of the box. Similar things are happening on the inside. Serial ATA is starting to deliver a simpler four-wire interface compared to the wide ribbon cables used for parallel ATA and floppy disk interfaces. Unlike USB or 1394, Serial ATA is a point-to-point connection, so multiple Serial ATA ports are necessary to handle more than one drive.
Serial ATA 1.0 operates at 150 Mbits/s. It is comparable to parallel ATA implementations, but its reduced cable size and point-to-point architecture appeal to companies delivering PCs. Serial ATA cuts down on cable clutter, provides more reliable connections, and eases tech support problems. Although parallel ATA can continue its performance climb, it does so at greater cost. Once Serial ATA drives are readily available, the move away from parallel ATA drives will significantly accelerate.
The CD-RW, DVD-RW, and DVD+RW drives will also migrate to Serial ATA. This will lead to another system reduction, the elimination of the floppy drive. The last bastion of hope for the lowly floppy is its ability to provide low-cost data interchange. Unfortunately, its capacity and overhead doom it to extinction.
Tomorrow’s system may be here today. It will have a pair of Serial ATA connections, a USB, Ethernet, and a video port. Use a low-power processor and the entire PC fits into less space than the DVD drive attached to it. This definitely opens a lot of possibilities.
Switch blades: Are 1U servers going to disappear like the legacy PC? Hardly. While blade servers tackle the high end, racks of 1U servers still make sense in many environments. Expect the two to coexist for some time while blade uses continue to grow.
Board-based solutions are breaking out into four categories. VMEbus and CompactPCI make up the pre-existing solutions, with switch fabrics being added to the specifications. AdvancedTCA takes the high road, but custom designs still abound.
For CompactPCI, the PICMG 2.16 standard leads the way. Although CompactPCI was hard hit by the drop in the communications market, it’s actually growing in popularity in other realms. Its small size and new standards provide a way to build high-reliability systems in a small package.
Meanwhile, VMEbus is the bus that never dies. It was a system that some thought would disappear, but it is at the other end of the spectrum now with lots of life. Adding switch fabric support seems to be a pleasant pastime, with many off-the-shelf options available from Gigabit Ethernet to RapidIO. In fact, for flexibility and a wide range of choices, VMEbus is second to none. The list of conduction cooled products is extensive and the mezzanine card standards make single-board computer enhancement a snap.
VMEbus will run into AdvancedTCA, but AdvancedTCA’s large board size, communications orientation, and wide support will make a large niche for this form factor. The standard shelf-management system is key to the high-reliability support that’s part of a high-availability design. The confusing aspect of AdvancedTCA is the plethora of switch fabrics it can support. A single implementation will almost always use one of the options, but choosing from Gigabit Ethernet, PCI Express Advanced Switching, InfiniBand, and StarFabric can be a bit daunting.
InfiniBand and StarFabric are doing well in the blade market. The technology is proven and offers performance advantages over Ethernet. They will eventually compete with PCI Express Advanced Switching, but having a two-year lead offers significant advantages. InfiniBand’s low overhead, high performance, and robust management will eventually win over blade server managers, especially where server farms are growing. Low overhead will make a significant impact on InfiniBand’s applicability compared with iSCSI and Ethernet.
StarFabric appears to be going for the title of PCI Express Advanced Switching-Lite. This is actually a good thing for a number of reasons. First, PCI Express is just showing itself while StarFabric has a track record. Second, there’s a clearly defined upgrade path. Finally, StarFabric fills a gap at the low end where PCI Express Advanced Switching is less competitive. Its incorporation into such standards as the CompactPCI PICMG 2.17 makes StarFabric worth watching, especially in the blade space where flexibility is often more important than raw power.
Finally, there are the custom blade servers. Be prepared to buy everything from one vendor. The architectures address deficiencies in standard solutions like CompactPCI. For example, board frames that are more robust make hot swapping easier and more reliable. Compact designs lead to ultra-dense systems. The advantage of a coordinated design is a system that can use all of the component’s features, such as low-power processors that work with conduction and convection cooling. Boards can be the proper height to accept standard components, such as hard disks and RAM. A standard may emerge from this fray, but don’t count on it happening soon.
The blade server approach offers benefits with respect to storage. In fact, the switch fabric employed by most blade-server systems is designed for three main purposes: communication with the outside world, other processor blades, and local peripherals. The latter can be further refined to storage.
InfiniBand is competing against Ethernet and iSCSI and, to some degree, FibreChannel. Each has its advantages. InfiniBand has its efficient architecture. Ethernet has availability of a familiar product. FibreChannel has a long and effective track record.
Behind these interfaces resides the disk subsystems. These are often FibreChannel or SCSI peripherals. Parallel SCSI drives have been the mainstay for years but will start to change as Serial attached SCSI (SAS) comes into play. SAS is a superset of Serial ATA 2.0, and the two use the same physical interface. SAS brings a more powerful command set needed in server and blade server environments.
The blade server approach has proven to be solid system design. Its popularity continues to grow, bringing along improved storage and interconnect technology. It will be interesting to see if the approach remains a high-end solution or migrates down to the midrange server market.