 connectivityZONE Products for the week of April 2, 2007
IDT Says…
PCI Express Interconnect Devices Manage All Communications In High Density Bladed Systems
Up to 64 lanes and 16 ports of predictable throughput supports flexible system resource sharing
IDT has announced the industry's first PCI Express-based (PCIe) system interconnect solution to predictably and efficiently manage all communications throughout the system while providing full-line rate throughput, reliability and high quality of service. The new devices are optimized to meet the capacity, scalability and predictable throughput requirements demanded by bladed system providers for consistent high performance and flexible system resource sharing. This robust solution - which includes the industry's first 64 lane and 16 port device - enables equipment providers to use PCIe as the primary high performance system interconnect for demanding server, storage, communications and embedded applications.
The seven new PCIe switches broaden the already industry-leading IDT family of PCIe system solutions to a total of seventeen. Three of the new devices are PCIe-based system interconnect switches that manage communications between roots and intelligent endpoints throughout a bladed system. Ranging from an 8-port, 32-lane configuration on the low end to the industry's largest monolithic PCIe switch with 16 ports and 64 lanes, the IDT switches offer unprecedented levels of switching capacity and system scalability and provide predictable, full-line rate performance for multiple simultaneous traffic flows regardless of system loading. The remaining four devices are new inter-domain PCI Express switches. This new class of switching solutions consists of both two and three port devices that support address translation and non-transparent bridging between PCI Express domains.
The IDT system interconnect solution surpasses simple peer-to-peer switching by offering robust support for communication between multiple roots and providing constructs for I/O virtualization and sharing that leverage today's PCI Express processors and endpoints.
Like other established IDT PCIe solutions, these devices have been optimized for low power. In fact, this family offers the lowest power consumption per port of any PCIe system switching solution available in the market today, enabling scalable and increasingly dense bladed system architectures and reducing the total cost of ownership by significantly minimizing system thermal management requirements.
"We have worked closely with top-tier system OEMs to ensure our interconnect architecture supports their increasing needs for scalability and flexibility," said Mario Montana, vice president and general manager of the IDT Serial Switching Division. "Our 64 lane/16 port device - the largest device available today - is already enjoying great traction with our customers. We are pleased to offer them a system interconnect solution with performance and advanced functionality that far exceeds the modest capabilities of the peer-to-peer switches commonly used today."
Development Tools and Customer Support
All of the new switches have a dedicated evaluation and development kit for device testing and analysis, and system emulation. Each kit consists of a hardware development board with representative upstream and downstream connectivity, and an IDT-developed, GUI-based software environment that enables the designer to tune system and device configurations to meet system requirements. Moreover, to ensure that each OEM system design is optimized for production and meets its time to market objectives, IDT provides customers with extensive, collaborative technical support, including system modeling and signal integrity analyses, and schematic and layout review services.
EN-Genius Says . . .
Although the PCI Express protocol has some inherent shortcomings left over from its origins as a motherboard interconnect technology, IDT is one of several companies hard at work making it easy to extend it into the world of blades and backplanes. This concept of dis-aggregating I/O by moving each type of connection onto a separate blade is very attractive because it enables quick cost-effective configuration and upgrades. IDT is to be congratulated on its new family of switches whose mix of ports and features is specifically designed for these kinds of systems. And while not everyone will be building multi-processor systems, it's good to know that part of the family includes PCIe inter-domain switches whose address translation capabilities allow for easy peer-to-peer inter-processor operations.
 IDT's switch family has been designed to move PCIe from desktop environments to carrier/enterprise-grade applications. This starts with the ability to maintain wire-speed transactions regardless of system loading for guaranteed deterministic performance in any non-blocking situation. The switches accomplish this with physically separate input buffers and output queues connected by a wide, over-provisioned, shared memory. Although it's hard to tell from the crude block diagram supplied by IDT (see Fig. 1), each port has its own dedicated scheduler that supports two virtual channels and eight traffic classes. Scheduling can be based on a combination of virtual channel and ingress port address. You can program each scheduler with its own selectable scheme (RR, WRR, and fixed priority) address enabling you to tune the switch's response to your application.
In keeping with the hi-rel philosophy, the switches are also equipped with failover capabilities that make it easy to implement systems with redundant hosts. Advanced systems also require resource sharing and virtualization as well as dynamic resource mapping that can allocate I/O resources to any compute element at will.
 The first members of the family include three system interconnect switches and four inter-domain switches. The inter-domain devices are especially interesting because they provide the address translation capabilities required to get around the PCIe protocol's inherent inability to support peer-to-peer operations. Just like earlier bridging products, the switches' low-latency look-up tables do non-transparent bridging that lets each processor think it's the master, even though one of them might be an end node on a system (see Fig 2). Cleanly segmenting PCIe's master/slave domains enables you to easily map an external processor into your compute domain. It also lets you easily share I/O between two processor blades, making so-called any-blade-any-slot designs possible.
In addition to its silicon, IDT has gone out of its way to give you the software tools you need to develop blade-based systems including support for hot-plug protocols and drivers to support discovery processes as well as system and domain address initialization. You also get drivers to support I/O sharing of FibreChannel-based peripherals over PCIe. And while they made no promises, IDT expects to provide drivers to support SAS/SATA storage devices as soon as the market expands enough to justify the development effort.
The lower costs and enhanced capabilities that IDT's new PCIe switches deliver should also help significantly cut the BOM and development costs of blade-based ATCA/uATCA products. While blade-based systems will always cost a bit more than the equivalent stackable pizza boxes that have dominated the market for the past few years, narrowing the price gap should enable small enterprises and SMBs to enjoy the same flexibility and scalability that's previously been reserved for enterprise-class systems.
By adding multi-processor capabilities support for in-band inter-processor communications, and extremely consistent low-latency connections, IDT's switches allow PCIe to realize its potential to deliver many of the advantages that Intel's late, unlamented Advanced Switching (AS) technology attempted to deliver -- but without the torturous overlays and bloated encapsulation overhead that contributed to its demise.
All of the new devices are currently sampling to qualified customers. PCIe interconnect switches are in BGAs with 900 balls for 32 lanes/8 ports, 1156 balls for both 48 lanes/12 ports and 64 lanes/16 ports with volume pricing at $85, $120 and $160 respectively. PCIe inter-domain switches are in BGA ranging from 324 balls to 484 balls and volume pricing from $21 to $32.
Product Page
|
