AdvancedMCs are finding their place
| By |
|
AdvancedMCs are ideal for introducing modularity, flexibility, and scalability while addressing many telco requirements including hot swap. However, the requisite thermal and form factor specifications place limits on the computational and I/O configurations these mezzanine cards can implement.
Understanding the AdvancedMC solution space, many system designers use modules for adjunct functionality, as evidenced by most AdvancedTCA implementations incorporating a module site, particularly compute blades. The market for modules is strong, creating demand for a large number of available products that support solutions for computing, storage, and I/O, just to name a few. In addition to telecom, there are AdvancedMC success stories in Military, Aerospace, and Government (MAG) and also in low-end gateway appliances such as WiMAX base stations and other access boxes. Access deployments exceed core and edge, so the sheer size of the opportunity bodes well for AdvancedMC.
The excitement over AdvancedMC benefits, which include hot swap, fewer field replaceable units, and a Lego Block design approach, has some enthusiasts believing modules should be the staple for system designers. Designers need to consider these data flow and system requirements, however:
- 1 GbE versus 10 GbE fabric
- Which processor architecture is used
- Options for addressing security, line cards, and redundancy
After considering the points just noted there will be circumstances where modules are too limiting. It's essential to determine the right fit for AdvancedMCs in light of thermals, connectivity, usage models, and functional modularity requirements.
Thermal testing
Whether a designer is working with a quad module AdvancedTCA carrier blade or a compute blade with a single site, it's necessary to evaluate the airflow. For example, a storage module with a hard disk drive mounted on top can block the air needed to cool downstream components. Thermal solutions require engineering and validation effort, and it's not enough to just dial down the power or the number of AdvancedMC modules. Having conducted thermal modeling on our carrier card with four module sites, we published guidelines informing customers how to work with it.
There are no universal thermal guidelines for AdvancedMC because environmental conditions are codependent on the type of modules employed and how they affect the airflow. System designers should be careful, because even if all the specs are met, it's still possible to run into cooling issues.
"One of the pain points for most people working in AdvancedTCA is cooling the blades. To address that, we've developed a set of interoperability requirements based around thermals and, of course, test procedures corresponding to those requirements," says Todd Keaffaber, Communications Platforms Trade Association (CP-TA) Technical Work Group Chair.
Module designers might not be able to max out the memory or use the highest performance CPU and still comply with the 40 W limit for a single-width board. As the silicon technology advances, shrinking die sizes and lowering thermals, AdvancedMCs will become increasingly more attractive. Progress is being made on performance-per-watt by devices such as the 2.2 W Intel Atom processor, which might be a good fit for AdvancedMC. Still, more help is needed from silicon makers to lower the power consumption of processors, memory, and physical layer chips.
Connectivity challenges
When it comes to I/O, it's really about Ethernet, and the question is whether to deploy 1 GbE or 10 GbE. E1/T1 and STM-1 aren't going away, but the dominance of Ethernet will continue for at least the next few years. Has any technology taken on Ethernet and won?
AdvancedMC is a good fit for I/O functionality, such as generic E1/T1, but designers must match their I/O through-put with their processing capability. Placing multiple 1 GbE and 10 GbE interfaces requires a lot of processing horsepower, and without a multicore processor and hardware acceleration the design could be unbalanced. We implemented an OCTEON CN58xx/CN38xx from Cavium, which can provide wirespeed packet processing for L2-L7 for the full line rate of 4 Gbps (4 x 1 GbE), as shown in Figure 1.
As networks transition to 10 GbE, module designers anxiously await optimized 10GBASE-T physical layer components (PHYs) that consume less power than the 10 to 12 W solutions launched last year. More costly 10 GbE optical modules consume about 3 W, but they are not costeffective for some applications. Fortunately, lower-cost copper-based solutions, which operate within the same power specifications as today's optical modules, are projected in the not too distant future.
For faster line rates, transceivers that support the new IEEE 802.3ap standard (also referred to as 10GBASE-KR) are available for running 10 Gbps serial data over backplane systems. Historically, Ethernet-based backplane interfaces were not standardized because Ethernet was designed for box interfaces. As a serial protocol, 802.3ap is simpler to route than Ethernet. Still, designers should be judicious when dealing with 10 Gbps, which might call for a beefed-up processor to handle the high throughput.
AdvancedMC is a clear winner for access devices where modularity is a requirement, GbE performance is sufficient, and there's less emphasis on availability and redundancy. Performance bottlenecks are less common in systems with lower I/O and computing requirements, and OEMs can easily employ standards based modules that provide I/O scalability and reduce development effort.
|
|
|
Figure 1: Isolate and protect (click graphic to zoom by 2.2x) |
The modularity of AdvancedMCs supports some practical usage models. It provides a means for customers to protect their Intellectual Property (IP) by isolating functionality onto custom modules. For example, customers with proprietary accelerators or failover mechanisms can deploy their functionality without revealing crucial details to third parties. Customers can also create a family of products using module options, like offering storage modules with a range of capacity or computing modules with different levels of performance.
MAG and some other verticals value standards based modularity because it reduces certification and inventory costs. Modules can be upgraded as more processing power becomes available without having to upgrade and recertify the entire system. This is valuable because MAG design cycles are extremely long due to the need to addressing special requirements for security, ruggedization, sealed enclosures, and multiple radios. So CPU processing technology can be outmoded by the time the system actually starts to be deployed, and a modular approach allows the prime contractor to swap out the module when instructed to do so.
Is AdvancedMC a commodity? Not exactly. Yes, AdvancedMCs offer more reuse and economies of scale for relatively generic functionality (general-purpose computing versus application-specific protocols). But because low value subsystems are typically commoditized, this doesn't apply to many AdvancedMC modules that implement fairly sophisticated functionality. AdvancedMCs are used to standardize and reuse functionality, but it's difficult to achieve the large volumes needed to make them commodities. So don't expect to buy AdvancedMC modules at your neighborhood electronics store.
Functional modularity
It's important to take a top-down approach when identifying functional modules. System designers should take a platform or system approach by first partitioning the architecture into efficient subsystems and then modularizing those subsystems. In this way, they will define the right modules without messing up the data flow and the architecture.
When using a bottom-up approach - defining and piecing together a set of common modules bottlenecks might arise from poorly provisioned modules lacking sufficient processing or I/O resources. In other words, designers shouldn't skip performing thorough data throughput analysis when building complex equipment out of modules.
Cost/performance trends are key
AdvancedMC modules won't solve every computing and I/O challenge, but they have their place, and developers are figuring out where they make the most sense. AdvancedMCs may play a key role in access equipment located at the network's edge, where the scalability, size, and relatively low cost of MicroTCA are compelling. This growing access equipment market can deploy barebones, low-end MicroTCA systems, including chassis, backplane, and virtual carrier card, for less than $500. Customers in other vertical markets, where cost, modularity, and standards-based solutions are highly valued, will pay close attention to the cost/performance trends of AdvancedMC-based solutions.
RadiSys Corporation
www.radisys.com
vp@radisys.com
