February 2004

The Debate Fizzles: Chassis Versus Stackable At The LAN Edge

BY TONY RYBCZYNSKI & PHIL EDHOLM

 
Over the past 10 years, an ongoing debate has raged over the comparison between modular chassis-based LAN edge solutions with hot swappable blades and component stackable solutions with some form of resilient high-speed interconnect mechanism between units. While there is little disagreement that for core switching applications, the modular chassis, with its combination of bandwidth, protocol layers, and fan-out is ideal, the LAN wiring closet has been an area of great contention� until now.

Let�s examine a number of critical points of comparison (reliability, scalability, manageability, serviceability, LAN powering, physical restraints) to enable the reader to develop a logical framework within which to analyze his or her particular needs and then make a decision based on requirements and attributes, rather than on unverified claims or religious fervor.

THE PUTS AND TAKES
Switch Reliability is a critical measurement for edge devices, in that, in a typical network, such devices have a single Ethernet link to each PC. Modular Chassis employ redundant power supplies, fan trays, control plane/switching fabrics, and uplink cards to assure that a physical device failure will not impact overall operation. By limiting the number of ports per card, the impact of a failure can be limited to the number of interfaces on the card, in the 12 to 48 range. Stackables are often thought to have much lower mean time between failures (MTBF). However, enterprise-class stackables can achieve the same level of redundancy by having per-unit redundant power and fans in each unit and a robust interconnection mechanism. Most stackables provide some level of redundancy by employing a unidirectional simplex ring architecture. More sophisticated stackable architectures support duplicated bi-directional rings with intelligence built in to ensure the shortest path is used. As self-contained units, stackables have a potential advantage in that they are more immune to electrostatic discharge, a growing concern as components move to lower voltages for increased speed.

Scalability is critical to assuring that the investment made will continue to deliver value for an extended period. The essential difference between stackable and modular scaling is in the mechanism employed to grow the number of ports, increase bandwidth, and add protocol capability. In the modular example, in general, blades are added or higher-density blades inserted to add capacity. Since there is a fixed number of slots, the addition of uplink blades comes at the expense of limiting the number of user interface blades. That said, a chassis has engineered parameters such as a fixed number of slots, fixed backplane speed, and specific capabilities in terms of voltages and power delivered to each blade. Requirements that cannot be met within these design constraints require the introduction of a new chassis.

A stackable solution generally scales by adding new units, possibly of higher capacity. As the power for each unit is separate, the capability to include power levels and voltages/currents specifically adapted to a new capability is built in. However, there are limitations to the scalability of the stack: the capacity of the interconnect bus in terms of bus speed (typically tens of Gbps) and number of units supported (typically eight). Only units in a stack without uplinks use up capacity on the bus. Essentially, stackables can provide a virtually limitless capacity for uplink upgrade, and where there is a cost effective relationship between desktop links and uplinks, they can create a virtual or true non-blocking environment. For example, for a stack of eight modules (e.g., supporting 192 ports with 24-port density), an initial configuration would typically have uplinks from the two end units. To increase capacity by 100 percent, the stack connection is broken in the middle (yielding two separate four-unit stacks), and uplinks are added. This process can be repeated until each individual stackable unit has two uplinks to the core switch. A 24-port 10/100 Mbps unit that has two active 1 Gbps uplinks effectively provides non-blocking operation.

In the end, both stackables and modular chassis switches are highly scalable, however, the flexibility of stackables may be an advantage in some cases.
Serviceability is another consideration. A chassis solution has a master processor or controller. Upgrading the software in this single point upgrades the entire chassis. In the stackable implementation, each unit has its own software load, a potential disadvantage unless all units in a stack run the same software and an automated system is used. On the other hand, this can potentially be an advantage as software can be loaded and tested in each unit without risking the impact of a glitch bringing down the entire stack. With free slots, it is easy to add or replace cards in a chassis configuration. On the other hand, adding or replacing units in a stackable configuration requires more planning to avoid having to reconfigure uplinks.

Manageability requirements dictate that both stackables and modulars be managed as a single system. However, while the management system graphical interface represents a modular by its physical attributes, it represents a stackable as a logical assembly of units. The chassis presentation has significant advantages in assuring that troubleshooting isolates the right component during routine or emergency field repair, though this can be mitigated for stackables with properly designed management systems and associated processes.

Power over Ethernet support is another area of consideration crucial to IP telephony, WLAN Access Points, and IP security and surveillance devices. To provide LAN power in a chassis, power capacity is typically engineered for the expected slot utilization and per card power consumption. Installing too many power hungry cards can result in a system failure, if the available power cannot support the new demand. By comparison, the stackable has a �clean� story in that each unit is individually powered. As the ports requiring power are a small subset of the total, this provides a mechanism to get all those ports needing power to a single unit without impacting the other ports.

Rack space is the final consideration. While chassis blades occupy approximately 75 percent of the area of stackable units, stackables generally occupy 30-50 percent less vertical rack space than an equivalent chassis solution. This is a result of chasses (which come in multiple discrete sizes between three and up to 14 slots) carrying a fixed overhead for controller cards, power and fans and often being operated with only 50-75 percent of slots filled.

A MATTER OF CHOICE
In the 90s, the sophisticated network architect routinely engineered chassis solutions due to their inherent superiority in reliability, maintainability, and manageability. Stackables just didn�t meet the needs. Today, the enterprise-grade stackable solutions have evolved to match and in some cases surpass chassis solutions.

Initial and lifecycle costs, and general IT preferences may be the main factors in deciding between the two approaches. Stackables generally have lower initial cost. Lifecycle costs will depend on how well initial and future requirements can be defined, on the degree of churn and the nature of that churn. For example, what�s the expected penetration of power over Ethernet, Gigabit desktops, VoIP, and WLANs? In the end, IT needs to decide which approach they are most comfortable with, since modulars and stackables can both provide the required feature/functionality, flexibility for growth, and evolvability to accommodate technology discontinuities.

Tony Rybczynski is Director of Strategic Enterprise Technologies at Nortel Networks. He has over 30 years experience in the application of packet network technology. Phil Edholm is Chief Technologist and VP Enterprise Network Architecture in Nortel Networks.

If you are interested in purchasing reprints of this article (in either print or HTML format), please visit Reprint Management Services online at www.reprintbuyer.com or contact a representative via e-mail at [email protected] or by phone at 800-290-5460.

[ Return To The February 2004 Table Of Contents ]