Network Computingwww.networkcomputing.com

The OmniStack 6024 was designed to be deployed at the edge of the network. It has 24 fixed-configuration 10/100 Ethernet ports and three expansion slots. The first slot is reserved for an SNMP management module. The second slot accepts a single uplink module. The third slot accepts a switch stack module, letting you cascade up to four OmniStacks into a single logical switch. The stacking capabilities of the OmniStack are superior to most competitors'. The stacking bus supports a full 2 Gbps in each direction (aggregate 4 Gbps across the stack). Most competitors offer only half this capacity.

With the addition of the VoIP module, the OmniPCX is not only a full-featured voice switch, but also an attractive VoIP solution for large enterprise networks. The VoIP gateway board adds full VoIP encoding to the OmniPCX switch and serves as an H.323 gateway. Each module installed in the OmniPCX 4400 adds support for 30 concurrent VoIP connections. Scaling the capacity of the switch is as simple as adding another module to the OmniPCX 4400 switch. I installed 18 boards to handle 500 simultaneous VoIP calls.

Put to the Test

I tested several aspects of QoS and performance against the OmniCore-OmniStack-OmniPCX solution. My test configuration consisted of a pair of OmniCore 5022 switches at the core of the network. I connected the 18 modules from each OmniPCX directly to 18 Fast Ethernet ports at the network core. I also connected an OmniStack 6024 switch via Gigabit Ethernet to the core 5022 switches. Four Chariot end-point clients were attached to each OmniStack switch. Then I attached a SmartBits 2000 via Gigabit Ethernet to each core switch so that I could flood the gigabit backbone with wire-speed traffic. Finally, I deployed several digital and IP telephones on each end, which I used to subjectively measure the quality of calls placed across the LAN.

I tested three facets of the products' QoS and VoIP. I began by baselining the network to show that the lack of QoS severely limits the effectiveness of VoIP. Then I used SmartBits to saturate the test network backbone. While SmartBits was generating traffic, I was unable to place IP calls and several of the phones on our test network lost their connections to the IP PBX. I then proceeded to define QoS policies to protect the voice data. The policies were based on the port, Layer 3 address and Layer 4 TCP protocol number. Then I re-enabled the SmartBits wire-speed backbone traffic and placed several calls across the IP telephone network to validate that QoS was indeed functioning.

Alcatel's QoS implementation is based on a weighted round-robin scheme. This method ensures that the highest queue is guaranteed a certain fraction of bandwidth. The lower-priority queues are guaranteed a lesser fraction, and the sum of all the queues' weights is 100 percent. The version 2.7 code I tested does not let the user adjust queue weights; however, version 3.0 code does allow the end user to provision queue weights by specifying bandwidth parameters per queue.

My second test involved using multiple levels of QoS to guarantee delivery for both voice and mission-critical data traffic. I used Ganymede's Chariot to simulate SAP purchase-order queries through the OmniStack switch and across the backbone network. With no congestion on the network, I measured a baseline of 3,169 transactions per second and 334.6 Mbps between 32 Chariot clients and servers on the network. With SmartBits-simulated congestion, the Chariot traffic was unable to pass across the network. Using Alcatel's TrackView policy manager, I defined a policy to apply 802.1p tags at the edge of the network and convert the 802.1p network tags set by the OmniStack edge switch into an IP DiffServ tag at the core of the network. Rerunning the Chariot tests under load, I measured approximately 3,060 transactions per second and 323 Mbps of throughput with full wire-speed load across the gigabit core. The slight loss in throughput is due to Alcatel's nonstarving weighted round-robin queuing scheme.

My final set of tests involved measuring VoIP performance and capacity, as well as verifying that the quality of the VoIP calls is guaranteed via QoS mechanisms. Using a special hardware card, Alcatel set up 500 simultaneous voice calls across its OmniPCX voice-switching platform. This test was run for several hours while SmartBits traffic was blasting at full speed. I measured the system's steady state response and found that the switch is capable of sustaining 500 simultaneous calls at a rate of 9,600 calls per hour. Each simulated call lasted around one minute. I also performed analog-phone-to-IP, digital-phone-to-IP, and NetMeeting-to-analog-and-IP-phone interoperability tests. I found that the Alcatel solution could make calls from any service to any service.

Vendor Information:

OmniCore 5022,
Starts at $482.40 per port (10/100) or $1,999 per port (1000BASE);

OmniStack 6024,
Starts at $2,200;

OmniPCX 4400,
Starts at $40,000 (100 users).

Available: Now.
Alcatel Corp., (800) 999-9526. www.ind.alcatel.com

Send your comments on this article to Joel Conover at jconover@nwc.com.