MMC Networks Provides Per-User Services and Speeds Network Traffic With Per-Flow Queuing.

SUNNYVALE, Calif.--(BUSINESS WIRE)--May 27, 1997--MMC Networks, Inc. today announced a way to provide priority services for Intranet and Internet customers on a per-user basis -- a breakthrough that also speeds overall network traffic by quickly identifying and resolving connections that cause congestion.

MMC Networks' new Per-Flow Queuing (PFQ) technology, available in the XStream chip set, revolutionizes the design of Ethernet, ATM (Asynchronous Transfer Mode) and Frame Relay switches and routers so they can support hundreds of thousands of separate queues, one for each traffic flow. Because of this, PFQ enables true Quality of Service (QoS) on a per-user or per-class basis so some users can pay more for better access. Additionally, it also speeds IP (Internet Protocol) multicast delivery and the way it handles data, queuing both frames and cells, considerably speeds the network, especially for customers with Ethernet connections.

"With over 50 million people accessing the Internet, doing everything from viewing bandwidth-intensive multimedia programs to getting real-time stock quotes to checking their email, the demands on the networks for different services are rapidly increasing," stated Brent Bilger, MMC Networks' Vice President of Marketing. "In today's networks, it's a challenge to provide priority service, even for users willing to pay for it. That's why Per-Flow Queuing is so important. It provides an easy way to give different users the services they require."

Messages are transmitted over the Internet on a first-in, first-out (FIFO) basis. Users can buy faster connections to the Internet, such as an ISDN line, but once online, they're in line behind everyone else. When there's a traffic delay, they're stuck. And this problem is not reserved for the Internet; many companies are finding that their internal networks cannot provide certain users with the high-priority service they need.

Internet planners are developing new classes of service that are similar to what the airlines offer, like separating the traffic into first class, business class, coach, and even stand-by. But these innovations will not work properly in today's networks full of FIFO queues because too many different users will have to share these queues. Invariably, some connections will cause problems, and every connection behind it in the queue will be delayed.

Per-Flow Queuing Solves the Problems of FIFO Queues

With FIFO queues, one user's traffic sits in line behind the traffic of other users, whose behavior is unpredictable. For example, if one user is sending a huge file, response time can suffer for everyone else. In most switches, when a FIFO queue gets full, new traffic is refused until the queue is drained to an acceptable level. Once the queue is opened again to incoming traffic, there's no way to tell if it's just opening up to the same traffic that originally caused it to fill up. Most switch makers have implemented up to four FIFO queues offer different classes of service, but that's no solution for the poor user who's stuck in line behind someone who is using more than his fair share of network resources.

Per-Flow Queuing eliminates these problems by isolating the behavior of each traffic flow. PFQ can manage hundreds of thousands of flows through a network switch by setting up separate queues for each flow, then scheduling the queues based on a set of priorities.

Cisco Systems, a company that has worked closely with MMC on new technologies over the past five years, has made the Per-Flow Queuing technology available as a field upgrade to their award-winning LightStream 1010 campus and metropolitan area network ATM switch. According to Anthony Alles, Cisco's Director of Product Marketing, "When enabled by the Cisco IOS, the Per-Flow Queuing technology provides the most advanced traffic engineering and management mechanisms at unprecedented price/performance points." As a direct result of the close working relationship with MMC on the development of this technology, Cisco plans to build PFQ into future products in the LightStream family, according to Alles.

Per-Flow Queuing Speeds IP Multicast Delivery

MMC Network's fundamental switch architecture, called ViX, provides a natural advantage for the delivery of IP multicast. IP multicast is a standards-based way to send data once to many recipients, saving network bandwidth. The ViX architecture is ideal for multicast because only one copy of the data needs to take up memory. Once all the pointers have delivered the content to all the users, the memory can be cleared. PFQ can keep track of each multicast data flow, insuring high-quality delivery to those who want premium service.

Per-Flow Queuing Unifies Carrier ATM Services with the Internet

Another challenge is the fact that many of the major carriers of network traffic use ATM, which uses a standard-sized cell (5 bytes of header and 48 bytes of content), while the transmission standard of the Internet, IP (Internet Protocol), uses variable-sized packets or frames (up to 1500 bytes or more). It costs a lot of money to convert frames into cells to transfer through the faster ATM switches because this conversion must be done at each ATM source and destination using a function called Segmentation and Reassembly (SAR).

Per-Flow Queuing greatly lowers the cost of transmitting frames over ATM because it eliminates the need for this expensive frame-to-cell conversion at each source and destination. In addition to being able to receive and send either frame or cell traffic, the breakthrough PFQ-enabled ATM switch can be configured to receive frames and send them out as cells, or receive cells and transmit them as frames. This means that, with PFQ, frame switching, found in Ethernet networks, can work at the same speed and deliver the same advanced traffic management capabilities as cell switching, something no one believed possible just a few years ago.

"The frame mode of PFQ, co-developed by Cisco and MMC, will be vital to the operation of Cisco Systems' new Tag Switching mechanism," added Alles, "allowing us to build much more scaleable Internet backbones using our ATM switches by allowing the consolidation of traffic onto a smaller number of connections, without loss of advanced functionality."

The Per-Flow Queuing technology is embodied in the 5-chip XStream chip set, which is available now for a price of $2,000 in quantities of 1,000. The XStream chip set can be added to switches based on MMC Networks' ATMS2000 networkprocessor via a simple add-in card. XStream acts as a co-processor to the ATMS2000 networkprocessor. This breakthrough lets MMC Networks provide PFQ technology as a cost-effective add-in solution, facilitating the employment of these capabilities in network switches.

MMC Networks, Inc., is the leading independent supplier of networkprocessor chip sets for multigigabit wire-speed networks. With headquarters in Sunnyvale, Calif., MMC Networks is privately held and has received financing from Institutional Venture Partners, U.S. Venture Partners, Kleiner Perkins Caufield & Byers, and JAFCO. For more information on MMC Networks products, please call [phone] or visit http://www.mmcnet.com .

CONTACT: MMC Networks

Paula Jones, 408/731-1605

[e-mail]