|
|
|
Netli's NetLightning
The company's NetLightning service accelerates Web applications such as e-commerce, support portals, and online business communication services by optimizing IP transport over the core of the network, or the "middle mile."
Network Magazine, By Doug Allen: July 7, 2003
Bringing high performance to IP remains the holy grail for IT managers. While QoS protocols, traffic shaping, load balancing, and caching all help, best-effort service is still often subpar. This can frustrate customers and partners, and often forces companies to build out a more distributed data center infrastructure at high cost. The problem is delay caused by geographic distance; the farther the run between the end user and the data center, the worse the response time. In the case of TCP and HTTP, Web pages often require dozens of round-trip connections to download a multi-element Web page. Packet loss and congestion compound the problem, and bandwidth alone can't fully fix it.
Netli (www.netli.com), an overlay provider start-up founded in 2000, may have solved a piece of the puzzle with what it calls the first application delivery network. The company's NetLightning service accelerates Web applications such as e-commerce, support portals, and online business communication services by optimizing IP transport over the core of the network, or the "middle mile."
NetLightning works by setting up a network-based layer-4 VPN of sorts between a business customer's data center and the end user. Outgoing data center traffic is routed onto Netli's overlay network, which consists of 13 global POPs linked to multiple Tier-1 and Tier-2 ISPs through neutral exchange and colocation provider Equinix and intelligent routing provider Internap. Once on the network, traffic is converted to a proprietary protocol that optimizes transport by minimizing the number of round-trip connections. This speeds up the rate of exchange, as well as bypasses IP's slow-start transmission feature. (Netli also adds caching and compression where helpful.) Then it's back to the POP closest to the end user, where the session is converted back to IP for the final hop to its destination. The process works in both directions.
This approach reduces global response time to a few seconds or even sub-seconds, while supporting Secure Sockets Layer (SSL) encryption if desired. Users benefit from increased application availability without having to distribute data centers, reengineer applications, or install remote clients to fit the Netli service. However, NetLightning is out of its depth when it comes to improving streaming content or server slowdowns due to flash crowds-that's a job for content delivery networks.
KEYS TO THE SUPERHIGHWAY
How does it work? NetLightning is built on Netli's Reliable Application Platform for Instant Delivery (RAPID) architecture (see figure). The key elements here are the Virtual Data Centers (VDCs) and the Application Access Points (AAPs), which act as gateways to the end users and Web application servers, respectively-on- and off-ramps to the Netli overlay network, if you will. Using a global DNS redirection and IP address mapping system, the AAP (the on-ramp) intercepts incoming requests for a specified Web application and sends them to the VDC closest to the end user. Netli's secret sauce protocol reduces the number of round-trips and request responses inherent in TCP over the middle mile, offsetting any packet loss and congestion.
The customer can dictate where an AAP is placed (typically as close to the data center as possible), while the VDCs are distributed globally so as to reduce the final hop distance as much as possible, thereby minimizing total response time to the end user. A 24-by-7 NOC handles management, monitoring, and support chores.
"Netli works best for highly dynamic and/or interactive translation-type Web applications with little repeated traffic, aimed at a global audience that is not limited to branch offices (i.e., the customer doesn't own the remote premises or clients)," says Peter Firstbrook, a META Group (www.metagroup.com) analyst. "It is pretty expensive when compared to other content delivery solutions, but it is really the only solution for this type of problem. It only really offers a significant payback when compared to using regional data centers to accelerate application performance, yielding a three-year Return on Investment ROI of 50 percent or more."
Early customer wins include multinational companies HP and bioscience vendor Millipore, which have reported significant performance improvement and, consequently, greater customer and partner satisfaction and service uptake, as well as a quick ROI in data center savings. Specifically, Millipore cites consistent global transaction times in the one- to three-second range; an increase in site traffic by 33 percent from the previous year and rapid growth in online sales, particularly in Asia; and the virtual elimination of performance complaints from Asian customers, partners, and staff.
A typical customer implementation costs between about $10,000 and $15,000 per application. The cost is broken down into a per-Mbit/sec charge and a separate service rate, which includes NOC support and service monitoring.
|