25.02.2013 Views

TCP/IP Tutorial and Technical Overview - IBM Redbooks

TCP/IP Tutorial and Technical Overview - IBM Redbooks

TCP/IP Tutorial and Technical Overview - IBM Redbooks

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

24.8.2 Encapsulation<br />

As shown in Figure 24-6 on page 922, NAT offerings sometimes enforce the<br />

need to see both inbound <strong>and</strong> outbound requests by obliging the NAT device to<br />

be installed as a bridge (without permitting bridges of any other kind), thus<br />

forcing the servers on to what is essentially a private segment. This can<br />

complicate installation, because it requires a significant physical change to<br />

existing networks. All traffic for those servers must pass through the load<br />

balancer whether the traffic is to be load balanced or not.<br />

The one advantage of NAT as originally conceived (the ability to forward packets<br />

to remote destinations across a wide area network) cannot be usefully deployed<br />

because the wide area network connection is behind the bridge <strong>and</strong>, therefore,<br />

can only be within the site's private network. Additionally, the same NAT device<br />

must still be the only exit from the wide area network link.<br />

To attempt to overcome these limitations, some NAT solutions add to the overall<br />

inefficiency that is fundamental to NAT by providing add-ons. For example, the<br />

capability to map one port address to another. Refer to “Network Address Port<br />

Translation (NAPT)” on page 93.<br />

To check if a server is up, NAT-based load balancing solutions need to sacrifice<br />

an actual client request, <strong>and</strong> so a server outage is typically perceived only as a<br />

result of a timeout of one of these real client requests.<br />

NAT devices often only map affinity or stickiness based on the client's <strong>IP</strong><br />

address, <strong>and</strong> not at the port level. This means that after a client has contacted a<br />

server, all traffic from that client that is intended for other applications is<br />

forwarded to the same server. This drastically restricts configuration flexibility, in<br />

many cases rendering the sticky capability unusable in the real world.<br />

Another approach to load balancing is proxies that encapsulate packets rather<br />

than modifying them, <strong>and</strong> then pass them to the server. This approach has some<br />

merit, particularly because it permits the load balancer to forward traffic across a<br />

wide area network, unlike the bridging NAT solutions. But other implementations<br />

use encapsulation for all traffic, <strong>and</strong> this requires an agent of the load balancer to<br />

be installed on each server. This agent reverses the encapsulation process. As a<br />

result, the choice of server platform is, by definition, restricted to the platforms for<br />

which the server agent is available. Also, like NAT, it entails further processing of<br />

the packet, which increases the likelihood that it will not be scalable to the levels<br />

required for major sites.<br />

Encapsulation is discussed in further detail in RFC 1701, which appears to be<br />

updated by RFC 2784.<br />

Chapter 24. Availability, scalability, <strong>and</strong> load balancing 923

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!