Microsoft KB Archive/932018
Article ID: 932018
Article Last Modified on 5/16/2007
- Microsoft SoftGrid for Terminal Services
- Microsoft SoftGrid for Windows Desktops
This article describes how SoftGrid Networking works together with Microsoft Windows Server 2003 Network Load Balancing.
Note For more information about streaming protocols and about the corresponding ports that are used by SoftGrid, click the following article number to view the article in the Microsoft Knowledge Base:
932017 Introduction to SoftGrid networking
Do this before you set up a load balancer.
Network Load Balancing (NLB) is built into all versions of Windows Server 2003. Unlike hardware appliance-based solutions, NLB is a kernel-level software driver that runs on each Windows host computer that participates in the SoftGrid Virtual Application Server NLB cluster.
Each host in the cluster receives all incoming virtual IP traffic on a cluster network adapter on each computer. One computer is selected to respond by using an algorithm that is run on each host.
The primary benefit of NLB is significant cost savings over hardware load balancers. Hardware load balancers cost thousands of dollars each. However, NLB has some limitations, as follows:
- NLB will not detect and remove a host from the cluster if the following conditions are true:
- The host is still running.
- The Softricity SoftGrid Server NT Service is not responding.
- NLB heartbeat packets are sent out every second by every host in the cluster. These packets are received by every computer in the same subnet. Non-cluster computers discard the packets.
These broadcast packets are currently smaller than 1500 bytes each, and they will not travel across a router.
Note These heartbeat packets are distinct from the “switch flooding” option that is discussed in the NLB documentation in the Windows Server 2003 Help and Support Center. For more information, visit the following Microsoft Web site:
- After a SoftGrid client computer is associated with a particular cluster host, that mapping is never broken unless cluster membership changes.
In many hardware-based load balancers in direct-dispatch mode, there is a configurable time-out value. However, for performance reasons, we do not recommend this direct-dispatch mode.
- Because each cluster host receives all incoming cluster traffic and then responds to or discards a particular packet, memory usage and CPU usage are increased on each cluster host. However, in SoftGrid, the ratio of incoming to outgoing cluster traffic makes this extra processing in NLB insignificant.
Before you implement NLB, read the Microsoft documentation on this topic. To do this, visit the following Microsoft Web site:
Keywords: kbtshoot kbexpertiseinter KB932018