One-to-Many Server Farm Administration

Abstract: This whitepaper discusses the research and development of a next generation administration solution targeted toward XSPs with medium to large server farms.

Introduction

With the introduction of networked servers came the need to administrate these machines from other computers and workstations. This is especially true with the rise of the Internet and the creation of companies that provide connection to the Internet and other Internet-related services.

Telnet Administration

The first program used for administration of remote servers was telnet, a program that simulated a console shell on a remote computer. This allowed the administrator to do everything they were accustom to do on the server's console, but on their remote workstation.

While this was very successful (and its child, SSH, has changed much more than encrypt the channel), it required the administrator to be technically saavy on the various computer servers in the farm. It also meant that an administrator would have to "log in" each server independently. She could not simply perform similar commands on multiple servers at one time.

Web-based Administration

To address the need for techical administrators came the programs that offered web-based administration. While they did not always hide or abstract the complexity, they often allowed the administrator to more easily perform complex administrative tasks.

Next Generation

The next generation of remote administration is often called "one-to-many" as it allows a single administrator to perform a single task on multiple servers at the same time. The other advantage is that summary analysis can be done on all machines at one time, instead of requiring the administrator to log on to each machine in particular.

While the concept of one-to-many administration has been around for many years (HP's Openview being the most widely known), the emphasis has been on SNMP reporting of problems rather than of configuration. The newer versions of one-to-many that many companies are now starting to develop build upon web-based platforms.

Benefits

The largest benefit for a one-to-many solution is the increase efficiency. If few system administrators are required, the ISP can now increase the number of servers without increasing the cost of operating the servers.

Competition Solutions

These benefits are such a compelling argument, that many solutions are currently being developed by many companies. Here is a small listing:

• VA Cluster Manager is an open-source project sponsored by VA Linux.
  
  
• ManageRaQ is Cobalt's solution. It is primarily a method for monitoring, but includes maintainence of both system and network resources.
  
  
• Dynamic Management Kit is Sun's solution for one to many on their hardware. Not suprisingly it uses Java and Jini and a few of their other "closed and costly" API's.
  
  
• Linux Utility for cluster Installation (LUI) is one of IBM's host of tools that are going after the one-to-many market. IBM already has HA tools including clustering and management for it's aix marktet. It is currently going after the linux market pretty strongly with their "blue hammer" line. Very little info. is available on this yet though. This is for remote installation. It's all open source.
  
  
• Parallel System Support Programs (PSSP) is for management of AIX systems currently, but it is being ported to Linux.
  
  
• Cluster Computing Whitepaper is a whitepaper describing IBM's clustering solution. Note: There is in this paper, a study that IBM did on competiting HA solutions.

First Example- Analysis

Let's describe the benefits of one-to-many server farm management by way of an example. Tim, the administrator of tim.com, a small ISP, has a farm of 9 servers. He has 20 customers with one or two domains per customer. These are spread out among the 9 servers.

When a new customer calls in with a new domain to add to his farm, he needs to decide which server should virtually host the domain. Under tradtional methods, he would need to log in to each machine to determine the load and the availability.

With a one-to-many solution, he would simply pull up a collective load/availability graph of all the servers in his farm.

sol.tim.com		tim.com   teletubbyland.com   immunizeyou.org kidsRpeople2.org
mercury.tim.com		whatsupdoc.com   funnysmallpeople.org   bugs-n-taffy-pull.com kill-your.tv
venus.tim.com		john-reynolds.com   howardabrams.com   flavorCrystals.org GMOControversy.org   new-stakes.com   wharIsYou.com   whatIsYou.com
mars.tim.com		latejohnny.com
jupiter.tim.com		hushpuppies.com   gwbush.com   lovemenot.com
saturn.tim.com		whosOnFirst.com   johnWatsonMemorial.org   bigLoads.com sherwoodForest.com   iceColdOne.com
uranus.tim.com		tinkywinky.com   frankRego.com   frankRego.org   tealeaves.com   iLoveCaffeine.org
neptune.tim.com		flora-n-tina.org   national-arbor.org   suzyQ.com lazySusans.com
pluto.tim.com		antiques4sale.com   net4you.com   jini-solutions.org dominicReynolds.com   skroahtum.org

Now, if Tim had simply visited each server, he would have thought that mars would be the best server since it only has one site. However, according to the graph, jupiter, with three domains, actually has more capacity to host a new domain.

Of course, Tim would like to know why mars has such little capacity for only hosting a single domain. Tim would then click on the mars server and be displayed the following:

latejohnny.com   Bandwidth:     Disk Space:     CPU Use:     Memory Use:

From the graph, Tim can tell that the latejohnny.com domain is actually using quite a bit of disk space. Tim can decide to either place disk space quotas or investigate further.

All this analysis was done without needing to log into multiple servers. This analysis was distributed and combined to present a more efficient view of the data.

Second Example- Multi-Host Configuration

The next example demonstrates a more complex situation. Tim received a call at 1:00 in the morning from the owner of iceColdOne.com complaining that his web site was down. A quick analysis told him that 5 servers were not responding to the outside world, but could still be seen via his one-to-many console.

In order to increase his capacity and his redundancy, Tim got two separate DSL accounts from two different providers. It seems that his Qwest DSL was down again. Since all 9 servers use a single switch, he just needs to set the gateway of the 5 down servers to the GTE DSL modem.

Instead of logging in to each of the 5 machines, he can use his one-to-many console. He first goes to the menu to select, "Change Network Settings." The next screen asks him which servers should be altered. The list box not only shows him his 9 servers, but also various "groupings" that he predefined.

He could either select the five servers individually, or simply select the "Qwest Servers" grouping. The next screens would allow him to change the network settings, including the gateway address. None of these configurations required him to log on to particular servers.