3.3 Scalability of Satellite Devices

The second area that you should carefully consider is the scalability of the Satellite devices. Satellite devices are designed to reduce load on the network, not to reduce load on the Primary servers. Satellite devices help reduce redundant traffic, load, and utilization from the WAN. Even if devices are located at a remote site, they connect to the central site for authentication and checking in with the Primary Servers to see if there is work to do. The actual work should be performed with remote Satellite devices strategically placed to service work requests from managed devices.

The following sections contain more information:

3.3.1 Factors Influencing Scalability

The major factors influencing scalability of Satellite devices include:

  • Disk I/O and the requests that the Satellite device is concurrently managing

  • Size of the subnets and their respective network speed

  • Services the Satellite device is performing (imaging, inventory collection, and distribution)

  • Disk capacity (the Satellite device must have enough capacity to cope with the required content)

  • Physical memory (RAM) installed on the Satellite device

  • The number of managed devices the Satellite device is managing

  • The frequency of distributions and the number of concurrent connections

  • Whether inventory collection or software distributions are randomized

  • Whether an L4 switch fronts the Satellite devices at a particular location

  • Whether Satellite device groups are used

  • Class of hardware (server-class hardware performs better than workstation-class hardware)

  • Class of operating system (a Satellite device running on Windows Server can handle more requests and workload than a Satellite device running on Windows XP or Windows Vista)

Keeping these factors in mind, you should build your design to manage the known devices and estimated ongoing workload.

3.3.2 Load Testing in the Novell SuperLab

Novell performed Satellite device scale tests using both server-class and workstation-class operating systems. As with testing that was performed on the Primary Server, the purpose of the tests was to find the point where the systems began to reach their limits. This gives us an understanding of scale under severe load.

The tests included the following:

  • Server-class Satellite devices: Dell PowerEdge 2950, Dual Quad Core 2.0Ghz, 8 GB RAM, 2 X 300 GB SAS hard drives running Windows 2003 Enterprise.

  • Workstation-class Satellite devices: White box, AMD 3400+, 2 GB RAM, 80 GB SATA hard drive running Windows XP SP3.

The Novell Corporate Configurations Test (CCT) team, using the baseline test of 250 bundles (1 KB file bundles), had the following results:

  • A server operating system (Windows 2003 Enterprise) machine can scale to 1,000 managed devices.

  • A workstation operating system (Windows XP SP3) machine can scale to 250 managed devices.

The following sections contain information about three tests performed in the SuperLab with Satellite devices:

Test 1: Bundle Size

Test results show how large a single bundle can be to be distributed by a Satellite device to managed devices. These results are useful when you begin to estimate how many Satellite devices are required based on the ongoing estimated load. You need to calculate estimated load during the design phase as you work with the different groups that manage processes around software distribution, patch distribution, inventory collection, and image distribution. You need to know how frequently these processes are required, and the estimated size of the typical distribution or collection.

Bundle Size (MB)

Server OS

Number of Managed Devices in Test

Server OS

Number of Managed Devices Successful

Workstation OS

Number of Managed Devices in Test

Workstation OS

Number of Managed Devices Successful

6

1004

1004

259

259

10

1004

1004

259

259

25

1004

968

259

259

50

1004

801

259

248

The following graph shows what you can expect in terms of Satellite device scalability given the conditions of the test:

Figure 3-3 Bundle size.

The graph illustrates the probable scalability limitations of the Satellite device. Under normal load, and through the use of proper configuration parameters (for example, randomizing distributions and collections) we can comfortably conclude that a Satellite device running on server-class hardware and Windows Server can scale to approximately 1,000 managed devices, and a Satellite device running on workstation-class hardware and Windows XP or Vista can scale to approximately 250 managed devices.

Test 2: Server OS Delivering Multiple Chained Bundles

This test shows the results of a Satellite device running on a server OS delivering multiple chained bundles to connected managed devices. This information is useful when calculating how much load you can place on the Satellite device running on a Server OS when deploying applications that are chained. This obviously can differ greatly from deploying single bundles.

Number of Bundles

Bundle Sizes in MB

Total Number of Managed Devices

Number of Devices Successful Managed

5

10, 15, 20, 25, 30

999

996

4

5, 10, 15, 20

999

996

Test 3: Workstation OS Delivering Multiple Chained Bundles

This test shows the results of a Satellite device running on a workstation OS delivering multiple chained bundles to connected managed devices. This information is useful when calculating how much load you can place on the Satellite device on a workstation OS when deploying applications that are chained together. This obviously can differ greatly from deploying single bundles.

Number of Bundles

Bundle Sizes in MB

Total number of Managed Devices

Number of Devices Successfully Managed

5

10, 15, 20, 25, 30

259

259

4

5, 10, 15, 20

259

259

3.3.3 Achieving Scalability in the Real World

The testing in the SuperLab tests the upper limits of a Satellite device. It is important to note that if you are deploying a bundle that is less than 25 MB in size, we see successful distributions to both 250 and 1,000 managed devices from workstation-class and server-class Satellite devices. With this as a known, we can safely assume that by doing the following, you can achieve these levels of scalability, and quite possibly much more:

  • For larger sites (more than 250 managed device):

    • Have a dedicated Satellite device for imaging purposes.

    • Have a dedicated Satellite device for inventory collection if the collection frequency is high. In other words, if you are collecting daily, you want the server to be dedicated, but if you are collecting monthly, you can collapse this service into another Satellite device onsite.

    • Have a dedicated set of Satellite devices for software and patch distributions if the frequency of distributions is high. You want to randomize the distribution of software and avoid massive numbers of devices hitting the Satellite device at the same time.

    • Randomize the refreshes of managed devices at the site with Satellite devices.

  • For smaller sites (fewer than 250 devices):

    • Have multiple Satellite device that share load and responsibility.

    • Do not be significantly concerned about designating specific servers for specific functions.

    • Randomize the refreshes of managed devices at the site with Satellite devices.