Notes on Stargate Testbed Nodes

Contents

• Preface
• Stargates are not imageable
• One kernel for all
• No experimental network interfaces
• Directly attached Mica2 motes
• Limited resources
• Special note on Stargates attached to Emulab mobile nodes (garcias)

• Preface

  Emulab has deployed Intel Stargates on a limited basis for use in experiments. So far, they are mounted on a handful of Garcia robots, with a couple more standalone units coming soon.
  
  Emulab Stargate nodes are a lot like their PC node counterparts, with support for events, startup commands, tarball/rpm installation, account setup, and home directory access via NFS. Users with local_root or greater permissions also have full root access on them, allowing arbitrary changes to the filesystem, running kernel, etc.
  
  However, there are some important differences between stargates and regular PCs that should be noted:

• Stargates are not imageable

  It is not currently possible for users to create and use arbitrary OS images on these units (in contrast to a normal testbed PC node). After experiment swapin, a Stargate will contain the default Intel "PlatformX" linux kernel (2.4.19 with modifications for the Stargate), and a default Emulab root filesystem (of type jffs2). Users may, however, overload this default filesystem with their own files using Emulabs tarball/rpm installation mechanism, or by making changes manually. On experiment swapout, the units are resynched with a the standard filesystem image, reverting all user modifications.
  
  For more info on the base linux software running on the Emulab Stargates, visit the PlatformX Project page at Sourceforge.

• One kernel for all

  Users are currently limited to using the stock PlatformX kernel provided by Emulab. Once we begin to provide OS image loading and saving support, replacing the kernel should become possible. It may be possible to use the Linux "two-kernel monty" trick to load another kernel, but this has not been tested. For info on building a Stargate kernel, please see the following page at the PlatformX project site: Kernel Building

• No experimental network interfaces

  The Emulab Stargates currently only have a single wifi interface, and it is used for control network connectivity. Therefore, users do not have access to any dedicated experimental interfaces for artifact-free experimentation. Communication between other machines is, of course, still possible via the control network interface.
  
  We plan to add at least one additional network interface for experimentation to the Stargates in the not too distant future.

• Directly attached Mica2 motes

  The Emulab Stargates do have Mica2 motes attached, which can, of course, be used for wireless experimentation. This can happen either in conjunction with the parent Stargate, or just on/between the motes themselves. Emulab can automatically manage loading the images on these motes (as documented elsewhere).

• Limited resources

  Its prudent to not forget that these units are not full-featured PCs; they run at 400MHz, do not have hardware floating point units, have 32MB of RAM, and only 64MB of flash disk (though we will likely expand the latter with an additional CF disk soon).
  
  Please also take care not to write at a high rate to the flash disks. Compared to regular old rotating magnetic media, these solid-state disks wear out quickly (good for about a million writes per block). The jffs2 filesystem amortizes the writes across the flash to help reduce wear on particular blocks, but please tread lightly anyway.

• Special note on Stargates attached to Emulab mobile nodes (garcias)

  The Stargates coupled with our mobile nodes run a daemon that enables the motion of robots as driven by Emulab central (garcia_pilot). If you don't want your experiment to (literally) stop in its tracks, please do not disable this process.