[Planetlab-users] information regarding sirius

[David E. Eisenstat]

On Mon, 28 May 2007, bikas gurung wrote:

> 1. Is Sirius running on all the planetlab nodes? Can we just randomly 
> select the nodes for our path or we need to check whether Sirius is 
> running on it or not?

Sirius is *supposed* to be running on all of the v4.0 nodes, but failures 
happen, and this isn't always the case. The handful of v3.3 nodes 
remaining are not running Sirius.

The list of nodes where the Sirius slice has a running process is here: 
http://summer.cs.princeton.edu/status/tabulator.cgi?table=slices/table_princeton_sirius

> 2. From FAQ - "Although your reservation is system-wide, your slice may 
> acquire resources at different times on different nodes." Do we really 
> need to check for each and every node we use whether we have been 
> assigned the resources or not? If that's the case, it might impose some 
> limitation on the number of tests we'll be able to conduct.

Sirius uses the node's clock to determine when to start and stop the 
resource loans. Thus if the local clock is significantly off (the cutoff 
isn't clear, but up to 30 seconds deviation is probably OK), you may 
receive resources early, late, or never at all.

Basically, Sirius delivers resources in a timely fashion on most nodes 
almost all of the time, but if you want reproducible measurements (insofar 
as PlanetLab measurements are reproducible) you should be keeping an eye 
on it. The FAQ is worded strongly because of a past incident where a 
NodeManager bug caused Sirius to do the wrong thing for months.

Best,

-David Eisenstat (Sirius maintainer)