attachment:notes_qiang.txt of msri07/notes

   1 11:30 a.m.
   2 Distributed SAGE
   3 Yi Qiang
   4 
   5 Main Objectives:  Accessible, scalable, cross-platform (SAGE-dependant), uses local and non-local resources.
   6 	Many times, the computing power will be there, but there will be no infrastructure to take use of those resources. "Department of Hidden Resources"
   7 
   8 Suited for coarse parallel tasks.  NOT "embarassingly" parallelizable problems.  Proudly parallelizable.
   9 Should have low requirements for synchronization
  10 
  11 A Worker in DSAGE consists of a monitor, which spawns multiple instances of SAGE to work on the tasks.
  12 
  13 Technologies:
  14 	Python
  15 	Twisted ("Build the engine of your Internet")
  16 	ZODB ("Zope Object Database")
  17 
  18 Why DSAGE?
  19 	Xgrid, BOINC, Chainsaw (IPython 1), et al. seem to do the same things.
  20 	These packages don't seem to make sense for the types of jobs the casual user (us?) may want to do.
  21 	Xgrid knows nothing about math/science, proprietary, and only works on OS X
  22 	BOINC is suitable for one massive job.
  23 	Chainsaw has no fault tolerance or authentication
  24 
  25 Fault Tolerance
  26 	Servers, workers, or clients can disappear at any time.  These factors must be taken into account in order to keep jobs running smoothly.
  27 
  28 Security
  29 	SSL communication by default
  30 	Public key authentication of clients
  31 	Run workers in a virtual machine (in the near future)
  32 
  33 Applications
  34 	Implemented Distributed Functions (such as Ray-Tracing or Integer Factorization)
  35 	Many other possibilities.
  36 	Distributed Factorization (qsieve just implemented)
  37 		DSAGE (8 processors) factored 11^150+1 in 2 minutes, SAGE factored it in 4 hours
  38 
  39 It's actually easy to write this yourself (see slides)
  40 
  41 The number of workers is dynamic.  Jobs cannot be sent to specific workers--they're just sent to the server, and the server passes them around.
  42 
  43 Since SAGE is based upon Python, makes use of Python's pickling to pass information from server to workers and back.
  44 
  45 Future Development
  46 	Tracking and scoring of users (similar to SETI@home)
  47 	Integrity checking schemes (especially an allowance for redundancy)
  48 	Run workers SETI@home-style
  49 Has been included in SAGE since version 1.7
  50 Latest dsage is version 0.2