Sigmod Demo - 03/03/1999

1. What is the status today?
2. What are the plans for the demo and the related issues?
3. Initial TODO list
4. Calendar

What is the Status Today?

What are the Plans for the Demo?

• Use a financial application as a common scenario for the different components of our demo.
  Roughly, the schema of a financial application contains the following relations:
  COMPANIES(Name, Ticker, WWWAddress, LastAnnualReport)
  INCOME(Ticker, IncomeStatementTS)
  STOCKS(Ticker, Year, Month, Low, High, Volume)
  PERSONEL(Ticker, EmployeeNumberTS)
  
  Issues: Obtaining the data and loading them into Predator
  Options: (1) UNC data for STOCKS, (2) Hoover's data for COMPANIES, INCOME, PERSONEL, (3) data obtained by Tobias at the Johnson school library
  
  Remark: selection of financial services on the Web
  http://www.stockmaster.com
  http://www.hoovers.com
  http://www.amex.com/floor/Resources/resource.htm (gives a list of financial links)
  http://quote.yahoo.com
  
  
• Tobias scenario:
  a - standard functionality: display company characteristics over last years
  b - define your own simple algorithm to analyze time series (Java UDF)
       (e.g.number of employees grows as the income statement)
  c - obtain sample of a typical query
  d - test Java UDF on client
  e - upload to server
  f  - execute the query with UDF (annual report of all companies whose number of employees grows as the income statement)/
       show execution plan on server (QPDisp) /
       show execution monitoring /
       show analysis of time series using Gui
  g - new function to analyze time series using local resources (my own portfolio)
       The problem is now to find the annual reports of the companies in my portfolio for whose number of employees grows
       as the income statement)
  h - first possibility is to postprocess the result of the previous query (we can deduce the execution time)
  i - second possibility is to integrate client UDF in query processing
  j - write java function / compile it
  k - register client UDF to the server
  l - for each execution strategy (server stub / client join / semi join)
        execute the query with client UDF /
        show execution plan on server (QPDisp) /
        show execution monitoring /
        show analysis of time series using Gui
  
  Issues:
  

  	Core Predator	Gui
  Tobias (Victor)	Time Series ADT Use of cost model for CSUDFs block optimization to determine the position of CSUDFs in execution plan	show execution monitoring refine Display of Time series integration of CSUDFs in QPDisp

• Zhiyuan's scenario:
  a - show memory usage on Windows CE device
  b - GUI on Windows CE / express query
  c - run query on server without compression | too large to fit on device
  d - run query on server with compression LZ77 | you can transfer | large footprint | show decompression cost
  e - run query on server with compression framework | you can transfer |  improvement on footprint | show decompression cost
  f - visualize compression plan

  Issues:

  	Core Predator	Gui
  Zhyuan (Kyle / Ryan?)	Compression optimizer (using stats)	Gui for expressing queries and visualizing results Means to visualize decompression cost Means to visualize compression plan

  A main issue in Zhyuan's scenario is the query which is asked.  It should (a) make sense and (b) be illustrative of the compression techniques.We can reuse the financial application and consider the following query:
  select Companies.name, Companies.WWWAdress, Stocks.year, Stocks.month, Stocks.low, Stocks.high
  from Companies, Stocks
  where Companies.ticker = Stocks.ticker
  

• Takako's scenario:
  a - run a query
  b - show execution plan / show discrepancy between estimates and actual numbers for cardinality and selectivity of different nodes
  c - feedback actual numbers in intermediate steps of query optimization
  d - run optimization from intermediate step and execute query
  f - show new execution plan / show estimates and actual numbers
  
  Issues:
  

  		Gui
  Takako (Ivan)	Feedback actual numbers in intermediate steps of query optimization + run optimizer from intermediate steps	Display estimates in compact form with QPDisp

  We have to define the query used for this scenario.

• Devices
  
  No concrete plans for devices yet.
  
  Remark:
  (1) Ryan's Gui can be used for the compression demo
  (2) Dan should be able to open the door with the Java Ring for March 25th.
  
  Options:
  (1) We can use the Java Ring for authentification within the financial application demo: java ring is needed to access a particular ressource. There should be no big implementation issue once the JavaRing ADT for opening the door is implemented.
  (2) Hubert & Randalph demo. In a few weeks, they should be able to capture the state of a PC application (IE ou WordPad), store it on the device and restore it on another PC.
  (3) to be defined
  
  

Initial TODO List

• Final version of demo paper
• Time Series ADT in Predator (Victor)
• CSUDF operator in execution plan (Tobias)
• cost model for CSUDFs (Tobias)
• Gui to display Time Series (Victor)
• Compression optimizer (Zhiyuan)
• Windows CE Gui for Compression (Kyle/Ryan?)
• QPDisp to display compact estimates and actual numbers (Ivan)
• Feedback of actual numbers in intermediate steps of query optimization (Takako)
• Define schema and load data into Predator (Philippe)
• JavaRing ADT for opening the door (Dan)

To be Defined:

• Common framework for the code to guarantee that all components fit together.
• Do we need a separate server for UDFs extensions?
• Which data are we using for the demo (UNC + hoover's data / Johnson school data)
• How to visualize query execution for CSUDFs ?
• How to visualize decompression cost?
• Experiment with queries for compression demo
• Which device stuff do we demonstrate? How?
• What is the target for March 25th?
• Timing for TODO list.

Calendar

March

April

May