Re: [logs] Data Reduction Strategies?

From: Sweth Chandramouli (svcat_private)
Date: Sat Sep 29 2001 - 23:48:16 PDT

  • Next message: jamie rishaw: "Re: [logs] Windows based Monitoring Tool"

    On Fri, Sep 28, 2001 at 10:11:56PM -0400, emf wrote:
    > So... What kinds of data reduction strategies are people using for their
    > logs?    Lets say we have a database full of various types of events (IDS
    > events, syslog messages, pacct data, and other assorted whatnot)  what kinds
    > of things are people doing to collapse this stuff over time and retain 
    > meaning, or are we all doomed to racks of disk servers and sql queries that
    > take ever-longer to complete?
    	The answer to this depends on what you want to do with
    the data later.  If it's just trending, and you know the variables on
    which you want to trend, rrdtool-style summarization is a very good
    solution.  For anamoly training, your data reduction is going to be
    directly dependent on the training methods you use.  (Terran Lane at
    CERIAS has written some really good papers discussing some of the issues
    with data reduction for training sets, that should be available at
    CiteSeer; there are also some other good papers in the astronomy
    community that can be extrapolated to this type of use.)  For being able
    to go back and look for anything that you missed, unfortunately, you
    really can't do much reduction--you can just move the data offline and
    compress the heck out of it until you need to look at a particular set;
    if you can define an event model that makes sense for your environment,
    you could instead move the historical data to an OLAP-style DB, which
    may or may not save you space (depending on how bad your denormalization
    gets).
    
    > Is this a problem someone's solved in a generic enough way that there's 
    > stuff I can just go download and install, or are we all still cobbling 
    > together our own little sets of widgets?
    	I'm not aware of any product that deals with this in this
    context; there are a few tools I've read about (but not used) for
    reduction of astronomy data, but they don't really seem like they are
    flexible enough to be kluged into working on log data as such (unless
    someone can think of a way to map log data into a FITS export file).
    <blatant plug>There are also a couple of companies that provide services
    like this, including mine; I'm in the process of setting up a
    partnership with another company that might lead to our writing some
    commercial apps to deal with things like this.  If there's any interest
    in such an app, please contact me off-list.</blatant plug>
    
    	-- Sweth.
    
    -- 
    Sweth Chandramouli ; <svcat_private>
    President, Idiopathic Systems Consulting
    
    
    



    This archive was generated by hypermail 2b30 : Sun Sep 30 2001 - 10:54:30 PDT