home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.engr
- Path: sparky!uunet!gatech!usenet.ins.cwru.edu!eagle!venus.lerc.nasa.gov!ecaxron
- From: ecaxron@venus.lerc.nasa.gov (Ron Graham)
- Subject: Question about CPU internal clock
- Message-ID: <31DEC199212262168@venus.lerc.nasa.gov>
- News-Software: VAX/VMS VNEWS 1.41
- Sender: news@eagle.lerc.nasa.gov
- Nntp-Posting-Host: venus.lerc.nasa.gov
- Organization: NASA Lewis Research Center
- Date: 31 Dec 1992 12:26 EDT
- Lines: 30
-
- Since this is a computer-related question, I can understand if this
- question would be better asked elsewhere, so pointers to where it belongs
- are welcomed.
-
- We have an application where we are trying to estimate the order of
- a certain algorithm (whether O(N), O(N^2), etc.) by running it a certain
- number of times for various values of N. We measure the CPU time used
- by the algorithm using the FORTRAN "SECNDS" command, executing it just
- before and just after the algorithm runs through its paces.
-
- What we are finding is that the CPU measure we get back varies wildly,
- even for large values of N (~250) and for large numbers of runs (~30).
- The questions are as follows:
-
- (1) Are there aspects of the SECNDS command we don't know about, such as
-
- o sensitivity to system functions (user load, time of day, etc.)
- o inherent uncertainty
-
- (2) Is there a more reliable way to estimate the order of the algorithm?
-
- As you have guessed, we're running in FORTRAN, we're on a shared system,
- and we're not taking advantage of parallel processing or any other advanced
- programming tool.
-
- Conscientious replies welcomed, will summarize if there is interest.
-
- RG
-
- Who became an engineer thinking "free-body diagrams" might be interesting.
-