home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.umcs.maine.edu
/
2015-02-07.ftp.umcs.maine.edu.tar
/
ftp.umcs.maine.edu
/
pub
/
WISR
/
wisr6
/
proceedings
/
ascii
/
jackowitz.ascii
< prev
next >
Wrap
Text File
|
1993-10-19
|
8KB
|
167 lines
Simple, Unobtrusive Metrics For ASSET's Software Reuse Library
Larry A. Jacowitz
IBM Federal Systems Company
2611 Cranberry Sw.
Morgantown, WV 26505
Tel: (304) 594-1762
Email: jacowitz@source.asset.com
David Irving
SAIC
2611 Cranb erry Sq.
Morgantown, WV 26505
Tel: (304) 594-1762
Email: irvingd@source.asset.com
Abstract
Current experience at ASSET provides insight into the way users navigate through a software
reuse library. Simple, unobtrusive metrics are described for measuring library effectiveness and
library efficiency. These metrics are used to analyze typical library usage.
Keywords: metrics, reuse libraries, networks
Jackowitz- 1
1 Background
The evolution of a network of collaborating software reuse libraries implies the need for a common
set of metrics relating to user fulfilment. These libraries include domain-specific libraries at several
levels (project, department, enterprise) and non-domain specific (or reference) libraries. A wide
range of subject matter, user communities, and library mechanisms are represented. Yet it is
highly desirable to seek a common set of metrics to be applied across the board. Such a common
set will necessarily be a "lowest common denominator" and simplistic in nature. However, this
simplicity is advantageous because it enables one to easily gauge how well each library is serving
its users. It is general practice to employ user surveys to provide this measure. It is obvious that
this may be perceived as highly intrusive to the user and it is labor intensive for the library staff.
Response rates are less than 100Therefore, it is desirable to have an alternative approach, such as
a preliminary assessment made by tracking the user's search and retrieval path within the library
mechanism without contacting the user. The results of analyses to date at ASSET show promise
for this approach.
2 Position
When a user enters his software reuse library of choice he will follow a typical sequence of commands
which will permit him to identify the parameters of his problem, perform a search of the library,
browse candidate assets, and extractthose which meet his criteria. By tracking the user's actions in
this process one can synthesize the user's objective and determine the degree of success in achieving
that objective. Two indicators are sought: first, the effectiveness of the library inmeeting the user's
requirements; second, the efficiency with which the library does this. Define an extraction ratio,
ER = number of user extractions per search; this is an indicator of the overall effectiveness of
the library. Next, define an extraction index, EI = ratio of the number of user extractions to the
number of search candidates found by the library mechanism; this is an indicator of the efficiency
with which the mechanism finds candidates. There are other possible intermediate parameters
which can add insight to the process, notably the ratio of browses to searches or extractions.
When this approach was applied to ASSET's user activity over a four-month period certain modal-
ities became apparent. The typical user search is an iterative process;the user may be unsuccessful
on the first try in matching his needs with the search mechanism schema. In this case the typical
user will converge on a matching path within two or three tries. It has b een noted that this same
user learns from the experience and converges more quickly on subsequent searches. An often cho-
sen alternative approach is to scan the catalog headings and call out specific candidates by their
unique library identification number. It is certainly true that the extraction index and extraction
ratio are a function of the classification scheme, ease of use, search mechanism peculiarities, extent
of the library's holdings, and so forth. However, the simple metric technique has merit for the
reasons stated.
Analysis of the cumululative statistics of ASSETusage since inception of this approach shows some
expected, but some unexpected results. First, the ER, thenumber of extractions per search on a
monthly basis has been quite consistently within a band of 0.9 to 1.5, with a cumulative average of
1.3. So users have been finding somewhat more than one component per search. Second, EI, the
ratio of extractions to candidates turned up by the search, varies over a wide range. Results are
erratic and not statistically significant. This is because at one extreme users may invoke the entire
catalog (hundreds of candidates) just to browse through it; at the other extremea user extracts
Jackowitz- 2
a component each time he calls for a known component by identification number. Third, we have
found that the ratio of searches:browses:extractions tends to be stable from month to month about
roughly 1:3:1
3 Comparison
Traditionally user surveys are the only means of determining the effectiveness of a library to meet
user needs. This method is time- consuming for the library staff and obtrusive to the user. Often
low response rates pertain. An indirect approach is desirable.
The approach described in this paper offers an alternative which is simple and unobtrusive. In
addition to providing information about the users' functional requirements, this technique is useful
in evaluating the ease of use of the library, the effectiveness of the classification scheme, etc.
This technique should be extended to other software reuse libraries. Comparative analysis of
different, heterogeneous libraries may provide ideas for improving the state of practice in software
reuse libraries.
4 Biography
Larry Jacowitz is the ASSET Director for IBM's Federal Systems Company, the prime contractor
for ASSET. The ASSET facility is located in Morgantown, WV, and is implemented under ARPA's
STARS program. Dr. Jacowitz previously held management positions in a variety of federal civil
programs, including DOE, Dept. of Agriculture and Immigration &Naturalization Service, and
NASA's space shuttle and space station programs at locations in Gaithersburg, MD, Downey, CA,
Owego, NY, and Huntsville, AL. Heholds BS, MS and PhD. degrees in engineering from MIT and
OSU.
David Irving is a Senior Software Engineer at ASSET. He is responsible for the maintenance and
improvement of the software reuse library mechanism. He also developed the tracking technique
described in this paper. Mr. Irving's previous assignments at SAIC have b een in the area of Ada
code development and IV&V on projects for DOE and missile projects for the Navy. Mr. Irving
holds a BA degree in mathematics from Univ. of Pennsylvania and M.S. in physics from Univ. of
Tennessee.:
Jackowitz- 3