On Thursday, May 4, 2006, some of the PALNI staff, armed with stopwatches, fanned out to PALNI sites in the area to test response time of the WebOpac.  Staff traveled to DePauw, Butler and UIndy to do the testing.  Three others tested response time on keyword searches from the PALNI offices. The testers used a sample script of 96 actual user searches pulled from web server log files and each person was assigned a different starting point to avoid caching issues.  The searches were conducted between 1:30 pm and 3:30 pm, when the system is most active.

 The tests did not reveal any major surprises.  They do confirm, however, that some PALNI libraries are experiencing real OPAC response time problems.  In addition, the results suggest that

1. Except for sorting, keyword search response times do not seem to be related in any direct way to the number of hits, 
   
2. Searching the PALNI union catalog is faster than searching any individual library catalog,
3. PC problems can contribute to significantly slower response times,
   
4. Keyword search response times can vary significantly from one library to another.

 Attached is a small chart that shows, for each library, the number of searches that retrieved more than 250 records and did not require sorting, the average search time, the median search time, and the number of searches that took more than 3 seconds.  Also attached is a spreadsheet of all of the search terms and the results of the searches.

 Some information that should be considered as you are viewing these statistics:

1. We found, after the fact, that the computer used to search the DePauw OPAC from the central PALNI office had a problem that caused it to run slow, and this set of results is questionable.
2. Sorting occurs on searches that retrieve less than 250 records, and this does seem to slow things down.
3. All searchers ran the same 96 searches over the same 2 hour time period.
   

Addendum [ Vince Lucas - 5/23/2006 ]:

1. For comparison, duing the same 1:30-3:30pm time period on  May 4, response times on our standard performance "benchmark" Boolean search ranged from 2.02  to 2.17 seconds.  However, keep in mind that this  is a search of the PALNI Union Catalog that is run automatically every 15 minutes, and it is also a search that is not sorted.   As noted above, searches run here at the PALNI offices, Union Catalog searches, and searches that do not get sorted are all factors that seem to yield better OPAC response times.
2. Also for comparison, looking at timings of actual user keyword searches recorded in our web server log files for the same two-hour period on May 4, we have the following results:

These results include all keyword searches entered by PALNI users across all PALNI libraries during the two-hour test period.  Of course, response times recorded in web server log files do not include the time to transmit the results across the network, nor do they include the time required for the web browser to actually format and display the search results web page on a PC.