• An INDEPENDENT TEST
  A customer's quick test showed ESAWEB to be 28 times faster loading a slightly complex page on a recent Amdahl processor. Only one significant digit is needed here!

  Eweb:   1.68 CPU seconds                                                   
  ESAWEB: 0.06 CPU seconds

• Another INDEPENDENT TEST:
  I'm running a version of my Java ticker with nil program logic except to generate a time stamp: eg: result = translate(diag(8,'QUERY TIME'),' ','15'x) 'OUTPUT' result
  
  On ESAWEB - 10 hits costs 0.03 or 0.04 CPU seconds TTIME (as reported by QUERY TIME)
  On EnterpriseWeb - 10 hits costs 0.80 CPU seconds
  Even using the higher time for ESAWEB for round-up. that's a 20:1 ratio.
  
  
• And the internal measurements:
  I ran the benchmarks Saturday morning. I let about 5 minutes lapse between each run to allow the system to quiet down a bit.
  I ran three different benchmarks from SFS then from minidisk for a total of six benchmarks. The first type was retrieval of HTML requiring EBCDIC to ASCII translation and appending CR/LF to each output record. The second type was of a GIF file requiring no server massaging of data. The third type was the execution of a CGI that used a PIPE state to read the test HTML file and 'OUTPUT' it.
  Each test generated 8304 bytes of header and 8196 bytes of output data per request.
  Results according to benchmark program:

                                                                             
  Time    Duration  Filetype  Storage   hits                                      
  11:54    240.008  HTML      Mdisk     1849 (first run of day)                   
  12:04    240.118  GIF       Mdisk     1753                                      
  12:17    240.047  GIF       SFS       1693                                      
  12:26    240.387  HTML      SFS       1529                                      
  12:52    240.042  CGI       SFS       1383                                      
  13:03    240.017  CGI       Mdisk     1363                                      
  

  I would be interested in seeing how many CGI cached hits there are relative to CGIs invoked. I had a couple of runs where the CGI failed and had to be modified.
  

If you can't measure it, I'm just NOT interested!(TM)