[ QUOTE ]
<font class="small">Code:</font><hr /><pre>
CPU Resource Differences - Upgrade Factor
EnterpriseOne| iSeries| pSeries| xSeries
Workload
Database| 1.41| 2.33| 2.8|
Application| 1.28| 1.18| 1.42|
Batch| 1.34| 1.62| 1.79|
HTML| 1.79| 2.25| 1.91|
Virtual 3-tier| 1.34| 1.62| 1.79|
Three-in-one| 1.48| 1.76| 1.82|
</pre><hr />
[/ QUOTE ]
Where did you get this graph from ? I'm a little blown away by it - and would love to understand the graph a little better - wow - did Oracle actually do some benchmarks on 8.11 ?!?! This is the very first comparison between AS/400, Unix and Intel I've ever seen.
Lets look into these a little more deeply.
According to the CPU resource differences, running the 8.11 database on an iSeries is substantially more efficient than on Unix or Intel - I assume the comparison is versus DB2 across all platforms to be fair. On Intel, UDB is twice a hog as it is on the iSeries.
The OneWorld Appserver code, however, seems to run more efficiently under Unix than it does under AS/400 - which I can understand, since much of the Appserver code has always been initially tested on HP-UX prior to compatibility testing on other platforms. I don't understand why Batch should be dramatically more efficient however, since the code is identical. This doesn't make sense.
HTML shows that Websphere and the Java code runs more efficiently on the iSeries - substantially quicker than on the Unix platform (again, I've seen this on the Proprietary unix platforms) - but I'm surprised that its more efficient than Intel. My guess is the "tweaks" that IBM put in the code have helped it along with 8.11.
Virtual 3-Tier, in my opinion, means that the application server and the database runs on the enterprise server, and the JAS server runs seperately. For some reason, these figures are identical to Batch running alone - so I might be wrong in assuming what "virtual 3-tier" is.
Lastly, we have "everything in one" - ie the database, application code AND the JAS Server running together. According to IBM, the iSeries trumps the other platforms.
Of course, we don't have the baseline for our examples. Obviously, the numbers mean that the database load between Xe and 8.11 requires 2.8 times the CPU load on an intel server comparitively. With intel servers performing substantially faster than when Xe was released, I'm not sure this a big issue - but are iSeries platforms running 4-5 times faster between 2000 and today ?