Fred Schmidt wrote:
The latest z/Journal has a study by Microsoft comparing Windows against
the mainframe in terms of electrical power usage for CICS web-based
applications. It claims that Windows is many times more efficient.
You can find the PDF document at
http://www.zjournal.com/redir.cfm?rid=939
Comments?
Well, some of this has been stated by others:
They never ran on a real 2094-704. They assumed the transaction rate
based on using a specific test set of LSPR tests.
They assumed that somebody would actually run a distributed Intel based
Windows box a 84+% busy the majority of the time. Most Windows server
guys I talk to start looking at adding new servers once they get above
any extended time period running at 30-40% busy (extended being more
than 1 or 2 hours). This would mean 2-3 of the distributed servers to
run the same level of workload at the 30-40% busy level.
I don't know where the got the 5,221 watts from, because the paper they
reference showing the z9 power consumption showed 3,916 watts for a
2094-704.
They don't talk about what happens if you need to support a workload
greater than the test environment could handle, or greater than the
extrapolated z9 workload could handle. From the z/9 power paper a
2094-716 uses 5,920 and had 3x the throughput.
I doubt if the Prime system could be upgraded enough to handle 3x the
throughput, so you would need to add more distributed boxes and then add
equipment to do the load balancing between those servers. Then if they
are like most distrusted environments, 1 or 2 extra servers just in case
one of the other ones fail.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
