A few first impressions...

I would have liked to see part of the test include multiple LPARs running
Linux, as compared to the same number of guests running on z/VM.  It would
have given some data to compare ease of management versus performance
tradeoffs (if any exist).

While avoiding customized kernels is probably realistic for the general
users of Linux/390, also avoiding adjusting kernel "constants" through the
/proc file system is not.
They allocated way too much storage to the Linux/390 guests.

They actively de-tuned the z/VM system by turning on QDSP for all the Linux
guests.  That's unforgivable.

The way they talk, they didn't adjust the z/VM SRM values for
over-committing resources for better performance.

They talk about "mainframe Linux experts" doing the tests, but don't
identify them.  This community is small enough that anyone who is
experienced enough to be considered an "expert" would likely be known by
name.  I would be interested in knowing who they are.

I'm not an expert at picking apart performance test results, so I can't tell
if these tests were particularly revealing of anything or not.  They're
enough to make me wish for a similar report from someone I could trust to
not have a hidden agenda.  I would love to see the tests repeated, and the
results published, by someone that knows what they're doing for each
platform, and preferably not funded by either MS or IBM.


Mark Post

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
Sent: Sunday, October 05, 2003 6:13 AM
To: [EMAIL PROTECTED]
Subject: benchmark comparison of l390-zVM vs win 2003


Here's a chance for IBM and others to comment on some benchmark experience:

http://download.microsoft.com/download/7/3/e/73e77129-db34-4c95-b182-ab0b9bd
50081/MainframeBenchmarkProj.pdf

Reply via email to