Hi Edwin,
Regarding your question about the # Sim Req setting in the tests, the
results table on Page 12 simply documents the *optimal* setting on each
version of the product. We tested all settings to find the optimal values
for CF4.5, then re-tested to find the optimal values for CF5. Because CF5
has been so significantly tuned internally, the # Sim Req that was optimal
for your CF4.x server is no longer the optimal value for your CF5 server.
You'll need to re-tune to achieve the absolute maximum performance out of
CF5. You'll see significant benefit over CF4.x just by upgrading, but to
really make things scream, tweaking the # Sim Requests under your typical
max load will get you some extra mileage.
As the brief shows, the most dramatic results (5X and 4.5X) were achieved on
configurations where CF4.x performed the absolute *worst*: on machines with
more than one processor running Windows 2000 and Linux.
We spent several months analyzing and tuning the internals, brought in top
hired guns from Microsoft at one point in the case of Windows 2000-specific
tuning, and our discoveries actually resulted in several Windows 2000
performance bottlenecks & tweaks that should be rolled into future Microsoft
Windows 2000 Service Packs at some point. It was a very fruitful exercise,
as you can tell from the results in the CF5 Performance Brief.
I would encourage all customers to spend a few minutes/hours re-tuning the #
Sim Req after upgrading to CF5 if their servers experience busy periods to
ensure they're getting the maximum possible scalability CF5 is capable of
(and it's nigh-and-day compared to the old CF4.x internals, believe me).
Results from our test application won't be as interesting to you as results
from testing your application on CF4.x vs CF5, obviously. Like we did,
however, be sure to tune the # Sim Requests to be sure you're getting
maximum performance in both cases.
Hope that helps. Since I'm heavily interested in ensuring customers get the
maximum performance & scalability benefit out of CF5, feel free to email me
directly if you have any other questions about the performance & scalability
of CF5. I'll be doing the CF5 Windows Performance Tuning talk at the User
Conference as well, if you'll be there and want to discuss "what happened"
to CF5 internally further.
- Damon
-----Original Message-----
From: Edward Smith [mailto:[EMAIL PROTECTED]]
Sent: Monday, July 16, 2001 10:08 AM
To: CF-Talk
Subject: Re: Review: CF 5.0
I'm of the opinion that Macromedia cooked the benchmark results to get
what they wanted. Either that, or I'm reading them wrong. Macromedia's
benchmark document is here:
http://www.macromedia.com/software/coldfusion/productinfo/performance_brief/
cf5_perf_brief.pdf
Please check out page 12 of the PDF.
In the table of settings, note that Macromedia set the 4.5 server to 1
thread per cpu. This is contrary to their own 4.5 tuning specs, and
certainly nothing any of us would do on a production server
('simultaneous requests' in the table). Am I reading this wrong? On
the 5.0 server, they set to 4 threads per cpu.
I find it hard to believe that they would skew their testing so
obviously, but I can't find any other conclusion.
Comments?
Archives: http://www.mail-archive.com/[email protected]/
Unsubscribe: http://www.houseoffusion.com/index.cfm?sidebar=lists
