Amazon CloudFront / S3 Small Object Test Results Post: 302325537

Sponsored Content

Special Forums UNIX and Linux Applications Virtualization and Cloud Computing Amazon CloudFront / S3 Small Object Test Results Post 302325537 by Neo on Friday 22nd of May 2009 08:01:00 AM

05-22-2009

Administrator

Amazon CloudFront / S3 Small Object Test Results

After some initial confusing results measuring "before-and-after CloudFront/S3" performance, Dennis Opacki, Systems Operations Manager for Amazon Cloudfront offered to help. Working with the test results from a number of members of The UNIX and Linux Forums pulling a small 1623 byte gif object with curl -s -w, we compiled the test results. For each location, we discarded the fastest and slowest times for the Origin Web Server (in Ireland) and the CloudFront (CF) sites and averaged the remaining three samples. The last two columns in the table below represents the difference in performance between the objects hosted. The data seems to suggest a fairly significant performance improvement (with Amazon CloudFront) for all locations worldwide except Ireland, which is not surprising, since Ireland is the location of the original server.

.	Web Server (WS) in Ireland		CloudFront (CF)		Compare Performance (CF/WS)
Client Location	WS First-byte(s)	WS Total Transfer(s)	CF First-byte(s)	CF Total(s)	First-byte (%)	Total Transfer (%)
Ireland	0.096	0.097	0.101	0.101	6%	4%
Netherlands	0.123	0.124	0.081	0.081	-34%	-35%
St Paul, MN	0.472	0.474	0.284	0.286	-39%	-40%
Sofia, Bulgaria	0.134	0.134	0.081	0.081	-40%	-40%
Milan, Italy	0.111	0.112	0.042	0.042	-63%	-63%
Seattle, WA	0.515	0.516	0.133	0.134	-74%	-74%
Redwood City, CA	0.324	0.327	0.041	0.045	-86%	-86%
Ashburn, VA	0.188	0.189	0.022	0.022	-88%	-88%
Hong Kong	0.944	0.944	0.048	0.048	-95%	-95%

As expected, the most improvement was from users in the US and Asia (Hong Kong), since the original web server resides in Ireland.

I want to thank all the members of The Unix and Linux Forums who took the time to run the tests and for Dennis of Amazon for helping compile the results.

Congratulations Amazon Web Services and CloudFront!!

Note: Our test results suggest that static web content hosted on Amazon's CDN can certainly help global user response time downloading (small) objects. This is what you would expect from a CDN.

Neo

View Public Profile for Neo

Visit Neo's homepage!

Find all posts by Neo

3 More Discussions You Might Find Interesting

1. UNIX for Advanced & Expert Users

Printer Error(the Object Instance Test Does Not Exist)

Hello, i need some help about how to set up a high velocity impact printer in UNIX SCO 5.05, this printer is attached with a parallel port in a PC(host), the host use tunemul to access unix.(this reference is just to ask you if this is a local or remote connection, just to be sure), so, i...

2. Virtualization and Cloud Computing

CEP as a Service (CEPaaS) with MapReduce on Amazon EC2 and Amazon S3

Tim Bass 11-25-2008 01:02 PM Just as I was starting to worry that complex event processing community has been captured by RDBMS pirates off the coast of Somalia, I rediscovered a new core blackboard architecture component, Hadoop. Hadoop is a framework for building applications on large...

3. Shell Programming and Scripting

PERL - traverse sub directories and get test case results

Hello, I need help in creating a PERL script for parsing test result files to get the results (pass or fail). Each test case execution generates a directory with few files among which we are interested in .result file. Lets say Testing is home directory. If i executed 2 test cases. It will...

LEARN ABOUT XFREE86

x11perfcomp

X11PERFCOMP(1)						      General Commands Manual						    X11PERFCOMP(1)

NAME

       x11perfcomp - X11 server performance comparison program

SYNTAX

       x11perfcomp [ -r | -ro ] [ -l label_file ] files

DESCRIPTION

       The  x11perfcomp program merges the output of several x11perf(1) runs into a nice tabular format.  It takes the results in each file, fills
       in any missing test results if necessary, and for each test shows the objects/second rate of each server.  If invoked with the  -r  or  -ro
       options, it shows the relative performance of each server to the first server.

       Normally,  x11perfcomp  uses  the first file specified to determine which specific tests it should report on.  Some (non-DEC :) servers may
       fail to perform all tests.  In this case, x11perfcomp automatically substitutes in a rate of 0.0  objects/second.   Since  the  first  file
       determines  which  tests  to  report  on, this file must contain a superset of the tests reported in the other files, else x11perfcomp will
       fail.

       You can provide an explicit list of tests to report on by using the -l switch to specify a file of labels.  You can create a label file	by
       using the -label option in x11perf.

OPTIONS

       x11perfcomp accepts the options listed below:

       -r	       Specifies that the output should also include relative server performance.

       -ro	       Specifies that the output should include only relative server performance.

       -l label_file   Specifies a label file to use.

X DEFAULTS

       There are no X defaults used by this program.

SEE ALSO

       X(7), x11perf(1)

AUTHORS

       Mark Moraes wrote the original scripts to compare servers.
       Joel McCormack just munged them together a bit.

XFree86 							   Version 4.7.0						    X11PERFCOMP(1)