[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [eclipse.org-committers] Release 3.1 is severely crippling eclipse.org
|
Jim des Rivieres wrote:
3.1 has already reached 95,000 downloads
I'm not sure where this particular number comes from (perhaps the Million
download challenge), but I doubt it's an accuracte reflection of actual
number of 3.1 downloads. Some simple math reveals it's bogus:
The download counter measures each time someone selects a mirror site
(or the main site) when accessing this link:
http://www.eclipse.org/downloads/download.php?file=/eclipse/downloads/drops/R-3.1-200506271435/eclipse-SDK-3.1-linux-gtk.tar.gz
(Linux shown, Windows and others similar, including eclipse Platform,
Eclipse RCP, and any other download that constitutes an "Eclipse Download").
Is it accurate? Hell no. How do I know if someone is cancelling a
download from a mirror site? How do I identify 20 clicks from one IP as
being one individual or simply 20 students in a classroom behind a
Proxy? How do I track bittorent feeds from the individuals who have set
them up? How do I count a downloads that go directly to mirrors as a
result of the numerous mirror site URLs posted here:
http://developers.slashdot.org/article.pl?sid=05/06/28/197256&tid=108&tid=156
How do you accurately measure downloads? Someone can easily write a PHP
script that loops and gets the file, but cancels after 1 bit has been
transferred. Someone can hold their finger on the F5 button. It's so
wildly inaccurate it's funny.
Apache logs? Impossible. We send out so many requests to mirrors.
Bandwidth? Impossible to differentiate download bandwidth without
setting up one single box to handle all the downloads. Hardly scalable.
Click Tracker? Close, and that's what's in place. What happens if
someone posts direct links to mirror sites, bypassing the click tracker?
Screwed. What happens if someone ballot-stuffs the click tracker? Screwed.
> While Eclipse 3.1 is certainly popular, we should ensure that the
numbers
> discussed in public are factual and accurate.
They are factual and accurate: eclipse.org has registered 162,500
downloads so far. If you downloaded it ten times, the number is still
factually correct. If you clicked 100 times and cancelled your download,
we still provided you the download 100 times. Please don't confuse one
download as representing one individual person.
Denis
Eclipse 3.1 SDK download for win32 (eclipse-SDK-3.1-win32.zip) is 105MB
--- call that 10**8 bytes
95000 downloads --- call that 10**5 downloads
3600 sec/hr or 86400 sec/day --- call that 10**5 sec/day
95000 downloads in 1 day would be 10**13 bytes in 10**5 sec, or 10**8
bytes/sec. That's 100 megabytes/second, which is a very large pipe indeed
(and I believe about 10x large than the current pipe to eclipse.org
download servers). So unless the 95000 number is somehow factoring in
downloads from mirrors, it's inflated. A 10 megabyte/sec connection could
physically serve up at most 1 download every 10 sec., to a limit of 10**4
downloads/day.
While Eclipse 3.1 is certainly popular, we should ensure that the numbers
discussed in public are factual and accurate.
---jim
"Eclipse Webmaster (Denis Roy)" <webmaster@xxxxxxxxxxx>
Sent by: eclipse.org-committers-bounces@xxxxxxxxxxx
06/29/2005 06:32 AM
Please respond to
Eclipse Committers
To
eclipse.org-committers@xxxxxxxxxxx
cc
"Sujay D'Souza" <sujay.dsouza@xxxxxxxxxxx>
Subject
[eclipse.org-committers] Release 3.1 is severely crippling eclipse.org
Hi All,
Even doubling our available bandwidth couldn't save us from the
popularity of Eclipse. Although the new hardware is holding up just
fine, we are simply overwhelmed by the amount of requests. We were
saturated at 2,200 concurrent connections, and we are currently at 4,320
- and have been at that rate for the last 8 hours.
I'll be focusing my energy on setting up a high capacity mirror site on
a separate Internet link so I can redirect download traffic to it. Until
then, I ask [beg] that you refrain from downloading 3.1 directly from
eclipse.org. Postponing any bandwidth-intensive eclipse.org operations
would certainly go a long way.
By the way, 3.1 has already reached 95,000 downloads, and the count
would likely be higher if our site was accessible.
I apologize for the inconvenience, and I will be setting forth some
recommendations to prevent this for R-3.x.x and R-4.x.
Denis
_______________________________________________
eclipse.org-committers mailing list
eclipse.org-committers@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/eclipse.org-committers
--
Eclipse WebMaster - webmaster@xxxxxxxxxxx
Questions? Consult the FAQ at http://www.eclipse.org/webmaster/faq.html
View my status at http://www.eclipse.org/webmaster/main.html
