Friday, June 26, 2009

Galileo should go down ...

... as being the smoothest release I have ever participated in, despite the major constraint I had this year:  no extra bandwidth shall be purchased.

Here is how we pulled this off:

1. More bandwidth. Since we got sooo many new Friends of Eclipse, I bent the rules and added 12 megabits of bandwidth [1]. But -- get this -- for the pre-release only.  Marker (1) shows the increase, and marker (2) shows a significant drop three hours into the release.  Unheard of!



2. Early access. Friends of Eclipse, mirror sites and Member companies all had early access to the bits, relieving some of the stress on release day.

3. BitTorrent. In April, I didn't have options (4) and (5) below, so with the no-extra-bandwidth rule, I had to set up something.  Actually, I suspect many Friends took advantage of their benefits by downloading and pre-seeding many of the files before release day.  More seeds == faster downloads.

4. Eclipse Member companies. They provided lots of bandwidth (and still are!) via the Get It Faster section of the download page.  Many thanks!

5. Amazon Web Services. This was a last-minute deal that was/is a true blessing, since I was able to redirect lots of the download.eclipse.org requests to AWS instead of our stressed mirror sites.

6. Server virtualization. Our own version of 'the Cloud on demand' -- I enabled an extra virtual server node for www.eclipse.org on June 23 to handle the extra CPU power that would be required to handle the many requests.  I may just shut it off later next week, or I may just keep it around...

I think we're getting good at these annual release trains..

[1] In reality, we get billed by the 95th percentile, so if I played my cards right, the extra bandwidth may end up costing us next to nothing.

1 Comments:

Anonymous maarten meijer said...

Goof to hear Denis,

I wonder: How much of that traffic is now gzipped? I've done gzip for Mylyn and for ECF used by p2, but are there any significant types of server request left for further optimization?

9:41 PM  

Post a Comment

<< Home