More Download Statistics for GlassFish
By pelegri on Sep 07, 2008
This is an update to GlassFish Download Statistics from March 6th, 2008. It has an intro - mostly copied from the March entry, some minor corrections to it, and then the new charts.
Added (Sept 25, 2008) - It seems we found the problem with the NB d/l numbers. We are using a CDN to meet the strong demand for NB 6.1 and it seems the script substantially undercounted donwloads in that case. I'll recreate charts when/if the new numbers are available, sorry, no estimated time for now.
Readers of TheAquarium will have seen additional adoption indicators like the Geo Mashup based on GF Admin Console pings (live, geomap@TA), informal Adoption Stories, posts in user lists, and Other Anecdotal Data. All indicators have limitations; my usual comment is that if all the indicators are pointing the same direction, they are probably right.
• The download numbers quoted below are based on completed (not just attempted) downloads.
• We only count downloads from Sun.Com (under the Sun Java System Application Server name), from Java.Net (GlassFish AppServer) and NetBeans.ORG (NetBeans IDE).
• We are not counting downloads from our Maven Repository.
• We do not count Ubuntu, Solaris, Solaris Express nor OpenSolaris distributions.
• GlassFish is freely redistributable and we don't track other distributions.
• I believe these are solid numbers, but if we discover a problem I will let you know.
Below are several charts with commentary. The charts show monthly downloads starting in July 2005, right after we launched the GlassFish project, and going through last month, August 2008. The Y axis is normalized to 600K per month to simplify comparisons. The images link to a larger image for further inspection.
Download numbers go down during the (northern hemisphere) summer and at the end of the (gregorian) year. They also tend to peak during JavaOne, in May, and right after new releases.
Corrections and Known Issues
Correction - The graphs/data reported last time suffered from some double-counting in Dec07/Jan08 and the charts below show the adjusted numbers. The overall shape of the graphs only changes slightly.
Known Issues - The charts below show a sharp, unexplained, drop in downloads for NB 6.x during Jun 08 and Aug 08. We believe we have found the Aug 08 problem (added some servers but the scripts are not yet searching their logs) but we are still puzzled about the Jun 08 drop (found? - see comment at the top of this entry). I had been waiting for clean data to publish the numbers but I've decided to push it out now; I'll generate new charts when the data for Aug 08 comes, and will let you know if we ever figure out the Jun 08 issue.
As before, there are 3 data sets, collected into 5 charts.
GlassFish AppServer in SDKs
This chart shows downloads of the GlassFish AppServer either by itself (SJS AS 9.x and GlassFish AS) or in bundles like the Java EE SDK and the Java Application Platform SDK (Main Download Page).
The peak around May '06 (223,079) was the release of Java EE 5 and GFv1 at J1 '06. Other local peaks correspond to GFv1U1, J1'07, GFv2, GFv2U1 and GFv2U2.
D/l numbers for Jul08 and Aug08 are 149,435 and 147,575. There is always a d/l lull during the summer, and GFv3 is in a month, so these numbers are as expected.
GlassFish AppServer in Tools
GlassFish AppServer - Combined
This table just adds the previous two. The combined number for Jul08 and Aug08 are 291,888 and 229,986. I will also update this chart when/if I get corrected NB data for these months.
JBoss AppServer - From SourceForge
Finally, as a reference point, here is the chart from the Source Forge JBoss AS Download Stats. The d/l numbers for Jul08 and Aug08 are 67,046 and 71,952. Since there might be other bundles not at SF, pointers to additional data would be welcome.
Note - If anybody knows how SourceForge counts downloads (attempted?, completed?), please let me know. Completion ratios vary based on geography and bundle size; for example, GFv2 currently has a completion rate between 60% and 70%. Added - it seems that 10-15% of the d/l #s are for source bundles or for MD5/SHA-256 files.