Very bad performance for large artifacts

There are a lot of variables between the two methods. The first is the protocol, filesystem vs http and another is the maven wagon being used. A better comparison would be to compare using apache httpd against nexus.

Even in that test, i would expect Nexus to be slightly slower since it has to check other things like security and hunt around the repos in the group (although memory caches make this very fast for the second lookup). The benefit to Nexus comes when multiple users are trying to get the same artifacts and from active management of the repositories.

For example, if that 5mb file comes from central, it might take 2mins to get into nexus the first time. Compared to that, 22seconds is pretty fast.

Once we find the file we want in a local hosted repo or the proxy cache, we hand it over to commons httpclient to stream out, i don't know how easy it would be to crank that up, but again i'd like to see how fast you see it with apache httpd to rule out protocol and wagon.

Hrm. I guess we need to look closer at some of those http client settings. Maybe there's a buffer size that needs to be increased.

Good news is there is some upside it seems to making it faster on the Nexus side. Bad news is we need to make it faster on the Nexus side.

;-)

I created a jira for this: http://issues.sonatype.org/browse/NEX...

Hi,

i could not reproduce the problem! I placed a 125MB file into my local storage, and downloaded it using curl. To have some "reference", i did the same with placing the same file into webapp folder, to completely "avoid" Nexus from the game and measure Jetty speed only. Here are my results:

Getting the big-file from Nexus repository

cstamas@Marvin tmp]$ curl -O http://localhost:8081/nexus/content/r...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 120M 100 120M 0 0 38.5M 0 0:00:03 0:00:03 --:--:-- 38.6M
[cstamas@Marvin tmp]$


Getting the big-file from Jetty directly

cstamas@Marvin tmp]$ curl -O http://localhost:8081/nexus/clip-2008...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 120M 100 120M 0 0 35.2M 0 0:00:03 0:00:03 --:--:-- 35.3M
[cstamas@Marvin tmp]$


The machine specs:
MacBook Pro / Mac OSX
Java 1.5

My results are actually about the same, and this was across a wireless network with a 21mb file:

brianf@penryn ~
$ curl -O http://192.168.101.21:8081/nexus/cont...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 21.4M 100 21.4M 0 0 2748k 0 0:00:07 0:00:07 --:--:-- 2762k

brianf@penryn ~
$ curl -O http://192.168.101.21:8081/nexus/test...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 21.4M 100 21.4M 0 0 2863k 0 0:00:07 0:00:07 --:--:-- 2860k

rschlege, what jdk are you using? It's possible that the performance difference is with the jdk vs native access.

.21 is an ubuntu jeos VM running nexus with just 128mb for the whole VM. The vm host is another machine and I ran the test from my laptop over wireless. If the problem doesn't exist when you access locally the something outside of nexus is your problem because you can see the app is capable of streaming much faster than you see. Perhaps a firewall, proxy or virus scanner is at play. I could see where those could behave differently for http than file... Particularly virus scan.