The need for speed

How Artifactory became 10 times faster

Elliot Bentley

New version of binary management tool scraps Jackrabbit storage engine and reaps massive rewards.

takes a brave developer to release a major software update
featuring only improvements to performance. However, the developers
of Artifactory
– who are today releasing
version 3.0
– have achieved no mean feat in ramping up its
speed by a factor of ten.

Reportedly in use at LinkedIn, Netflix, Twitter,
Oracle, EMC2, Apple and VMware, Artifactory is a tool for managing
thirty-party binaries and libraries. Parent company JFrog, based in
Israel but now expanding to California, recently launched a
GitHub-like binary database with social features called Bintray. Having
opened to the public
in March, the company claims that it now
boasts almost 60,000 software packages.

However, that’s something of a side project:
Artifactory is JFrog’s flagship product, which is today receiving a
massive speed boost to their flagship product, Artifactory –
achieved by ripping out the existing Jackrabbit storage engine and
replacing it with a custom-made solution.

To eke out the performance improvements they wanted,
the development team had to “tune every SQL query”, JFrog Chief
Architect and co-founder Fred Simon told JAXenter. The company’s
engineers worked “like a game developer, when they go all the way
back to C”.

“We went very low-level, so people will say, ‘woah,
you’re crazy, why are you rewriting the wheel in terms of going so
low-level about data storage?’.” But despite “some doubt along the
way”, the eight-month project paid off. “To tell you the truth, we
surprised ourselves,” said Simon.

The benchmarks shared with JAXenter by JFrog support
these claims: with 50 users, Artifactory 3.0 is able to serve an
average of 1298 user requests per second – over ten times that of
current version Artifactory 2.6.7, which can only do 127 per
second. With 200 users, this difference extends to a 13x speed
boost. These improvements also mean that Artifactory requires less
disk space, CPU and memory.

If Jackrabbit was such a bottleneck, however, why did
they choose to use it in Artifactory in the first place?

“It was good for what we wanted to do [initially],”
responded Simon, “but once we saw that the model of the content
management was really stable, it was more of a burden than a help.”
With Artifactory becoming an increasingly mature piece of software,
the team were able to write a replacement for Jackrabbit tailored
to Artifactory’s needs.

“And the other thing is that we had enough resources
and time to decide that we need to push the performance of
Artifactory,” he added. “The main thing also is that there are lot
of other features we want to add to our software, and we knew that
if we kept adding features on top of the current storage of
Jackrabbit, we would have been pushing it to the limit.”

When asked if previous versions of Artifactory could
be considered slow, Simon drew a comparison with Twitter’s frequent
downtime in its early days. By moving away from “high level”
technologies and optimising their stack for specific use cases, he
said, both have achieved considerable performance improvements.

If it looks like a Black Duck…

The other big feature in Artifactory 3.0 is
integration with Black
, whose software analyses third-party libraries for
licensing issues. However, this is usually done after a build is
complete (or even sent to production), clashing with modern
continuous integration practises.

By integrating the two products, users can get the
“best of both worlds”, claimed Simon: the license-checking demanded
by the legal team and low-friction deployment demanded by the
developers. With Black Duck enabled, any binaries added to
Artifactory are automatically scanned, and the results sent to the
necessary party. Artifactory has included a license-checking
feature for a while, but – as Simon admits – it is “kind of simple”
in comparison.

“The process here is, for the developer, it’s easy,
and for the people that are watching [for licensing issues], it’s
immediate also. They are getting the real information about what’s
really built, as soon as it is built.”

This Black Duck integration is only available in the paid ‘Pro’
version of Artifactory, while the performance improvements will be
included in the open-source edition too.

Photo by Nathan

comments powered by Disqus