atlas-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apoorv Naik <an...@hortonworks.com>
Subject Re: Performance of relationships
Date Fri, 01 Jun 2018 20:48:00 GMT
Can you share the text of the benchmarks you performed ? My email is not rendering the attached
image.

From: Pierre Padovani <pierre.padovani@civitaslearning.com>
Reply-To: "dev@atlas.apache.org" <dev@atlas.apache.org>
Date: Friday, June 1, 2018 at 12:49 PM
To: "dev@atlas.apache.org" <dev@atlas.apache.org>
Subject: Re: Performance of relationships

I was in a hurry writing the previous email and realized that I did not specify that the relationship
type was COMPOSITION.

On Fri, Jun 1, 2018 at 2:37 PM, Pierre Padovani <pierre.padovani@civitaslearning.com<mailto:pierre.padovani@civitaslearning.com>>
wrote:
All,

I decided to spend a few cycles performing from basic performance tests to help us refine
our type system. The goal was to determine how we would add relationships to our entities
and if we would use the new relationship APIs or not. The test was simple. I created a type
called container and a type called child both extending DataSet. I think created an instance
of container, and added one child at a time to the container recording the cost of the add,
and the cost of fetching the container after each add.

There were three tests run:

- Legacy - Assign the entities to the attributes in the attribute map
- Hybrid - Leave the legacy attributes in the type system, and use the relationship APIs
- V2 - Remove the legacy attributes and use the relationship APIs

Here are the results for 100 children:

[cid:ii_jhwdc79w0_163bcd934880f531]

Is there a reason we are continuing to use legacy attributes in the type system(s) we are
shipping with Atlas? From a performance perspective this does not seem like a reasonable thing
to do.

Thanks!

Pierre


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message