You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Support a mode in the s2sTesting module where only server-side bids will be made
Description
One of the greatest limitations of client-side integrations is its reliance on the browser to process many responses. To improve publisher's ability to measure the impact of server-side header bidding, we've been asked to support a scenarios where all bidders happen via the Server path.
The proposal is to add a new optional testServerOnly flag that integrates with PBJS core to suppress client-side bids when:
testing: true
at least one server-side bidder matches the A/B test percent
This should be implemented as a filter that runs after all existing server/client determination runs -- if at least one bidder runs server, but some other bidders were chosen as client-side for A/B testing, they will not be run. This scenario shouldn't happen with the recent 'global random' unless different client/server ratios are chosen.
When this mode is activated, there will never be a mix of client and server.
In this configuration, 90% of the time all bidders happen client-side. 10% of the time, both rubicon and appnexus go through the server AND any client-side bidders are ignored.
The goal here is to get an accurate view of the impact of s2s, and having the browser be busy with bidders that don't support server may be minimizing the perceived benefit of server.
If we find that using server saves the browser cycles and improves header bidding, other bidders will have incentive to build server-side adapters.
In reviewing with the Prebid.js committee, it came up that session-based sampling is the best practice. Prebid.js should not be setting cookies to support implementation of that feature, but it can be done by the publisher.
Assuming the pub has set up a session-sampling cookie and has stored the results in pbjs.isSampledSession, they could configure the test with:
Bidders A and B are in s2sConfig. Bidder A is 10% server and 90% client. Bidder B is 50%/50%
The random number chosen is 34. Therefore, only Bidder B will go server-side. A,C, and D are client side
the testServerOnly flag is on, so A,C,D are suppressed. Only B participates in the auction server-side.
The text was updated successfully, but these errors were encountered:
bretg
changed the title
Feature: s2sTesting supports suppressing client-based bids
Feature: s2sTesting support for suppressing client-based bids
May 22, 2019
@bretg is the converse true? if some bidders are server only (in bidders list but not in bidderControl), the testServerOnly flag is true, and s2sTesting returns client - should it do client only? Or in that case should there be a mix of client and server?
Type of issue
Feature
Summary
Support a mode in the s2sTesting module where only server-side bids will be made
Description
One of the greatest limitations of client-side integrations is its reliance on the browser to process many responses. To improve publisher's ability to measure the impact of server-side header bidding, we've been asked to support a scenarios where all bidders happen via the Server path.
The proposal is to add a new optional
testServerOnly
flag that integrates with PBJS core to suppress client-side bids when:This should be implemented as a filter that runs after all existing server/client determination runs -- if at least one bidder runs server, but some other bidders were chosen as client-side for A/B testing, they will not be run. This scenario shouldn't happen with the recent 'global random' unless different client/server ratios are chosen.
When this mode is activated, there will never be a mix of client and server.
e.g.
In this configuration, 90% of the time all bidders happen client-side. 10% of the time, both rubicon and appnexus go through the server AND any client-side bidders are ignored.
The goal here is to get an accurate view of the impact of s2s, and having the browser be busy with bidders that don't support server may be minimizing the perceived benefit of server.
If we find that using server saves the browser cycles and improves header bidding, other bidders will have incentive to build server-side adapters.
In reviewing with the Prebid.js committee, it came up that session-based sampling is the best practice. Prebid.js should not be setting cookies to support implementation of that feature, but it can be done by the publisher.
Assuming the pub has set up a session-sampling cookie and has stored the results in pbjs.isSampledSession, they could configure the test with:
Finally, an example:
The text was updated successfully, but these errors were encountered: