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
I'm curious about this idea in the short run just for consistency and hard bounding, but especially so in the long run if we hope to one day move mission critical operations onto Aggregated Reporting, such as billing, budgeting, etc, and adopt something like the extended private agg reporting. I do see one issue with it but wanted y'alls thoughts.
Currently the browser controls and executes the delay for sending a report to the agg report TEE. Why not continue to let the browser control the delay (i.e. choose the time in seconds/minutes) but send it immediately to an additional TEE that has one job, to hang on to the report and send it out after that number of seconds. You could make it optional so that the ad tech could choose to accept the lack of bounding rather than additional TEE cost. In the case of Attribution in particular the event count is likely to be small (our attributions are 2-3 orders of magnitude smaller than our request counts).
I'd imagine one issue is that you'd in theory need to store the received reports-to-delay in a durable fashion somewhere, and storing anything opaque durably adds risk. We'd have to do some smart data partitioning and replication within a cluster, but that's doable, and if the max retention was 10 minutes + some reasonable recovery factor, I'd think it would be helpful.
Curious for your thoughts?
The text was updated successfully, but these errors were encountered:
In general, I like the idea of reducing delays via introducing server mediation. I'd need to think a bit harder about the TEE acting in this role but it's definitely something we can consider.
I'm curious about this idea in the short run just for consistency and hard bounding, but especially so in the long run if we hope to one day move mission critical operations onto Aggregated Reporting, such as billing, budgeting, etc, and adopt something like the extended private agg reporting. I do see one issue with it but wanted y'alls thoughts.
Currently the browser controls and executes the delay for sending a report to the agg report TEE. Why not continue to let the browser control the delay (i.e. choose the time in seconds/minutes) but send it immediately to an additional TEE that has one job, to hang on to the report and send it out after that number of seconds. You could make it optional so that the ad tech could choose to accept the lack of bounding rather than additional TEE cost. In the case of Attribution in particular the event count is likely to be small (our attributions are 2-3 orders of magnitude smaller than our request counts).
I'd imagine one issue is that you'd in theory need to store the received reports-to-delay in a durable fashion somewhere, and storing anything opaque durably adds risk. We'd have to do some smart data partitioning and replication within a cluster, but that's doable, and if the max retention was 10 minutes + some reasonable recovery factor, I'd think it would be helpful.
Curious for your thoughts?
The text was updated successfully, but these errors were encountered: