Serving optimization in a large-scale serving system

1 . A computing system, comprising:

one or more processors;

a plurality of serving servers, the plurality of serving servers including a first server and a second server;

one or more non-transitory computer-readable media that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations comprising:

classifying a first content item with a first-tier label based on a performance attribute of the first content item;

classifying a second content item with a second-tier label based on a performance attribute of the second content item, wherein the performance attribute of the second content item is less than a threshold value;

classifying a third content item with the second-tier label based on a performance attribute of the third content item, wherein the performance attribute of the third content item is less than the threshold value;

storing the first content item and the second content item in the first server; and

storing the first content item and the third content item in the second server.

2 . The computing system of claim 1 , the operations further comprising:

classifying a fourth content item with a first tier-label based on a performance attribute of the fourth content item, the performance attribute of the fourth content item being above the threshold value; and

storing the fourth content item in the first server and the second server.

3 . The computing system of claim 1 , wherein the plurality of servers includes a third server, the operations further comprising:

classifying a fifth content item with a second tier-label based on a performance attribute of the fifth content item being below the threshold value; and

storing the fifth content item in the third server.

4 . The computing system of claim 1 , the operations further comprising:

a plurality of first-tier content items, the plurality of first-tier content items comprising of content items classified with the first-tier label;

a plurality of second-tier content items, the plurality of second-tier content items comprising of content items classified with the second-tier label;

the operations further comprising:

storing the plurality of first-tier content items in the first server and second server;

storing a first subgroup of the plurality of second-tier content items in the first server; and

storing a second subgroup of plurality second-tier content items in the second server, the second subgroup having content items that are not included in the first subgroup.

5 . The computing system of claim 4 , wherein each content item in the plurality of first-tier content items having a performance attribute above a threshold value, and wherein each content item in the plurality of second-tier content items having a performance attribute below a threshold value.

6 . The computing system of claim 4 , wherein each server in the plurality of servers is a serving replica server, and each serving replica server stores a unique subset of the plurality of second-tier content items.

7 . The computing system of claim 1 , the operations further comprising:

obtaining first-tier performance metrics associated with a plurality of content items classified with the first-tier label;

obtaining second-tier performance metrics associated with a plurality of content items classified with the second-tier label; and

processing the first-tier performance metrics and the second-tier performance metrics, using a machine-learned model, to determine the threshold value.

8 . The computing system of claim 1 , wherein the first server stores content items that are served in a first geographical region, and wherein the second server stores content items that are served in a second geographical region.

9 . The computing system of claim 7 , wherein the second content item includes a first targeting condition associated with the first geographic region, wherein the third content item includes a second target condition associated with the second geographic region, and the operations further comprising:

determining to store the second content item in the first server based on the first target condition; and

determining to store the third content item in the second server based on the second target condition.

10 . The computing system of claim 8 , wherein the first target condition for the second content item is received from a merchant associated with the second content item.

11 . The computing system of claim 1 , wherein the performance attribute is associated with a number of impressions over a specified period of time.

12 . The computing system of claim 1 , wherein the performance attribute is calculated using a machine-learned prediction model.

13 . The computing system of claim 1 , wherein the performance attribute of the second content item is updated to be more than the threshold value, the operations further comprising:

re-classifying the second content item with the first-tier label; and

storing the second content item in the second server and a third server.

14 . The computing system of claim 1 , wherein the performance attribute of the first content item is updated to be less than the threshold value, the operations further comprising:

re-classifying the first content item with the second-tier label; and

removing the first content item from the second server.

15 . A computer-implemented method for serving content items, the method comprising:

classifying a first content item with a first-tier label based on a performance attribute of the first content item;

classifying a second content item with a second-tier label based on a performance attribute of the second content item, wherein the performance attribute of the second content item is less than a threshold value;

classifying a third content item with a second-tier label based on a performance attribute of the third content item, wherein the performance attribute of the third content item is less than the threshold value;

storing the first content item and the second content item in a first server in a plurality of servers; and

storing the first content item and the third content item in a second server in the plurality of servers.

16 . The method of claim 15 , wherein the plurality of servers includes a third server, the method further comprising:

classifying a fourth content item with a first tier-label based on a performance attribute of the fourth content item, the performance attribute of the fourth content item being above the threshold value;

storing the fourth content item in the first server and the second server;

classifying a fifth content item with a second tier-label based on a performance attribute of the fifth content item being below the threshold value; and

storing the fifth content item in the third server.

17 . The method of claim 15 , the method further comprising:

storing a plurality of first-tier content items in the first server and second server, wherein the plurality of first-tier content items comprising of content items classified with the first-tier label, and wherein each content item in the plurality of first-tier content items having a performance attribute above a threshold value;

storing a first subgroup of a plurality of second-tier content items in the first server, wherein the plurality of second-tier content items comprising of content items classified with the second-tier label, wherein each content item in the plurality of second-tier content items having a performance attribute below a threshold value; and

storing a second subgroup of the plurality second-tier content items in the second server, the second subgroup having content items that are not included in the first subgroup.

18 . The method of claim 15 , the method further comprising:

obtaining first-tier performance metrics associated with a plurality of content items classified with the first-tier label;

obtaining second-tier performance metrics associated with a plurality of content items classified with the second-tier label; and

processing the first-tier performance metrics and the second-tier performance metrics, using a machine-learned model, to determine the threshold value.

19 . The method of claim 15 , wherein the first server stores content items that are served in a first geographical region, wherein the second server stores content items that are served in a second geographical region, wherein the second content item includes a first targeting condition associated with the first geographic region, wherein the third content item includes a second target condition associated with the second geographic region, and the method further comprising:

determining to store the second content item in the first server based on the first target condition; and

determining to store the third content item in the second server based on the second target condition.

20 . One or more non-transitory computer-readable media that collectively store instructions that, when executed by one or more computing devices, cause the one or more computing devices to perform operations, the operations comprising:

classifying a first content item with a first-tier label based on a performance attribute of the first content item;

classifying a second content item with a second-tier label based on a performance attribute of the second content item, wherein the performance attribute of the second content item is less than a threshold value;

classifying a third content item with a second-tier label based on a performance attribute of the third content item, wherein the performance attribute of the third content item is less than the threshold value;

storing the first content item and the second content item in a first server in a plurality of servers; and

storing the first content item and the third content item in a second server in the plurality of servers.