MediaGuard System Deployment
MediaGuard provides real-time, pre-bid bot predictions for each bid request you receive. By integrating directly with digital advertising bidding infrastructure, MediaGuard can identify fraudulent bids for removal from the monetization stream.
Cluster deployment and response time
We deploy MediaGuard server clusters in various regional data centers across the globe. To ensure optimal response times, your data center must be within 200 miles of a MediaGuard cluster.
HUMAN uses Amazon Web Services (AWS), Google Cloud Platform (GCP), and Equinix Metal data centers for our standard deployments. If you require a different provider not listed here, we may be able to make hosting arrangements accordingly, but this process often adds additional time to the MediaGuard integration process.
HUMAN offers MediaGuard under a software-as-a-service (SaaS) model. We do not currently support deployment at the client site on client hardware.
Cluster size
At a minimum, each MediaGuard server cluster needs at least two nodes to provide redundancy in the event of failure. Each node can process up to 20,000 queries per second (QPS) if your MediaGuard integration is built with HUMAN’s MediaGuard SDK.
We use the maximum expected QPS in a given region is used to calculate the total number of nodes to deploy in the MediaGuard cluster. For example, if you expect to receive a maximum of 90,000 QPS for a given region, HUMAN will provision six nodes for the corresponding MediaGuard cluster, based on the following formula:
⌈90,000 / 20,000⌉ = 5 (plus 1 for redundancy, for a total of 6 nodes)
The total number of nodes deployed per cluster can change significantly if you create your own SDK built on gRPC or via HTTPS connections rather than using HUMAN’s MediaGuard SDK. HUMAN's MediaGuard SDK uses round robin DNS to load balance between the nodes in the cluster. However, if you are not using our SDK, you must implement your own mechanism for load balancing traffic between nodes in the MediaGuard cluster.
Even with this load balancing in place, nodes in the MG cluster may be overwhelmed by large connection spikes that prevent the nodes from responding to lookup requests, regardless of your integration method. Because of this, we recommend maintaining fewer than 1,000 connections per node and opting to use a minimal number of high-throughput connections. We also recommend keeping connections open for as long as possible to avoid wasting resources with unnecessary reconnects.
If you expect your traffic to increase by 20% or more in a particular region, please contact HUMAN to discuss potential cluster resizing. This helps ensure that you do not experience any adverse performance impacts.