Bypass the load balancer on the return path. With DSR, response traffic flows directly from your servers to clients, delivering maximum throughput for bandwidth-intensive workloads without bottlenecking your balancer.
DSR fundamentally changes the return path for traffic, eliminating the load balancer as a bottleneck on outbound data.
The client sends a request to the load balancer's virtual IP address, just like any standard deployment.
Tula selects a backend server and forwards the request using NFTables for high-performance, kernel-level packet manipulation.
The backend server sends its response directly to the client, bypassing the load balancer entirely on the return path.
Traditional:
Client → LB → Server → LB → Client
DSR:
Client → LB → Server → Client
The load balancer only handles inbound requests. All return traffic takes the direct path, freeing the balancer to process more connections.
DSR delivers measurable performance gains where they matter most.
Return traffic flows directly from server to client with no bottleneck. Your load balancer never limits outbound bandwidth, no matter how large your responses are.
One fewer network hop on the return path means lower response times for your users. Every millisecond counts for latency-sensitive applications.
The load balancer only handles inbound traffic, dramatically reducing its workload. Scale your backend servers without upgrading your balancer infrastructure.
DSR excels in workloads where response sizes far exceed request sizes.
Small inbound requests produce massive outbound responses. DSR ensures your load balancer never becomes the bottleneck for media delivery.
Distribute static assets and cached content at full wire speed. Each origin server responds directly without routing through a central point.
Queries that return large result sets benefit from direct return paths. Keep your database proxy layer lean while delivering results at full speed.
Any application where response size significantly exceeds request size. API endpoints returning bulk data, image processing services, and report generation all benefit.
Built on proven Linux networking primitives for reliability and performance.
Packet rewriting happens at the kernel level using NFTables, avoiding costly userspace processing. This delivers near-native network performance with minimal CPU overhead.
Choose between Layer 2 (MAC rewrite) for same-subnet deployments or Layer 3 (IP tunneling) for routed environments. Both modes are fully supported and configurable per service.
Full protocol compatibility for both TCP and UDP traffic. Deploy DSR for any Layer 4 workload without protocol-specific limitations.
DSR is transparent to your application layer. No code changes, no special libraries, no SDK integration. Configure it at the load balancer and your servers respond directly.
Deploy DSR load balancing in minutes. No credit card required for the free trial.