Realizing unprecedented performance with cost efficient scalability
Deliver 2x Transaction Rates with Onload®
Transaction-intensive cloud applications can’t afford poor performance. Solarflare’s Onload acceleration solution enables faster and more consistent response times – creating better customer experiences. When distributed databases utilizing Solarflare’s Onload underpin these applications, cache tiers – typically built on key/value NoSQL stores such as Redis and Memcached – meet unprecedented service levels for performance and scalability.
The Problem: OS Overhead Penalty
Kernel based networking is limited in efficiency due to large numbers of memory copies, lots of context switching, high interrupt rate, and lock contention; further aggravated by the kernel being pinned to CPU0.
The Solution: Bypass the OS and Run
Networking Services in Lightning-Fast User Space
Solarflare’s Onload kernel bypass, eliminates the network bottleneck and overhead penalty of operating systems—allowing industry standard caching servers to support 2x more requests. With Onload, data centers can deploy fewer remote caching servers supporting more requests, realizing significant cost savings.
Acceleration Creates Opportunities for Fewer Caching Servers or More App Servers
IO Through OS Kernel
200K RPS per Cache Node
IO Through Kernel Bypass
800K RPS per Cache Node
Onload Increases Capacity
Support More Web/App Servers
Solarflare’s Onload Solution
How Solarflare’s Onload kernel bypass solution accelerates performance
- By streamlining and reducing interrupts, context switches and data copies
- By reducing latency 50% and increasing message rates 4x or more
Seamlessly integrates into existing infrastructure
- Binary compatible with industry standard APIs
- No software modifications are needed
- Standards-based solution uses TCP/IP and UDP
- No specialized protocols needed
Available with Solarflare XtremeScale Ethernet adapters
Test Results Validate Solarflare’s Onload-Enabled Database Caching Servers Handle 2X More Redis Requests
For in-memory databases, Onload provides 2X performance improvements vs. the same Ethernet adapter and a kernel-based driver on a Redis database caching server.