Check out the newly published July/August 2016 IEEE Micro article on Implementing Ultra Low Latency Data Center Services with Programmable Logic by John Lockwood and Madhu Monga, pp. 18-26.
An association of a key to a value is an essential network service in modern datacenters. Keys refer to alphanumeric strings or binary data; values are associated with keys and may change over time. Key-value store (KVS) systems are typically deployed in datacenters to exchange data between servers as users shop, trade, and send data over the Web. Widely used KVS deployments include Dynamo at Amazon (http://aws.amazon.com/dynamodb) and Memcached at Zynga, Twitter, and Facebook. Telecommunication carriers use KVS servers to track the status of mobile users, devices, and network activities.
Today, millions of database servers are used every second to associate billions of keys with values due to human-to-human and human-to-machine interactions (see the "Database Services" sidebar). In the future, far more powerful servers will need to handle trillions of machine-to-machine interactions per second. Reducing latency improves network service responsiveness. Efficient KVS implementations reduce the capital expenses of building out a datacenter service and the operating expenses to power it.
Critical KVS metrics are latency, throughput, and power. Stock traders, brokers, and dealers, as well as option exchanges trading in co-location facilities, need to look up order IDs with low latency. In this article, Algo-Logic reports record-breaking KVS performance through the use of a new, open-standard message format.