The Gateway to Algorithmic and Automated Trading

iPinYou chooses Azul Zing to increase bidding win ratios for daily advertising transactions

First Published 15th October 2015

Zing enables iPinYou to boost its Java heap size capacity and reduce the number of JVM instances, while simultaneously increasing the throughput of iPinYou's platform by 20%.

Scott Sellers, Azul Systems

Scott Sellers, Azul Systems

"By deploying Azul Zing, Java-based ad serving applications can handle far more ad requests while meeting strict SLAs."

Sunnyvale, California and Beijing, China - Azul Systems, the provider of Java runtime solutions, has announced that iPinYou, the independent Demand Side Platform (DSP) in China, is using Azul Zing, Azul's highly scalable Java Virtual Machine (JVM), to power its Java-based online advertising infrastructure. Zing improves ad success rates by providing low latency for Java-based ad servers, driving peak latency and jitter.

iPinYou processes up to eighteen billion advertising transactions per day for web service providers in China, implementing campaigns which combine customer insight with high performance algorithmic technology to maximize ad performance. iPinYou's partners include Google, Alibaba, Tencent and Baidu. From partners, iPinYou receives bidding requests which are processed to determine which ad to show and how much to bid. In China, the total transaction time for this process must be below the Exchange SLA of 50~100 milliseconds.

Java applications rely on the underlying JVM to perform memory management. With typical JVMs, high volumes of ad requests fill memory quickly and trigger GC pauses. Historically, a significant number of iPinYou transactions had not met SLAs due to GC artifacts - and slow or unpredictable response times resulted in lost opportunities for a sale or click through - and ultimately lost revenue. Since Zing has been deployed in iPinYou's infrastructure, GC pauses have been removed and exchange SLAs have been consistently met for all online transactions.

Improving data consistency and throughput is particularly important for iPinYou partners that are geographically distant from data centers. Prior to Zing, iPinYou had deployed multiple JVMs so that it could minimize the size of its heap and lower pauses - by deploying Zing, larger JVM heap sizes have been implemented and the total number of JVMs reduced, simplifying system operation and reducing chance of operational error.

Xuehua Shen, CTO and co-founder of iPinYou, said: "We use industry leading algorithms and Azul Zing has helped us get maximum value out of these for our customers. We are delighted we have chosen to deploy the Zing JVM, having seen exceptional improvement in our ability to meet our strict SLAs on our ad serving infrastructure. We are currently moving other parts of our system, including our Hadoop deployment, to Zing as well."

Scott Sellers, CEO and President of Azul Systems, said: "By deploying Azul Zing, Java-based ad serving applications can handle far more ad requests while meeting strict SLAs. Zing has improved iPinYou's bottom line by achieving better and more consistent response times for their business-critical application infrastructure. There is no other JVM which delivers such high sustained throughput combined with consistent low latency, a perfect fit for high volume web-scale use cases."