Activ Splits Opra Processing To Pre-empt Data Growth

Ticker plant and datafeed vendor Activ Financial has almost completed implementing changes to its technology that allow users to scale their hardware requirements to avoid latency from the growing data traffic from the Options Price Reporting Authority (IMD, March 6).

Activ has updated the messaging middleware technology used by its content gateways and downstream servers that receive and process data by splitting the component that handles Opra data across multiple hardware processors.

"Up until this new release, we funneled all Opra streams—a dozen or more—through a device in our datafeed infrastructure for last value caching," says Mike Dunne, chief technology officer at Activ. "With the new release, we have partitioned the cache across multiple processors, which can be run on multiple servers… into an arbitrary number of components."

So far, the vendor has deployed it over two processors, but Dunne says it could be split over any number to meet requirements simply by adding more hardware. "So we automatically support parallel processing of Opra data now," he says.

He says the work on the changes, which are known internally as version 1.7 of Activ's middleware, began last fall. Activ notified clients of the forthcoming change before Christmas and rolled it out to client servers and servers at Activ facilities that provide hosted datafeeds over the last four weeks. "We've done it on our end that generates our consolidated feed and on our hosted model. But also for clients who run our direct feed product, they run on-site servers, and we have partitioned them…. We've now got it running on all but three servers in the field," he says.

"The infrastructure upgrade will be transparent to any subscribing application," says Activ president Frank Piasecki. "No one will receive a notification that they have to upgrade anything, just that they can run it on separate machines. We ourselves deploy [our middleware] for our ticker plants and hosted [points of presence], and we have deployed split Opra processing," he says.

Piasecki says that with the expectation that Opra data rates will double during the course of this year, the vendor decided that this was a good time to split its own servers to pre-empt any latency arising from the increased traffic.

"If clients don't go for a split architecture and keep the infrastructure as it is, the latency they will experience will start to grow. Adding extra servers will help reduce the latency," he says. "For the price of a $4,000 piece of server hardware… they can deal with the increased traffic rates for now and beyond."

Dunne says the move amounts to no more than reconfiguring servers to be able to add more as required. "At the API level, there is no software release to support," he says. The move from single-stream to parallel processing could be applied to any other data set distributed by the vendor, Dunne says, although "there is no technical reason to. With all the other data sets we have, we keep Opra on a separate stream… because it is a gigantic feed."

Activ runs its hosted servers on dual-processor AMD Opteron chip-based servers and recommends the same configuration to clients who run their own hardware on-site. "Clients can get dual-core machines, or you can get a four-core server for a modest amount of money now," Dunne says.

He says Opteron chip-based machines are more suitable because they have lower memory latency than Intel chips. "When you're updating a multi-gigabyte database, those nanoseconds that it takes to fetch data from memory and pull it into the CPU all add up," he says.