A year ago, my colleague, Ellen Brigham, posted a blog that reported the results of the rigorous STAC M2 benchmarking of our first QFabric family member: the QFX3500 Switch. We were proud to post the best scores ever with that product, proving that Juniper was a leader in the low-latency networking space. My more technical friends tell me the STAC benchmarks are brutal—with precise tuning of applications, buffers and threads, placing particular stress on the multicast functions in the network (or just the switch, as was always the case up ‘til now).
Today we are releasing even more impressive results with the benchmarking of a full QFabric System, based on the same STAC M2 standards—the first-ever STAC benchmark on a data center network fabric (or any network environment more complex than a single switch). This is significant because in a traditional tiered network, latency and jitter add up, meaning the larger the network gets, the less deterministic it is. Only QFabric has proven to deliver fast, predictable performance across the more complex environment. The benchmark determined the mean latency for the QFabric System, which in the STAC M2 test was rated at 10 microseconds (μsec). For comparison purposes, the best published score Cisco recorded for its Nexus 5010 switch had a mean latency of 14 microseconds (μsec). Full results are available at http://www.STACresearch.com/
Since Cisco’s version of a “fabric” uses the Nexus 5000 as an access device (before you get to a Lotta Links running Cisco’s own FabricPath protocols), and a full QFabric System has shown a lower latency benchmark than a single N5k, it is safe to assume that people building data centers with QFabric will have BIG performance advantages.
With this test, we have shown that the QFabric System can serve as the heart of a high-performance data center, with the latency and management simplicity of a single low-latency switch. This offers tremendous advantage to bankers and financial markets in particular, but it’s also exciting news for anybody dealing with large sets of real-time data such as SmartGrids, air traffic control, transportation, retailers, and national security & defense.
Based on these results demonstrating QFabric’s low-latency performance at data center scale, imagine the other performance-sensitive applications that can deliver a real advantage if “turbo charged” on a QFabric network:
Risk and Compliance: Some trading firms implement as many as 500 pre-trade compliance checks that run across multiple systems before a trade is approved. Removing network latency accelerates those approvals and gets winning bids into the market faster.
Analytics: Big data, Hadoop clusters and other large-scale data sets are taking over data center real estate. As the data sets become more massive, best practice means you need to move the query or application to the data, rather than the other way around. The faster you can do that, the more timely—and valuable —your analysis will be.
Real-time Core Banking: Managing a bank with days-old data from batch reports is like flying a plane looking out the rear window. Banks like BBVA are learning that updating the general ledger in real-time is more like looking where you are going—and managing cash and risk more effectively because of it. But real-time banking creates a flood of transactions in the network core. This is exactly QFabric’s strength, as proven in the STAC test.
Floor Planning: Determining where to locate servers and storage may not be the most exciting topic, but if you’ve read this blog before, you know that location matters in a data center. System administrators carefully deploy those applications that communicate most frequently in close proximity to each other to maintain performance. With QFabric, that careful planning is no longer necessary—you can place servers, data and applications anywhere in the data center with no performance penalty
One of the most interesting stories to come out of the QFabric System benchmark and tuning was actually an oversight. While the network team was setting up the network, they connected a Spirent TestCenter as a traffic generator for basic load testing. Once the network was up and running, the application engineers came in to set up the servers and applications and begin tuning the STAC test harness to determine the optimum configuration for the formal benchmark.
Running through some final quality checks, engineers discovered that the Spirent system had been running the entire time,broadcasting at 80Gbps across the QFabric System while the preliminary testing was going on. The Spirent system was disconnected and the tests were run again—with no difference in the results.
Put another way, while running the most demanding, low-latency switching tests in the industry, the QFabric System absorbed an extra 80Gbps of traffic with no impact on the results. While not part of the official audit, this was a great proof-point demonstrating the performance headroom in the system.
In short, a low-latency, non-blocking, highly scalable network fabric is ideal for almost any performance-sensitive application. Add in the benefits of location flexibility, a single system to manage, and the proven capabilities of Junos, and you have the foundation for building the Fastest Data Center in the World.
For more information on the actual testing and the full report - please visit us at the QFabric site here on Juniper.net.
With these two test reports - Juniper QFabric is the most throroughly tested and docuemented 10 Gigabit solution in the industry. Lots of vendors claim to be the best - Juniper has provdided the proof to support our claim: The Fastest Data Center in the World.