The business benefits of running Neo4j on IBM Power Systems Hardware
Choosing the optimal hardware for complex workloads results in significantly better business benefits and outcomes verses simply viewing the choice of hardware as a commodity. IBM has designed the POWER8 system to deliver significantly higher business value for modern complex workloads that often are called upon to deliver the highest qualities of service for a combination of Relational and NoSQL databases and new Cognitive applications. IBM POWER8 Systems are delivered with 4x more threads per core, 4x higher memory bandwidth, and 6x more processor cache than x86 systems. Faster processors with more threads per core enable much faster execution of today's complex algorithms. The superior memory bandwidth of POWER8 maximizes business insights, and the larger processor caches on POWER8 ensure lightning fast response times by keeping the data closer to the CPUs. In today’s high performance compute environments, acceleration of algorithms is key to meeting end customers’ service requirements. Acceleration can deliver both higher performance and lower cost, but only if the compute platform is designed for heterogeneous computing.
Many software applications and workloads, if based on Linux, can immediately inherit the superior qualities of service of the POWER8 hardware and the end result is significant business advantages. In addition, The IBM POWER8 chip, with its Coherent Accelerator Processor Interface (CAPI) technology, delivers on the promise of accelerated computing and is currently leveraged by some of the hottest NoSQL database vendors in the marketplace, including both Neo4j and Redis Labs.
As one (of many) examples of a specific proof point where hardware can bring incremental business value to a modern workload, back in 1Q2017 IBM conducted some internal testing of two different hardware system (both running Linux bare metal images). A mixed Neo4j Graph transaction workload based with a 200GB data model was run and measured on both an IBM POWER8 System 822LC server with 20 cores and an x86 based system with 24 cores. Both systems measured were running the exact same Neo4j based Graph workload . At the time of the testing, the IBM POWER8 system was priced at approximately $19K USD and was able to comfortably run 711 total Neo4j Graph operations per second, while the x86 system was priced at approximately $17K and only able to run 390 Neo4j Graph operations per second. That's a 1.6X price/performance advantage and a 1.8X performance per server advantage in favor of IBM POWER8 hardware for the identical Neo4j based Graph workload !
But wait - The story gets even better - less hardware to run your workloads (like Neo4j) will often result in a significant reduction in software licensing costs, lower IT admin resources, reduce risk by lowering the need for constant performance tuning, require less expensive floor space, and can significantly lower ongoing maintenance costs.
All the benefits of running Neo4j on Linux on Power that are listed in the previous sections enable you to do alot more with less costs and risk. A Neo4j on Linux on POWER8 environment allows you to more efficiently and effectively manage the growth of your graph workloads and user base, serve more users at lower cost, and run more transactions per second while spending alot less money on a solution with higher qualities of service. Internal testing done by IBM on POWER8 systems on other highly popular workloads like Hortonworks HDP, MongoDB, Enterprise DB and RedisLabs also show similar or better price/performance improvements when running on IBM POWER8 vrs x86 systems.
Now, since the costs are lower -what about reliability ? More good news. Back on 23 Jun 2017 IBM announced that corporate enterprise users ranked IBM z Systems 'enterprise mainframe class’ servers the most reliable hardware platform available on the market today. In its ‘mainstream’ server category the survey found that IBM Power Systems achieved the highest levels of reliability and uptime when compared with 14 server hardware options and 11 server hardware virtualization platforms. The results come from the ITIC 2017 Global Server Hardware and Server OS Reliability survey, which polled 750 organizations worldwide between April and May of 2017. The survey findings also concluded the following:
- IBM Power Systems has the least amount of unplanned downtime, with 2.5 minutes per server/per year of any mainstream Linux server platforms.
- IBM and the Linux operating system distributions were either first or second in every reliability category, including virtualization and security.
In summary, if performance or price/performance or reliability of your IT environments are not meeting your expectations or your customers’ expectations, if your user base or application catalog is rapidly growing rapidly, or if your current software spend is inhibiting innovation, it may be a very good time to examine the business benefits of running your next wave of modern , data centric workloads (like Neo4j, MongoDB, Enterprise DB, Redis, Hortonworks, Machine Learning, Deep Learning and many others) on Linux on IBM Power Systems.
See more information on the detailed test results at https://bit.ly/wwpowersystems. All opinions are my own.