Validating 4 Million TPS using AerospikeDB with only 20 GCE nodes Today’s guest bloggers are Lynn Langit – Independent Cloud and Big Data Architect named the Google Cloud Developer Expert for the past three years and recognized as an AWS Community Hero, Microsoft MVP (SQL Server), MongoDB Master and ACM Distinguished speaker – and David Haley, startup engineer and former Google Cloud team member who was the first engineer on the Google Cloud Platform Americas sales team. He lives in Issaquah, Washington, where he's working on his latest novel,Storm Crow. His clients have included Fortune 500 companies and US and European government agencies. , specializing in document and data publishing, semantics, data analytics, visualization, and taming the Big Data beast. Kurt Cagle is an information architect, data scientist, industry analyst and, yes, consultant, who works for Avalon Consulting, LLC. ![]() Further context about the benchmark may be found in Avalon’s announcement regarding the benchmark here. If you're evaluating MongoDB relative to your specific NoSQL needs, you owe it to yourself and to your organization to go get the full benchmark report here. I have to admit to being surprised here - I've liked MongoDB for a while, but between the server numbers and the introduction by Couchbase of a new SQL like query language - N1QL - that I've written about before - if I was asked now to recommend an open source NoSQL database to my clients, it would absolutely have to be Couchbase. The results were unequivocal - Couchbase server, across the board, was just plain better than MongoDB. ![]() The benchmarks were run three times for each database, with the numbers staying very consistent between runs, ruling out the likelihood that other factors, such as latency from outside the system, may have been at play. Out of curiosity, Avalon ran additional tests for Couchbase, to find that it's performance crossed the 5 ms latency line established earlier at a whopping 800 requests from 23 simultaneously clients, more than three times above Mongo's performance (Figure 1)įigure 1. Couchbase too would eventually put up the flag, but it was well above the 500 requests per second that had been established as the upper end of the initial tests. Latency began to rise dramatically for Mongo after that, to the extent that its latency fell outside of Avalon's test parameters by 250 requests per second. When the same number of nodes (nine in both cases) were used, Couchbase definitely performed better, but even at 9 vs 6 node clusters, Couchbase had comparable performance under higher loads.īy two hundred client requests per second, MongoDB was struggling, Couchbase was just getting its second wind. The Couchbase architecture doesn't face that same problem, which actually meant that Mongo needed a nine node cluster just to do an apple to apples comparison with Couchbase running on six nodes. One of the key problems that MongoDB has, even with WiredTiger, is that its architecture requires coordination between three different server instances, creating a bottleneck when network loads begin to rise. Arguably, for relatively low demand uses, either provides about the same level of support. The Yahoo Cloud Serving Benchmark (YCSB) was used to simulate client load.Īt relatively low demand levels (up to about fifty simultaneous clients), both Couchbase and MongoDB held their own - MongoDB improving considerably on their previous version, while Couchbase continued to maintain an edge in performance even at these levels. Avalon established a set of real world scenarios that concentrated on three key variables - how quickly document-like records could be be read, how quickly they could be written, and how both databases scaled in a cloud environment, using Amazon EC2 servers. Given Mongo’s recent release of MongoDB 3.0, which incorporates WiredTiger, and the associated marketing related to performance improvements, Avalon Consulting, LLC, decided to put it to the test and pit these two NoSQL contenders against one another (Couchbase released their own Couchbase Server 3.0 late last year). MongoDB is the more established of the two, while Couchbase is the scrappier, making waves of its own with better performance and definitely better scalability. ![]() While there are a number of players in the open source NoSQL space, the two heavyweights in this area are MongoDB and Couchbase. A social media site, mobile platform or web application needs to be able to go from normal traffic loads to high demand in a heartbeat - a news event or favorable review can send demand for your content soaring, and if the database in the background can't handle it, the resulting fallout could be bad news for you. In the NoSQL market, speed, scalability and flexibility matter.
0 Comments
Leave a Reply. |