This document discusses the limitations of relational databases for modern applications and real-time architectures. It describes how NoSQL databases like Aerospike can provide better performance and scalability. Specific examples are given of how Aerospike has been used to power applications in domains like advertising technology, social media, travel portals, and financial services that require high throughput, low latency access to large datasets.
The document provides an overview of the Aerospike architecture, including the client, cluster, storage, indexes, RAM, flash storage, and cross datacenter replication (XDR). It describes Aerospike's goals of handling high transaction volumes at low latency while scaling linearly. The key aspects of the architecture are the smart client that routes to data in one hop, shared-nothing nodes, single row transactions, smart cluster management, and XDR for data replication across datacenters.
This document discusses database performance characteristics and benchmarks Aerospike on Google Compute Engine (GCE). It finds that with 50 nodes, Aerospike achieved a median latency of 7ms and 83% of requests under 16ms latency for 1 million writes per second. CPU utilization was only 50-60% due to overhead. Network bottlenecks were identified, and optimizations like DPDK helped achieve 4.2 million reads per second with 90% under 4ms latency. Live migrations can impact highly consistent databases and their applications. Local SSDs provide good performance as an alternative to RAM and were benchmarked positively with Aerospike.
The document provides an overview of Redis, a key-value store with optional persistence and various data structures. It details how Redis can handle different data types such as numbers, strings, lists, sets, and sorted sets, and explains transactional operations for reading and writing data. Additionally, it highlights upcoming features in Redis 2.6 and 3.0, including Lua scripting and clustering.
This document discusses the limitations of relational databases for modern applications and real-time architectures. It describes how NoSQL databases like Aerospike can provide better performance and scalability. Specific examples are given of how Aerospike has been used to power applications in domains like advertising technology, social media, travel portals, and financial services that require high throughput, low latency access to large datasets.
The document provides an overview of the Aerospike architecture, including the client, cluster, storage, indexes, RAM, flash storage, and cross datacenter replication (XDR). It describes Aerospike's goals of handling high transaction volumes at low latency while scaling linearly. The key aspects of the architecture are the smart client that routes to data in one hop, shared-nothing nodes, single row transactions, smart cluster management, and XDR for data replication across datacenters.
This document discusses database performance characteristics and benchmarks Aerospike on Google Compute Engine (GCE). It finds that with 50 nodes, Aerospike achieved a median latency of 7ms and 83% of requests under 16ms latency for 1 million writes per second. CPU utilization was only 50-60% due to overhead. Network bottlenecks were identified, and optimizations like DPDK helped achieve 4.2 million reads per second with 90% under 4ms latency. Live migrations can impact highly consistent databases and their applications. Local SSDs provide good performance as an alternative to RAM and were benchmarked positively with Aerospike.
The document provides an overview of Redis, a key-value store with optional persistence and various data structures. It details how Redis can handle different data types such as numbers, strings, lists, sets, and sorted sets, and explains transactional operations for reading and writing data. Additionally, it highlights upcoming features in Redis 2.6 and 3.0, including Lua scripting and clustering.
28. ノードのクラスタイン確認。02:42:26
Let’s Try!
バージョンアップ後同じデータに対してプロセスのリスタート実施
Mean of Partitions log
absent: Number of partitions not owned by this node
sync: Number of non-master partitions owned by this node
actual: Number of master partitions owned by this node
actual + sync + absent = 4096 partitions for the namespace