Neki平台完成大规模性能基准测试，在512个分片的配置下达到了1.18亿次每秒查询（QPS）的持续吞吐量1。具体来看，该平台在连续16分钟内保持118,538,803 QPS的稳定性能，最高峰值达到118,747,267 QPS1。这一成绩验证了系统的线性可扩展能力，从5个分片扩展到512个分片时，吞吐量实现了10倍的线性增长1。
基准测试的系统架构由512个分片组成，每个分片运行一个Postgres主库，配合480个Neki路由器1。在此规模下，系统处理了1.22PiB的数据，实现了15.8M读取操作每秒（IOPS），网络带宽超过2Tb每秒1。单个分片可持续支撑231k QPS的性能水平1。
性能表现方面，系统的P99延迟分别为路由器端6.06毫秒和客户端端13.95毫秒1。在可靠性方面，错误率控制在约180万个查询中出现1个错误，每秒产生67个错误1。
Neki has demonstrated substantial performance capabilities through large-scale benchmarking, sustaining 118,538,803 queries per second over a 16-minute period across 512 shards, with peak throughput reaching 118,747,267 QPS 1. The test processed 1.22 pebibytes of data and validated the system's linear scalability, with each individual shard consistently handling 231,000 QPS 1.
The infrastructure supporting this performance comprised 512 shards, each running a single PostgreSQL primary instance, coordinated through 480 Neki routers 1. The system achieved a P99 latency of 6.06 milliseconds at the router level and 13.95 milliseconds at the client level, while maintaining a low error rate of approximately one error per 1.8 million queries 1. The platform delivered 15.8 million read input/output operations per second and sustained network bandwidth exceeding 2 terabits per second 1. When scaled from 5 shards to 512 shards, throughput demonstrated tenfold linear expansion, confirming the system's proportional scalability 1.
评论
还没有评论，欢迎留下第一条。