Cassandra Datastax最佳合并选项

问题描述

我正在使用cassandra作为后端的spring / java网络应用程序上工作；该应用程序可能同时被数百名客户使用。

我看到默认的Cluster PoolingOptions连接池设置（使用协议v3）是：

 本地主机：core = max = 1 
远程主机：core = max = 1 
  

，默认的maxRequestsPerConnection设置（使用协议v3）为：

 对于本地主机为1024，对于远程主机为256。 
  

这些默认设置是否足以满足我们的使用要求？

如果没有，我可以从哪些池选项开始？

解决方案

我建议使用默认的 PoolingOptions 。除非您有充分的理由，否则不应该真正接触它。

请注意，该配置是针对每个主机的 ，而不是针对整个集群。 / p>

此外，默认情况下，cassandra一次仅处理128个本机传输请求（），因此配置 PoolingOptions 处理每个低1024个请求

最后，在基准测试和具有少量节点的高吞吐量场景中，将每台主机的连接数增加到8个可以表明吞吐量有所提高，但是随着您添加更多的C *主机，改进变得微不足道。

I'm Working on a spring/java webapp using cassandra as backend; the app would be used by potentially hundreds of customers simultaneously.

I see that default Cluster PoolingOptions connection pool settings (With protocol v3) are:

LOCAL hosts: core = max = 1
REMOTE hosts: core = max = 1

And default maxRequestsPerConnection setting (With protocol v3) is:

1024 for LOCAL hosts, and 256 for REMOTE hosts.

Will these default settings be sufficient to fulfill our usage requirements?

If not, What poolingoptions can I start with?

解决方案

I would recommend using the default PoolingOptions. It should not really be touched unless you have a good reason.

Note that the configuration is per host, not for the entire cluster.

Also, by default cassandra only processes 128 native transport requests at a time (native_transport_max_threads), so configuring PoolingOptions to handle more than 1024 requests per connection has low utility.

Finally, during benchmarking and high throughput scenarios with a small number of nodes, increasing the number of connections per host to something like 8 can show some improvement in throughput, but as you add more C* hosts that improvement becomes marginal.

这篇关于Cassandra Datastax最佳合并选项的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！