AWS OpenSearch 实例类型 - 拥有几个较大的实例还是更多较小的实例更好?
我是一名初级开发人员工程师,并且有一个非常基本的问题。
我的团队目前正在提供AWS OpenSearch群。由于我们的问题类型,我们需要存储优化的实例。从亚马逊文档中,我发现他们建议至少3个节点。我知道所需的存储大小是我知道的,在OpenSearch Service定价计算器中,我发现我可以选择10 i3.large实例或5 i3.xlarge。我检查了价格,它们是一样的。
因此,我的问题是,当我面对这样的问题时,我是否选择较小的实例或更大数量的较小实例?我对原因特别感兴趣。
谢谢你!
I am a junior dev ops engineer and have this very basic question.
My team is currently working on providing an AWS OpenSearch cluster. Due to the type of our problem, we require the storage-optimized instances. From the amazon documentation I found that they recommend a minimum number of 3 nodes. The required storage size is known to me, in the OpenSearch Service pricing calculator I found that I can either choose 10 i3.large instances or 5 i3.xlarge ones. I checked the prices, they are the same.
So my question is, when I am faced with such a problem, do I choose the lesser bigger instances or the bigger number of smaller instances? I am particularly interested in the reason.
Thank you!
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(1)
每个VM的OS都有一些开销,因此10个较小的实例将使 的计算和RAM总共可用于5个较大的实例。另外,如果您只保留默认索引设置(5个主片,1个复制品),并且一次单独写入1个索引,那么您将有效地只有5个节点为您索引数据(这些节点将具有较少的带宽,因为他们很小)。
因此,我通常建议运行一些较大的实例,而不是许多较小的实例。在某些特殊情况下,这是不正确的(例如同时进行搜索的群集),但是对于那些人来说,我建议首先使用更大的实例。
Each VM has some overhead for the OS so 10 smaller instances would have less compute and RAM available for ES in total than 5 larger instances. Also, if you just leave the default index settings (5 primary shards, 1 replica) and actively write to only 1 index at a time, you'll effectively have only 5 nodes indexing data for you (and these nodes will have less bandwidth because they are smaller).
So, I would usually recommend running a few larger instances instead of many smaller ones. There are some special cases where it won't be true (like a concurrent-search-heavy cluster) but for those, I'd recommend going with even larger instances in the first place.