FAST'22 KeyNote: 25 Years of Storage Research and Education: A Retrospective

Lessons Summary
• When working on something, ask: how can I be the best?
• Never underestimate the power of textbooks
• Use papers you like as inspirations
• Sometimes ignore advice
• Keep eyes open when doing research
• Read widely and take notes
• Ideas can be “close to right”; keep thinking and refining
• Always think about what can be measured, and how to learn from it
• Ask how new technologies change how we build systems
• Use good ideas again (in different contexts)
• Modeling can be useful
• Explore ideas from other (sub)fields

小idea：
数据分层对精确度要求不一样然后通过插值，或者通过机器学习加强来实现类插值？
专利：提出一个预处理系统模块cassandra+希尔伯特
首先keans画圈，若是数据分布集中度较高那就没必要压缩了
对数据进行希尔伯特曲线分块，然后确定locality程度！d8_tree
k-v系统的double - hash，一次模拟双系统？
paraview to use rdma to transfer data
基于rdma的分布式数据刻
可视化
ai预测的可视化
计算机系统会议论文 - 知乎 https://www.zhihu.com/column/c_1424714267832967169
基于RDMA的分布式内存池的优化
节点之间的信息读取
现有的都是通过多层转发
可以通过直接赋予密钥来实现减少overhead

is it possible to implement ml in building minibounds when partioning data？

qq文件还有图片未上传，关于高性能可视化和深度学习可视化～

是否可以搞一个自适应的节点数的k-d树？以求优化并行性能？

大内存多机集合机器：集群部署，应用无感；分布式操作系统 laxcus，k8s，