随着科学数据集规模的不断增大，高性能计算(HPC)领域的I/O瓶颈问题日益突出。用户通常下载的数据量远超实际所需，加之数据量的增长速度远快于网络带宽的提升速度，这为高效的数据服务带来了严峻挑战。因此，服务器侧的数据子集抽取和其他预处理变得越来越不可或缺。同时，面对多样化的数据访问模式，服务器必须能够进行有效的调解。
1. Even with multi-terabyte local disk sub-systems and multi-petabyte archives, I/O can become a bottleneck in HPC.
2. Users download 10x more data than needed
3. While data volumes grow,network bandwidth does not at the same rate, and hence serving such data efficiently is a challenge. Therefore, service flexibility like subsetting and other processing on server side prior to their shipping is more and more considered indispensable.
4. Server must mediate between access patterns.
5. 80% of all data have some spatial connotation.
6. Arrays at the heart of large-scale scientific data analytics, along withgraphs (incl meshes), sets, trees.
扫一扫
9582
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
