Joining the petabyte club with direct attached storage
暂无分享,去创建一个
Our site successfully runs more than a Petabyte of online disk, using nothing but Direct Attached Storage. The bulk of this capacity is grid-enabled and served by dCache, but sizable amounts are provided by traditional AFS or modern Lustre filesystems as well. While each of these storage flavors has a different purpose, owing to their respective strengths and weaknesses for certain use cases, their instances are all built from the same universal storage bricks. These are managed using the same scale-out techniques used for compute nodes, and run the same operating system as those, thus fully leveraging the existing know-how and infrastructure. As a result, this storage is cost effective especially regarding total cost of ownership. It is also competitive in terms of aggregate performance, performance per capacity, and – due to the possibility to make use of the latest technology early – density and power efficiency. Further advantages include a high degree of flexibility and complete avoidance of vendor lock-in. Availability and reliability in practice turn out to be more than adequate for a HENP site's major tasks. We present details about this Ansatz for online storage, hardware and software used, tweaking and tuning, lessons learned, and the actual result in practice.