問題文
A cluster will use a third-party storage system. During bring-up the administrator has to set the initial parameters for how the cluster mounts and uses it. Two kinds of use are expected: the datasets that training jobs read repeatedly, and the home directories where users keep code and results. What is the right starting point?
選択肢
- Treat the two uses separately, because their access patterns and their tolerance for latency are different and one set of parameters will not suit both.
- Leave the parameters at their defaults and adjust them after the cluster has been in production for a while, since the defaults are chosen to be safe and the real access patterns cannot be known in advance of actual use.
- Use one set of parameters tuned for the datasets, since those carry far more bytes and the home directories are small enough to live with them.
- During bring-up, mount the storage read-only, so that neither kind of use can write anything until the parameters are settled.