Christopher Hogan

We want to model a checkpoint/restart workload. * 4 client nodes, with 40 ranks each (out of 48) * 4 PFS server nodes (HDD) * 2 BB server nodes (SSD)...

Jaime didn't provide me with anything that works. I've asked him for his scripts that configure everything manually. In the meantime, I'm working on a manual setup myself.

* Weight recency, frequency, and size equally for now. * Higher score should imply a more important blob. The logic is currently the opposite.

This involves at least 2 steps: 1. Remove the `Bucket::RenameBlob` function from the public API. 2. Change the behavior of a `Put` on an existing `Blob`. Previously it would destroy...

Test Hermes on other systems

3

It would be great to get performance data on a couple NSF systems.

Create a stress test benchmark

2

We want a program that exercises several I/O patterns and library functionality that we can use to generate performance data.

Created from JIRA [HERM-68](https://jira.hdfgroup.org/browse/HERM-68) The algorithm can be improved (currently does a linear search, which gets worse as the heap gets fragmented).

Created from JIRA [HERM-53](https://jira.hdfgroup.org/browse/HERM-53)

A Hermes application sets a global mode (`Buffer Mode` by default), and each `Bucket` or `Put` can override the mode via the `Context`. ## Buffering Mode The standard mode, where...