perfview
perfview copied to clipboard
Smaller gcdump storage
- Use shared storage for module names
- Apply differential compression to m_nodes
- Apply differential compression to m_nodeAddresses
Can you share a comparison on what sort of benefit you see with and without this?
For a heap dump with 330M objects, this change reduced the storage from ~9GiB (which is greater than the 8GiB gcdump limit of my modified build) to 6.22GiB. This is a prerequisite to upcoming work that doubles the amount of data PerfView can load at once.