You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The current implementation of the PC tree Tensor Data Model I/O is very slow. A 2M tensor PC tree can be expected. A 1M node tree made from 1M "janky" test tracks takes 7 seconds to generate, 23 seconds to convert to tensordm, 14 seconds to index the tensors and 28 seconds to convert back to pctree with the index.
From google-perftools, about half of the time is spent constructing and destructing JsonCPP::Value.
Even if that magically goes away, it would be still too slow.
I believe the core problem is that the pctree TDM results in a large number of small files/objects and the per file/object I/O overhead is killing performance. The implementation must either remove this overhead or the pc tree tensor data model must be redefined to allow fewer, more monolithic objects.
The text was updated successfully, but these errors were encountered:
The current implementation of the PC tree Tensor Data Model I/O is very slow. A 2M tensor PC tree can be expected. A 1M node tree made from 1M "janky" test tracks takes 7 seconds to generate, 23 seconds to convert to tensordm, 14 seconds to index the tensors and 28 seconds to convert back to pctree with the index.
From google-perftools, about half of the time is spent constructing and destructing
JsonCPP::Value
.Even if that magically goes away, it would be still too slow.
I believe the core problem is that the pctree TDM results in a large number of small files/objects and the per file/object I/O overhead is killing performance. The implementation must either remove this overhead or the pc tree tensor data model must be redefined to allow fewer, more monolithic objects.
The text was updated successfully, but these errors were encountered: