| ... | @@ -8,3 +8,16 @@ In iterative execution, a chunk of input is processed in one operation and then |
... | @@ -8,3 +8,16 @@ In iterative execution, a chunk of input is processed in one operation and then |
|
|
|
|
|
|
|
### Challenges in Cross-Device Iterative Processing
|
|
### Challenges in Cross-Device Iterative Processing
|
|
|
|
|
|
|
|
|
1. Optimal chunk size
|
|
|
|
* chunk for individual devices
|
|
|
|
* chunk for individual operations
|
|
|
|
|
|
|
|
2. pushing data vs pulling data
|
|
|
|
3. optimal split vs merge operations
|
|
|
|
4. Preference for operation placement
|
|
|
|
5. Data partitioning with functional placement
|
|
|
|
* data vs functional parallelism
|
|
|
|
6. Type of linearization
|
|
|
|
* BFS
|
|
|
|
* Topological sorting
|
|
|
|
* DFS ( in-addition with data parallel execution) |
