[This is preliminary documentation and is subject to change.]
The Configuration type exposes the following members.
Properties
Name | Description | |
---|---|---|
AdditionalSettings |
Collection of application-specific settings.
| |
Broadcast |
The network protocol to be used for broadcasting control messages.
N.B. Support for TcpUdp and UdpOnly is experimental. The BroadcastAddress must be set to use these modes.
| |
BroadcastAddress |
The address and port to be used for sending or receiving broadcast control messages.
N.B. Support for broadcast primitives is experimental. This must be set to use the TcpUdp and UdpOnly broadcast protocols.
| |
CompactionInterval |
The periodic interval in milliseconds at which Naiad will attempt to compact its collection state.
| |
DeadlockTimeout |
Setting this to a value other than Timeout.Infinite will cause Naiad to print diagnostic information
after a process sleeps for that number of milliseconds.
| |
DistributedProgressTracker |
EXPERIMENTAL: Setting this to true enables the (original) distributed progress tracker. Default is false.
| |
DuplexSockets |
If true, uses the same socket for sending and receiving data.
| |
Endpoints |
The network addresses of the processes in this computation, indexed by process ID.
| |
KeepAlives |
If true, enable TCP KeepAlives on sockets
| |
MultipleLocalProcesses |
Setting this to true ensures that, if cores are pinned to CPUs, subsequent processes are pinned to different
sets of CPUs.
| |
Nagling |
If true, uses Nagling on sockets
| |
Processes |
The number of processes in this computation.
| |
ProcessID |
The ID of the local process in this computation (starting at 0).
| |
ReadEndpointsFromPPM |
Setting this to true causes the server to assume it has been spawned by a Peloponnese manager, and
contact the manager to find out the endpoints of the other processes in the system
| |
SendBufferPolicy |
Defines how the buffer pool is divided
| |
SendPageCount |
Number of pages used for sending network data
| |
SendPageSize |
Size in bytes of pages used for sending network data
| |
UseBroadcastWakeup |
Uses optimization to wake multiple threads with a single kernel event.
| |
UseInlineSerialization |
Uses a new code generation technique to generate more efficient code for serialization.
| |
UseNetworkBroadcastWakeup |
Uses optimization to wake multiple networking threads with a single kernel event.
| |
WorkerCount |
The number of workers (i.e. CPUs used) in the local process.
N.B. At present, all processes in a computation must use the same worker count.
|
See Also