Interconnection Network

From gem5
Revision as of 02:32, 25 September 2016 by Tushar (talk | contribs)
Jump to: navigation, search

The various components of the interconnection network model inside gem5's ruby memory system are described here.

How to invoke the network

Simple Network:

./build/ALPHA/gem5.debug \
                  configs/example/ruby_random_test.py \
                  --num-cpus=16  \
                  --num-dirs=16  \
                  --network=simple
                  --topology=Mesh_XY  \
                  --mesh-rows=4 

The default network is simple, and the default topology is crossbar.


Garnet network:

./build/ALPHA/gem5.debug \
                      configs/example/ruby_random_test.py  \
                      --num-cpus=16 \
                      --num-dirs=16  \
                      --network=garnet2.0
                      --topology=Mesh_XY \
                      --mesh-rows=4  



Topology

The connection between the various controllers are specified via python files.

  • Related Files:
    • src/mem/ruby/network/topologies/Crossbar.py
    • src/mem/ruby/network/topologies/CrossbarGarnet.py
    • src/mem/ruby/network/topologies/Mesh_XY.py
    • src/mem/ruby/network/topologies/Mesh_westfirst.py
    • src/mem/ruby/network/topologies/MeshDirCorners_XY.py
    • src/mem/ruby/network/topologies/Pt2Pt.py
    • src/mem/ruby/network/Network.py
    • src/mem/ruby/network/BasicLink.py
  • Topology Descriptions:
    • Crossbar: Each controller (L1/L2/Directory) is connected to a simple switch. Each switch is connected to a central switch (modeling the crossbar). This can be invoked from command line by --topology=Crossbar.
    • CrossbarGarnet: Each controller (L1/L2/Directory) is connected to every other controller via one garnet router (which internally models the crossbar and allocator). This can be invoked from command line by --topology=CrossbarGarnet.
    • Mesh_*: This topology requires the number of directories to be equal to the number of cpus. The number of routers/switches is equal to the number of cpus in the system. Each router/switch is connected to one L1, one L2 (if present), and one Directory. It can be invoked from command line by --topology=Mesh_XY. The number of rows in the mesh has to be specified by --mesh-rows. This parameter enables the creation of non-symmetrical meshes too.
      • Mesh_XY: Mesh with XY routing/
      • Mesh_westfirst: Mesh with west-first routing.
    • MeshDirCorners: This topology requires the number of directories to be equal to 4. number of routers/switches is equal to the number of cpus in the system. Each router/switch is connected to one L1, one L2 (if present). Each corner router/switch is connected to one Directory. It can be invoked from command line by --topology=MeshDirCorners. The number of rows in the mesh has to be specified by --mesh-rows.
    • Pt2Pt: Each controller (L1/L2/Directory) is connected to every other controller via a direct link. This can be invoked from command line by --topology=Pt2Pt.
    • Torus: This topology requires the number of directories to be equal to the number of cpus. The number of routers/switches is equal to the number of cpus in the system. Each router/switch is connected to one L1, one L2 (if present), and one Directory. It can be invoked from command line by --topology=Torus. The number of rows in the Torus has to be specified by --mesh-rows. This parameter enables the creation of non-symmetrical tori too. By default, this file models a folded torus topology, where the length of each link (including the wrap-around ones) is the same, and approx. equal to twice that of a mesh. The default link latency is thus assumed to be 2-cycles.

Topology_overview.jpg

  • Optional parameters specified by the topology files (defaults in Basic_Link.py):
    • latency: latency of traversal within the link.
    • weight: weight associated with this link. This parameter is used by the routing table while deciding routes, as explained next in Routing.
    • bandwidth_factor: For garnet networks, this is equal to the channel width in bytes. For simple networks, the bandwidth_factor translates to the bandwidth multiplier (simple/SimpleLink.cc) and the individual link bandwidth becomes bandwidth multiplier x endpoint_bandwidth (specified in SimpleNetwork.py).

Routing

Based on the topology, shortest path graph traversals are used to populate routing tables at each router/switch. The default routing algorithm tries to choose the route with minimum number of link traversals. Links can be given weights in the topology files to model different routing algorithms. For example, in Mesh.py, MeshDirCorners.py and Torus.py, Y-direction links are given weights of 2, while X-direction links are given weights of 1, resulting in XY traversals. adaptive_routing (in src/mem/ruby/network/simple/SimpleNetwork.py) can be enabled to make the simple network choose routes based on occupancy of queues at each output port.

Flow-Control and Router Microarchitecture

Ruby supports two network models, Simple and Garnet, which trade-off detailed modeling versus simulation speed respectively.

  • Related Files:
    • src/mem/ruby/network/Network.py
    • src/mem/ruby/network/simple
    • src/mem/ruby/network/simple/SimpleNetwork.py
    • src/mem/ruby/network/garnet/BaseGarnetNetwork.py
    • src/mem/ruby/network/garnet/fixed-pipeline
    • src/mem/ruby/network/garnet/GarnetNetwork_d.py
    • src/mem/ruby/network/garnet/flexible-pipeline
    • src/mem/ruby/network/garnet/flexible-pipeline/GarnetNetwork.py


Simple Network

Details of the Simple Network are here.

Garnet

Details of the original (2009) Garnet network are here. This design is no longer supported in the codebase.

Garnet2.0

Details of the new (2016) Garnet2.0 network are here.


Running the Network with Synthetic Traffic

The interconnection networks can be run in a standalone manner and fed with synthetic traffic. We recommend doing this with garnet2.0

Running Garnet Standalone