Permalink
Browse files

rename server to router (it is actually just a router)

  • Loading branch information...
PatWie committed Jan 17, 2018
1 parent 9157212 commit 25f0bf1a12041a9a217dcf20fa3f6752aab047a5
Showing with 7 additions and 7 deletions.
  1. +1 −1 Makefile
  2. +3 −3 README.md
  3. +1 −1 cluster-smi-node.go
  4. +1 −1 cluster-smi-server.go → cluster-smi-router.go
  5. +1 −1 cluster-smi.go
View
@@ -1,5 +1,5 @@
all:
go build cluster-smi.go config.go cluster.go
go build cluster-smi-server.go config.go
go build cluster-smi-router.go config.go
go build cluster-smi-node.go config.go cluster.go
go build cluster-smi-local.go config.go cluster.go
View
@@ -30,7 +30,7 @@ Additional information are available, when using `cluster-smi -p -t`.
<p align="center"> <img src="./cluster-smi.jpg" width="100%"> </p>
Each machine you want to monitor need to start *cluster-smi-node* (e.g. using systemd). They are sending information from the nvidia-driver to a *cluster-smi-server*, which further distribute these information to client (*cluster-smi*). Only the machines running *cluster-smi-node* require CUDA dependencies.
Each machine you want to monitor need to start *cluster-smi-node* (e.g. using systemd). They are sending information from the nvidia-driver to a *cluster-smi-router*, which further distribute these information to client (*cluster-smi*). Only the machines running *cluster-smi-node* require CUDA dependencies.
You might be interested as well in [cluster-top](https://github.com/PatWie/cluster-top) for CPUS.
@@ -76,7 +76,7 @@ To obtain a portable small binary, I suggest to directly embed the configuration
```go
...
c.ServerIp = "127.0.0.1"
c.RouterIp = "127.0.0.1"
c.Tick = 3
c.Timeout = 180
c.Ports.Nodes = "9080"
@@ -96,7 +96,7 @@ make all
## Run
1. start `cluster-smi-node` at different machines having GPUs
2. start `cluster-smi-server` at a specific machine (machine with ip-addr: `cluster_smi_server_ip`)
2. start `cluster-smi-router` at a specific machine (machine with ip-addr: `cluster_smi_router_ip`)
3. use `cluster-smi` like `nvidia-smi`
Make sure, the machines can communicate using the specifiec ports (e.g., `ufw allow 9080, 9081`)
View
@@ -23,7 +23,7 @@ func main() {
defer nvml.ShutdownNVML()
// sending messages (PUSH-PULL)
SocketAddr := "tcp://" + cfg.ServerIp + ":" + cfg.Ports.Nodes
SocketAddr := "tcp://" + cfg.RouterIp + ":" + cfg.Ports.Nodes
log.Println("Now pushing to", SocketAddr)
socket, err := zmq4.NewSocket(zmq4.PUSH)
if err != nil {
@@ -25,7 +25,7 @@ func main() {
var mutex = &sync.Mutex{}
// message loop
log.Println("Cluster-SMI-Server is active. Press CTRL+C to shut down.")
log.Println("Cluster-SMI-Router is active. Press CTRL+C to shut down.")
// receiving messages in extra thread
go func() {
View
@@ -41,7 +41,7 @@ func main() {
}
defer request_socket.Close()
SocketAddr := "tcp://" + cfg.ServerIp + ":" + cfg.Ports.Clients
SocketAddr := "tcp://" + cfg.RouterIp + ":" + cfg.Ports.Clients
request_socket.Connect(SocketAddr)
for {

0 comments on commit 25f0bf1

Please sign in to comment.