Remote parallel computation

class: center, middle, inverse, title-slide

# Remote parallel computation
### Facundo Muñoz <a href="mailto:facundo.munoz@cirad.fr">facundo.munoz@cirad.fr</a> famuvie
### Orléans, Sep. 18, 2018

---

background-image: url(img/remote_computation.png)

---
## Remote computation

- You have access to a **Linux** server through **SSH**

- You can perform `breedR`'s computations __remotely__

- Take advantage of more **memory** or **faster** processors

- **Parallelize** jobs

- Free **local resources** while fitting models

- See `?remote` for details

---
# Configuring a server

1. Windows users: install `cygwin` with `ssh` beforehand (http://cygwin.org/)

1. configure the client and server machines so that password-less [SSH authentication works](http://www.thegeekstuff.com/2008/11/3-steps-to-perform-ssh-login-without-password-using-ssh-keygen-ssh-copy-id/)

1. Set breedR options `remote.host`, `remote.user`, `remote.port` and `remote.bin` (see `?breedR.setOption`)

- Optionally, set these options permanently in `$HOME/.breedRrc`

```r
writeLines(
  c("remote.host = '147.99.222.196'",
    "remote.user = 'yourusername'",
    "remote.bin  = '/usr/local/lib/R/site-library/breedR/bin/'"),
  con = file.path(Sys.getenv('HOME'), '.breedRrc'))
```

---

## Fitting models remotely

```r
res <- remlf90(..., breedR.bin = "remote")
```

- Fit model **remotely**

- R-console stays in **stand-by** until job is finished

- When job finishes (provided that connection keeps alive), results are automatically **retrieved**

### Identical in use to local computing, but without the processor/memory burden

---

## Submitting jobs

```r
res <- remlf90(..., breedR.bin = "submit")
```

- Fit model **remotely**

- Connection is **closed** in the meanwhile

- R-console is **active**

- Typing `res` **queries** the server for the job status (Running/Finished/Aborted)

- __Retrieve__ results with `breedR.qget(job_id)`

---

## Parallel computing

- After you **submit** a job, you are free to submit more (specially with multiple-processor servers)

- Query the **status** of all jobs with `breedR.qstat()`

- **Kill** some job with `breedR.qdel(res)` or all jobs with `breedR.qnuke()`

---
class: inverse, center, middle
background-image: url(img/breedRhex.png)
background-position: 50% 90%

# Remote parallel computation