Regarding REMD using cp2k and i-pi

[Arzoo Jangra]

Hello folks,

I have been trying to run REMD using cp2k and i-pi but the process is extremely slow. I don't know how to speed up these calculations.I am attaching my.xml file and input.inp file. Kindly tell me how can i speed up the process as for now it is running only 130 steps in 4 days.

Thanking You

Sincerely

Arzoo

[Mariana Rossi]

Hi Arzoo, there is nothing wrong in general, but as a curiosity, for the system you are trying to calculate (I don't know which one it is) how long does one force evaluation in CP2K take? 
If you launch i-PI with 4 CP2K clients with your setup, you should be taking as much time as one force evaluation for your system. If you launch it with a single client, you should take 4 times as much. Hope it helps.

[Michele Ceriotti]

hard to tell why it is so slow, but it should depend entirely on the time taken for each cp2k evaluation. with your input it needs to evaluate 5 energies per step, and if you have a single cp2k instance running, your 130 steps should be about 650 energy evaluations. 4 days seems extremely slow for a small water box, at the level I'd expect if you were running on a single CPU. so I recommend you first check the timings for running cp2k on its own, checking it's actually compiled with parallel execution. then, as Mariana says, you can run multiple cp2k instances, but this is far too slow to start with, and it's a problem of your cp2k setup, not of the interaction with i-pi

cheers

michele

[Mariana Rossi]

yes, when I wrote 4 I meant 5, like Michele. miscounted the number of replicas you had!

[Arzoo Jangra]

Hello folks,

First of all I want to thank Marian and Michele for your fruitful inputs but now the problem is it works fine till 12 steps then again it stops( for HCl in 6 water). I am pasting the error message in this email. For reference I am also attaching my input files and job submission scripts in this email.

Kindly help regarding this issue.

Thanking You

Sincerely

Arzoo

!W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again
 !W!  @SOCKET: Couldn't find readable socket in 60s, will try again

[Michele Ceriotti]

I don't know why it's dying after a few steps, but you don't need to have separate ffsockets. You can have multiple cp2k instances connect to the same socket - so basically one single 
```
<ffsocket name="cp2k" mode="inet">

<address> IP_ADDRESS </address>
<port>31415</port>
<timeout>600</timeout>
<latency>0.01</latency>
</ffsocket>

```
block, and then no need to have cp2k_IDX in the system, just
```
<forces>
<force forcefield="cp2k"/>
</forces>
```

Then you can launche multiple cp2k instances, all connecting to port 31415.

This should make simulations a bit more resilient. From what you say it's likely that one of the cp2k instances crashed.

Best,

Michele