[Smeagol-discuss] stay for a long time when running the transport calculations!

Thu Dec 18 16:35:34 GMT 2008

Hi MingSu. How about trying the following? 

Try setting: 

NEnergImCircle 1 
NEnergImLine 1 
NPoles 2 

In smeagol the density is computed by integrating over the contour in the complex plane. This means in the numerical implementation, that the program is going to get a Green's function for you 

a) For each NEnergImCircle in the circle defined between EnergLowestBound and an energy close to the Fermi level, plus 
b) For each of the NEnergImLine points you requested. 

This means, your density is going to be proportional to the sum over NEnergImCircle+NEnergImLine of the imaginary part of the Green's function at those energies. 

What I am suggesting you try is to determine how long it takes for you to get a couple of those energies. Once you get the time it takes in this case; call it T2, the time it will take for realistic NEnergImline and NenergImCircle will be proportional to: 

Ttotal=T2*(NenergImLine+NenergImCircle)/2. In other words, the execution time is proportional to Nencircle+Nenergimline (assuming no bias is applied of course). 

Once you determine T2, you can try increasing these values and setting a sensible energLowestBound so that you get what you need in sensible time. 

An additional comment is that smeagol is paralll over energies, so if you have two processors instead of one, the execution time will reduce by half. The memory will increase twofold, though. 

Best, 
-Salva. 

----- Original Message ----- 
From: "Mingsu Si" <mingsusi at gmail.com> 
To: "Salvador Barraza-lopez" <sbl3 at mail.gatech.edu> 
Sent: Thursday, December 4, 2008 9:01:08 PM GMT -05:00 US/Canada Eastern 
Subject: Re: [Smeagol-discuss] stay for a long time when running the transport calculations! 

Hi Salvador, 

I check the use of memory in our calculations. It uses about 3.8 G. This job submitted is running with just one cpu node. The calculation speed is very slow. How solve this case? Increase the number of cpu? Give me some suggestions, please. 

Mingsu Si 

2008/12/4 Barraza-lopez, Salvador < sbl3 at mail.gatech.edu > 

Hi Mingsu, 

The "Leads decimation" messages will occur for each lead, for each spin component, and for each k-point in the extended molecule calculation. 
Your output indicates the size of the matrices involved for the extended molecule: 2006x2006 double complex (16 bits per entry). 

The leads might have sizes relatively smaller than the extended molecule, say 500x500 or so? 
In any event, I would suggest to log in to the running nodes and execute "top" on them. That way you will have a look at the memory usage. It may be that allocation of memory is taking long, or else that your job at some point started swapping, which will make things *very* slow. 

Best regards, 
-Salvador. 

----- Original Message ----- 
From: "Mingsu Si" < mingsusi at gmail.com > 
To: smeagol-discuss at lists.tchpc.tcd.ie 
Sent: Thursday, December 4, 2008 4:23:41 AM GMT -05:00 US/Canada Eastern 
Subject: [Smeagol-discuss] stay for a long time when running the transport calculations! 

Dear all, 

When I run the transport calculations, the program stay for a long time at the the following step. What's the possible problem about it? 
Any information are welcome. 

sims 

------------------------------------------------------------------------------------------------ 
siesta: 3.64339 9.66028 52.14890 1 147 
siesta: 3.05380 0.73510 49.84533 1 148 
siesta: 1.80091 5.98288 49.81887 1 149 
siesta: 1.59871 4.60337 52.16969 1 150 
siesta: 5.14006 -1.04691 49.85855 1 151 
siesta: 6.40050 -1.63650 52.19993 1 152 
siesta: 2.26200 1.91618 52.18292 1 153 
siesta: 10.49932 -1.41163 49.85100 1 154 
siesta: 9.15384 -1.81036 52.20182 1 155 
siesta: 12.81802 0.03402 49.84911 1 156 
siesta: 13.77611 1.06014 52.19615 1 157 
siesta: 14.83247 3.60938 52.19048 1 158 

initatomlists: Number of atoms, orbitals, and projectors: 158 2006 2486 

siesta: System type = chain 

* Maximum dynamic memory allocated = 3 MB 

siesta: =============================== 
SMEAGOL Bias step = 0, V = 0.000 Ry 
Begin CG move = 0 
=============================== 

InitMesh: MESH = 128 x 128 x 240 = 3932160 
InitMesh: Mesh cutoff (required, used) = 200.000 205.308 Ry 

* Maximum dynamic memory allocated = 1085 MB 
gensvd: Leads decimation 
gensvd: Dim of H1 and S1 : 520 
gensvd: Rank of H1: 57 
gensvd: Rank of (H1,S1): 217 
gensvd: Decimated states: 246 
gensvd: Decimation from the right 
gensvd: Leads decimation 
gensvd: Dim of H1 and S1 : 520 
gensvd: Rank of H1: 57 
gensvd: Rank of (H1,S1): 217 
gensvd: Decimated states: 246 
gensvd: Decimation from the right 
----------------------------------------------------------------------------------------------------- 

_______________________________________________ Smeagol-discuss mailing list Smeagol-discuss at lists.tchpc.tcd.ie http://lists.tchpc.tcd.ie/listinfo/smeagol-discuss 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.tchpc.tcd.ie/pipermail/smeagol-discuss/attachments/20081218/295b940e/attachment.html 