cancel
Showing results for 
Search instead for 
Did you mean: 

Cost GB memory for SNP optimizer and how far can push?

former_member584840
Participant
0 Kudos

Hi, We have ramped up our test SNP optimizer from 2GB to 8 GB memory, but it seems to be running at about 17GB using the full 8GB memory and grabbing swap space as well. (That only trying to run a small subset of our data with lots of details in the settings like discretization and lot sizes) Basis tells me that 8GB is big, but not being that technical in this area, and reading in the papers that memory is cheap now, and a quick google search mentions memory has droppod from $5000 per GB a few years back to under $100, I am wondering what the issue is, why dont we just ramp it up to 100GB and forget about it. But can anyone tell me the real story? ie vague approximation of cost of memory for a windows box  with an SNP optimizer on it (or any similar windows box) and are there any limitations? ie is it simply unlimited? Or can you only add so much memory before you run into problems? And does anyone have some big examples of how far you have pushed your SNP optimizer?

Accepted Solutions (0)

Answers (2)

Answers (2)

rajkj
Active Contributor
0 Kudos

Hi John,

We performed a similar exercise to verify the results of linear and discrete optimization with same input (demands, quotas, cost factors, etc.using SAP SCM Release 5.1) for a known scenario. Quality and accuracy of discrete optimization results are slightly superior compared to linear, but the response time is almost 3 times more and the execution was more resource intensive (CPU and RAM).

Thanks,
Rajesh

Former Member
0 Kudos

We found we were running out of memory too quickly.  There is a parm that can be specified to modify the CPLEX behaviour, and after consuming a few GB of RAM, it then starts building the remainder of the tree on disk.  Largest tree to date have been just under 60GB (4GB in RAM and 56GB on disk).  Although disk is slower than RAM, it is cheaper.

Former Member
0 Kudos

Hi John

SNP Optimizer can be very resource intensive depending on multiple factors like constraints, data volume etc.  I do not have a straight forward answer to your question neither I am an expert in Basis. However, from your description it looks like you are using Single Gateway i.e. dedicated server. From cost perspective, it is probably cheaper to use cloud as a solution. You can get the details in wiki

http://wiki.sdn.sap.com/wiki/display/SCM/SCM+Optimizers+in+the+Cloud+White+Paper+-+EC2+Image

Thanks,

Prasun