|
When we say NetMiner 3 supports ‘Large Network Data’, what does it mean? We explore the exact meaning of “NetMiner 3 supports large network data analysis” through four questions and answers below.
- Do all licenses of NetMiner 3 support ‘Large Network Data’?
Not true. The license of NetMiner 3 consists four levels (Tiny, Small, Medium, Large) in terms of data size to be handled. For processing ‘Large Network Data’, the appropriate level of NetMiner 3 license is required. Currently, the maximum size is 100,000 nodes.
- If I bought the appropriate level of NetMiner 3 license, would all ‘large network data’ be supported?
Not true. The level of data size just limits the maximum nodes to be opened in NetMiner 3. But the size of network data is defined not only by the number of nodes but also by the number of links. At this point we need an operational definition of ‘Sparse Network’ to define the upper limit of link size. NetMiner 3 defines ‘Sparse Network’ as the network of ‘m ≤ 10 x n’, where ‘n’ is the number of nodes and ‘m’ is the number of links. So NetMiner 3 supports ‘Large but Sparse Network’ where the number of links is smaller than ten times the node size. Currently, the maximum size is 1,000,000 links. Of course, NetMiner 3 can run data having more than 1,000,000 links. It is just not guaranteed and tested officially.
- If I bought the appropriate level of NetMiner 3 license and data was in the range of ‘Large but Sparse Network’, could I use all analysis algorithms for it?
Not true. NetMiner 3 guarantees for the performance of algorithms which is subquadratic in terms of time complexity. An algorithm is said to run in subquadratic time if its running time f(n) is under o(n2). Subquadratic algorithms in NetMiner 3 are listed in the test result below. Of course, overquadratic algorithms can be run in NetMiner. But it does not guarantee fast and smooth processing of ‘Large Network Data’. We will offer the results of performance test in the case of overquadratic algorithms.
- If I bought the appropriate level of NetMiner 3 license and data was in the range of ‘Large but Sparse Network’, and if I ran subquadratic algorithms of NetMiner 3, could they be always processed with high performance?
Not true. The final factor for the support of ‘Large Network Data’ is the performance of hardware of computer NetMiner 3 is installed in. The size of RAM and the speed of CPU, etc affect the performance of NetMiner 3. Of course, NetMiner 3 can be run in computer with lower-level specifications.
To summarize, When we say NetMiner 3 supports ‘Large Network Data‘, we mean that i) provided that user have appropriate level of license and, ii) the network data is sparse (m ≤ 10 x n) and, iii) users are running subquadratic algorithms and, iv) NetMiner 3 is installed in a computer with large size of RAM and high speed of CPU, NetMiner 3 can handle ‘Large Network Data‘ fast and smoothly.
The contents below are the results of performance test using ‘Large but Sparse Network Data’ under a normal computing environment.
- The results of performance test of NetMiner 3 -
The contents of this document are regarding the performance of NetMiner 3 under ‘Large Network Data’ condition. Although the same conditions are given to the other test, the different results may be occurred.
1. Data for test
Single network data with properties below was used for test.
- Directed network
- Weighted network
- Multiple links are allowed between two nodes
- No Link Attributes nor Node Attributes were loaded unless analysis requires it(for example, Blockmodeling or Node Extract)
# of Nodes |
100,000 |
# of Links |
1,012,705 |
Average Degree |
10.127 |
Network Density |
0.01% |
2. Computing Environments
NetMiner 3 was run under following environments, with other applications occupying approximately 800MB of Memory.
CPU |
Intel® Core 2 6400, 2.13GHz |
RAM |
2GB |
HDD |
149GB, 7200rpm |
OS |
Microsoft Windows XP® SP2 |
3. Selecting processing modules to test
All subquadratic algorithms in ‘Transform’ and ‘Analyze’ menu are tested.
4. Measuring Elapsed time
Only time for loading and running algorithms (from starting command to generating report pages) was measured, and time for generation of report pages was not included. ‘Elapsed Time’ is averaged over 5 runs. 5. Results
Menu |
Selection |
Time
Complexity |
Elapsed
Time |
Top-Level |
2nd Level |
3rd Level |
Analyze |
Neighbor |
Degree |
|
O(m) |
< 10 sec. |
Neighbor |
Assortativity |
|
O(m) |
< 10 sec. |
Cohesion |
Component |
Weak |
O(m) |
< 10 sec. |
Strong |
O(m) |
< 10 sec. |
Bi-Component |
|
O(m) |
< 10 sec. |
Centrality |
Degree |
|
O(m) |
< 10 sec. |
Coreness |
|
O(m) |
< 10 sec. |
Properties |
Network |
Number of Links |
O(m) |
< 10 sec. |
Density |
O(m) |
< 10 sec. |
Avg. Degree |
O(m) |
< 10 sec. |
Inclusiveness |
O(m) |
< 10 sec. |
Reciprocity |
O(m) |
< 10 sec. |
Connectedness |
O(m) |
< 10 sec. |
Efficiency |
O(m) |
< 10 sec. |
Position |
Blockmodel |
|
O(m) |
< 30 sec. |
Transform |
Direction |
Symmetrize |
|
O(m) |
< 10 sec. |
Transpose |
|
O(m) |
< 10 sec. |
Value |
Dichotomize |
|
O(m) |
< 10 sec. |
Reverse |
|
O(m) |
< 10 sec. |
Normalize |
|
O(m) |
< 10 sec. |
Recode |
|
O(m) |
< 10 sec. |
Node Set |
Ego Networks |
|
O(m) |
< 10 sec. |
Reorder |
|
O(n logn) |
< 10 sec. |
Link Set |
Incidence |
|
O(m) |
< 10 sec. |
(n: the number of nodes, m: the number of links) |
