1 |
54 |
alirezamon |
= SynFull Models
|
2 |
|
|
|
3 |
|
|
Several models are provided for you that have been generated based on ideal network traces as described in the paper.
|
4 |
|
|
These models are for a variety of PARSEC and SPLASH-2 benchmarks.
|
5 |
|
|
The model files themselves are in text format, so you can see the actual raw values that were modelled.
|
6 |
|
|
The code takes these values and converts them into probability distributions (with the exception of the Markov probability matrices).
|
7 |
|
|
|
8 |
|
|
== Manually Changing Model Data
|
9 |
|
|
|
10 |
|
|
You can potentially change the values found in the model files to create your own custom model, however I have not experimented with this.
|
11 |
|
|
You should also be careful, as the traffic generator expects certain values for certain parts of the model.
|
12 |
|
|
I would be weary of changing the Markov probabilities, for example, but changing the volume distributions (e.g. WRITE_INJECTION) is probably harmless.
|
13 |
|
|
|
14 |
|
|
== Model Description
|
15 |
|
|
|
16 |
|
|
The model files are divided into several sections with different headers.
|
17 |
|
|
Simple headers act as name-value pairs, while complex headers have a BEGIN and END and usually identify some probability distribution.
|
18 |
|
|
A description for each of these headers can be found below.
|
19 |
|
|
|
20 |
|
|
=== Simple Headers
|
21 |
|
|
|
22 |
|
|
HIER_CLASSES:: The number of macro phases in the model file.
|
23 |
|
|
TIME_SPAN:: The number of cycles long each macro phase is.
|
24 |
|
|
MEMORY:: The memory the Markov chain supports (currently, only 1 is supported).
|
25 |
|
|
NUM_NODES:: The number of nodes in the simulation (i.e. sources and destinations).
|
26 |
|
|
NUM_CLASSES:: The number of micro phases for a given macro phase.
|
27 |
|
|
RESOLUTION:: The number of cycles long each micro phase is.
|
28 |
|
|
|
29 |
|
|
=== Complex Headers
|
30 |
|
|
|
31 |
|
|
HIER_BEGIN_ID:: The micro model for the given macro phase begins after this line.
|
32 |
|
|
HIER_MARKOV:: The transition probability matrix for macro phases (Markov Chain).
|
33 |
|
|
HIER_MARKOV_STEADY:: The Markov chain's steady state for macro phases.
|
34 |
|
|
MARKOV:: The transition probability matrix for micro phases (Markov Chain).
|
35 |
|
|
MARKOV_STEADY:: The Markov chain's steady state for micro phases (not used).
|
36 |
|
|
*_SPATIAL:: The spatial injection distribution (i.e. who injects) per message type.
|
37 |
|
|
*_FLOWS:: The flow injection distribution (i.e. destinations) per message type.
|
38 |
|
|
*_INJECTION:: The volume distributions per message type. Starts at zero.
|
39 |
|
|
FORWARD_PROB:: The probability a directory forwards a request.
|
40 |
|
|
FORWARD_FLOWS:: The flow injection distribution for forwarded requests.
|
41 |
|
|
INVALIDATE_PROB:: The volume distribution for invalidates at each message type.
|
42 |
|
|
INVALIDATE_FLOWS:: The flow injection distribution for invalidates at each directory.
|