"Priority flow scheduling algorithm Research" Graduate design work log

2016/4/1--2016/4/17

Completed the reading of three articles

Use the Xshell tool to install a test environment on the Hust server ns2

2016/4/22--2016/4/24
PIAs the additional patches, recompile parts.
Read the paper script code in progress.
There are many parameters in the code, need to understand the specific meaning and corresponding in the paper, but also need to combine the C + + code patches to understand the specific implementation mechanism.

(1) tcp-common-opt.tcl Tcp_pair instproc init {args}{}
How does the eval $self next $args understand? Does next not represent a function with the same name that inherits from the parent class? But this class doesn't have a parent class?

(2) Tcp_pair instproc SETTBF {TBF} {
Global NS
$self Instvar TCPs TCPR
$self Instvar San
$self Instvar TBFs

$self Set TBFS $TBF
$ns attach-tbf-agent $san $tcps $TBF
}

(3) What do $tcps set_callback $self come for? Why does TCPR not need to be set up? What is the callback specific scenario?

(4) Tcp-common-opt.tcl in 233 rows
$self set rttimes [$tcps set Nrexmit_] Two set means, [] means command substitution, $ represents the value but set Nrexmit_ (TCP. cc file represents a time-out count), should not be the Set variable value?

(5) TCP-COMMON-OPT.TCL 242-244 lines
(6) 628 row set tmp_ [Expr ceil ([$rv _nbytes value])]

2016/4/28
1. Understand the syntax of the Argparse module in Python and get a sense of result.py
2, reread Tcp-common-opt.tcl and Spine_empirical.tcl, understand how TCL and C use of the combination of how to achieve.
3, the setting of the parameters in the run_[transport]_[workload].py and the literature comparison, output test results and analysis.

2016/4/30
Using Excel to sort out the data, there are a few questions to explore:
1, the data shows that the Pfabric large data stream (10MB,) and overall average FCT measured under datamining workload decrease with the increase of the workload. (Table 9 and table 10)
2, datamining workload under the measured large data stream (10MB,) and overall the average FCT,LLDCT situation better than PIAs, in the case of big data flow well understandable, the paper seems to have mentioned (to re-return to the paper textual research, Because the data center network has the most characteristics of small and medium data streams, ... ), but the reason that overall FCT is superior to PIAs in LLDCT condition remains to be seen.

"Priority flow scheduling algorithm Research" Graduate design work log