PANINI

1
PANINI

• Harshit Sahay
• Shuchi Agarwal
• Sumeet Jindal

2
Introduction

• Panini is a language for parallel computing over
  multiple machines
• We have named our language after the Sanskrit
  grammarian, Panini
• Using Panini the task is distributed among
  different machines specified by the user

3
Features of Panini

• Panini is a declarative language
• Panini is simple and easy-to-use
• Panini is flexible

4
Panini Architecture
VC compiler
Programmers VC project
sample.cpp
Panini compiler
sample.panini
5
Communication between the machines
Servicing machine
output2
Requesting machine
Libnamefuncnamedata2
output1
Servicing machine
output0
Libnamefuncnamedata1
Libnamefuncnamedata0
Servicing machine
6
A sample PANINI program

• PANINI
• CREATE NODE n1"209.2.235.32"
• CREATE NODE n2"209.2.226.71"
• LIBNAME D\libdll.dll"
• FUNCTIONNAME "word_count"

7
sample.cpp

• char n1"209.2.235.32"
• char n2"209.2.226.71"
• bstr_t nodes_array"209.2.235.32","209.2.226.71
  "
• void Create_Queue()
• void libdll_word_count_Fire(char
  function_inputs,int number_of_tasks,char
  guids)
• CString library_name"libdll.dll"
• CString function_name"word_count"
• IMSMQQueueInfoPtr spQInfo("MSMQ.MSMQQueueInfo
  ")
• IMSMQQueuePtr spQSend
• for(loop_i0loop_iltnumber_of_tasksloop_i)
  
• IMSMQMessagePtr spQMsg("MSMQ.MSMQMessa
  ge")
• spQMsg-gtLabel g
• spQMsg-gtBody BSTR(paniniMessage)
• spQMsg-gtDelivery MQMSG_DELIVERY_RECOVERABLE

8
sample.cpp(contd.)
void Wait_For_Output(char guids,int
number_of_guids,char output) ) IMSMQQueueInf
oPtr spQInfo1("MSMQ.MSMQQueueInfo") IMSMQMessag
ePtr spRMsg spRMsg spQRec-gtPeekCurrent(vtMiss
ing, vtMissing) while(1)
for(guid_i0guid_iltnumber_of_guidsguid_i)
if(_stricmp((char)spRMsg-gtLabel,guidsguid
_i)0) num_of_guids_matched
break if(guid_iltnumber_of_gui
ds) spRMsgspQRec-gtReceiveCurrent(vtMissin
g, vtMissing, vtMissing)
temp_strspRMsg-gtBody
while(altlength) outputguid_ia(char)tem
p_str.GetAt(a) if(num_of_guids_matched!numbe
r_of_guids) spRMsg spQRec-gtPeekCurrent(vt
Missing, vtMissing) else
spRMsgspQRec-gtPeekNext(vtMissing,
vtMissing)
9
Including and using Panini output in user code

• //user_code.cpp
• includesample.cpp//output of Panini
• int _tmain(int argc, _TCHAR argv)
• CoInitialize(NULL)
• char arr"C\\data1.txt","C\\data2.txt","C\
  \data3.tx t","C\\data4.txt"
• char output8
• char guids8
• Create_Queue() libdll_word_count_Fire(arr,4,gu
  ids) Wait_For_Output(guids,4,output)
• int no_words0
• for(int i0ilt7i)
• no_wordsatoi(outputi)
• coutltlt"Total number of words in all
  files"ltltno_wordsltltendl
• CoUninitialize()
• return 0

• //sample.cpp
• void Create_Queue()
• ...
• void libdll_word_count_Fire(char
  function_inputs,int number_of_tasks,char
  guids)
• ...
• void Wait_For_Output(char guids,int
  number_of_guids,char output)
• ...

10
Keywords of Panini

• PANINI
• CREATE
• NODE
• LIBNAME
• FUNCTIONNAME

11
Service running on all machines

• int Receive()
• /OPEN AND RECEIVE MESSAGE
  BEGIN/
• IMSMQMessagePtr ReceiveMsg ReceiveQueuePtr-gtRe
  ceive(vtMissing, vtMissing, vtMissing,
  vtReceiveTimeout)
• /PARSE RECEIVED MESSAGE
  BEGIN/
• while (resToken ! "")
• if(i 1) ipaddr resToken
• if(i 2) libName (LPCWSTR)resToken
• if(i 3) funcName resToken
• if(i 4) inputToFunc resToken
• resToken msg.Tokenize(_T(""), cur_pos)
• i
• /LOAD LIB FUNCTION CALL
  BEGIN/
• FunctionPtr (FunctionType)GetProcAddress((HMOD
  ULE)hLib, funcName_1)
• FunctionPtr(input_to_function,
  output_from_function)
• /SEND MESSAGE BEGIN
  /
• BSTR bstr_paniniMessageOut outMsg.AllocSysStri
  ng()
• ipaddr "DIRECTTCP" ipaddr
  "\\PRIVATE\\listening_queue"
• Send(bstr_FormatName, bstr_guid,
  bstr_paniniMessageOut)
• return 0

12
Development tools and environment

• Visual Studio 2005
• Microsoft Message Queueing(MSMQ)
• Lex and Yacc
• Google Docs

13
Test Plan

• Every grammar production was tested
• Testing communication between machines was a
  major component of testing

14
Conclusion

• Great learning experience
• Learnt about message queuing
• Realized that implementing parallel computing is
  not so easy
• We achieved our basic objective-to create a
  language for parallelizing tasks
• So, if you have some set of instructions which
  you want to easily execute over multiple sets of
  data, then use our language