Thursday 3 May 2012

File Sorting - 2 Source 1 Destination

Introduction
A file is a series of records. A record consists of one or more fields. One of the fields in the record is called as key field which is used to uniquely identify a record. E.g. A student’s record consists of Roll Number, Name, Class, Subjects, and Marks etc. fields; of this Roll Number is the key field as it is used to uniquely identify a student. One student is one instance with one record each. These instances are called series of records. Record in a file is synonym to tuple is the database. A record consists of field(s). Tuple is described with column(s).

While working with files, if the records are of fixed length with fixed number of fields in a fixed order then a structure can be defined to store the record. If records are of variable length then we use explicit delimiters to separate two records. The delimiter should be distinct and should not be used in the record. It is always predefined.

File Sorting
A file consists of millions of records. The size of a file is usually in GBs. To sort some data, all the data to be sorted must be visible (loaded in the RAM) at the time of sorting. As RAMs are in MBs hence the entire file cannot be loaded on the RAM. On a RAM only a small portion of the file is loaded in a frame. Each frame will hold a page of the file. The size of the page is defined by the register lines of the CPU.

Consider of the millions of records from a file, you are able to load only three records at a time in a frame. Hence at some point of time you can see only those records that can fit on a page depending on size of records fetched in RAM.

This is something which we need to adjust because in array sorting, you can see all elements but in files only a fraction is visible. Hence file sorting is different than arrays.

Why to sort a file?
  1. In database, we want to remove duplicate records.
  2. We are required to process some data in order. E.g. In a bank, highest depositors should be processed first.
  3. If there is a sort merge join.
While sorting, the things which are compared should be compatible. E.g. Comparison can be between roll numbers, date of births etc.

We will sort record on particular field where ordering exists and the rest (baggage) will not be considered. A record will be considered as consisting of Key & Baggage. If key is shifted then entire record including baggage will be shifted.

For us at this learning stage, a file consists of only keys which are to be sorted. To order keys, we need minimum 2 keys visible at a time. As only a page is loaded on a frame, we have a narrow window and can see only a couple of records; hence most of the file will be unsorted.

Two source files are must for any file sorting technique. Each file we talk about will be represented by a frame in the memory. We cannot open entire file in multiple frames as we have to work under the restriction of the operating system which are multitasking and many other processes also need RAM.

Minimum four frames are required for sorting (2 Source File + 1 Destination File + 1 Code File). If less than four frames then sorting is not possible.

The general steps for sorting a file are –
1. Start
2. Mark the runs in source file
3. Till number of runs != 1
       Distribute the runs in two or more files
       Merge the runs into a destination file
4. Stop
First we will open up discussion on how to mark runs.

Marking Runs


Note
reset(F0) -> Go to the beginning of the file and start reading.
rewrite(F0) -> Go to the beginning of the file and start writing.

1. Primitive Method

Consider a source file S with key values
76, 53, 5, 15, 20, 29, 31, 37, 41, 40, 50, 49, 91, 61, 81, 76, 1, 2, 3, 4, 5, 6
Now we will go through this file and mark runs. A run is a sorted subsequence of record.

1. Read 2 records
2. Sort those 2 records
3. Place them in F0
4. Mark eor (end of run)
5. Fetch Next two record and go to step 2
6. If no more records then stop

Now the file F0 will be
53, 76 | 5, 15 | 20, 29 | 31, 37| 40, 41| 49, 50| 61, 91| 76, 81| 1, 2| 3, 4| 5, 6|EOF 

But we have not taken advantage of the sortedness of the file hence this primitive method of marking run is not useful.

2. Marking Natural Runs

1. Reset(S)
2. Rewrite(F0)
3. Read record from S s.current
4. Place in F0
5. Read next record from S as s.next
6. If no record then go to step 8
7. If s.current<s.next Then
        Place s.next in F0
        Go to step 3
   Else
        Mark eor in F0
        Place s.next in F0
        Go to step 3
   End if
8. Mark eor in F0
9. Mark eof in F0
10.Stop

Now the file F0 will be
76 | 53 | 5, 15, 20, 29, 31, 37, 41 | 40, 50 | 49, 91 | 61, 81 | 76 | 1, 2, 3, 4, 5, 6 | EOF

<A procedure in C for marking natural runs>
markRuns()
{
 int runCount=0;
 int current, next;

 if( (d = fopen("Source.txt", "r")) == NULL)
 {
  printf("Source file not found.");
  exit(1);
 }

 f1 = fopen("File0.txt","w");

 fscanf(d,"%d",¤t);

 while(1)
 {
  fscanf(d, "%d", &next);

  if(next==eof)
  {
     fprintf(f1, "%d %d %d",current,eor,eof);
     runCount++;
     break;
  }
  else
  {
   if(current>next)
   {
    fprintf(f1, "%d %d ", current, eor);
    runCount++;
   }
   else
   {
    fprintf(f1, "%d ", current);
   }
  }

  current = next;
 }

 fclose(d);
 fclose(f1);
 fclose(f2);

 return runCount;
}


As the data is naturally sorted, hence these are called “Natural Runs”. The runs will be long, fewer and variable in length. This will have fewer passes. A pass is a run through entire data from 1st till last record.

Distributing Runs in two files F1 and F2
This is a very simple method. A flag is uses to switch between files F1 and F2.

1. Set flag=true
2. Reset(F0)
3. Rewrite(F1)
4. Rewrite(F2)
5. Read an element from F0
6. If element is EOF go to step 8
7. If flag is true
        Place the element in F1
        If the element is eor
           flag=false
   Else
        Place the element in F2
        If the element is eor
           flag=true
   End if
8. Mark EOF in F1 and F2
9. Stop

<A procedure in C for distributing runs>
void distributeRuns()
{
 int switchFlag=1;
 int e;

 f1 = fopen("File1.txt", "w");
 f2 = fopen("File2.txt", "w");
 d = fopen("File0.txt", "r");

 fscanf(d,"%d",&e);

 while (e!=eof)
 {
  if(switchFlag==1)
  {
   fprintf(f1, "%d ", e);
   if(e==eor) switchFlag=2;
  }
  else
  {
   fprintf(f2, "%d ", e);
   if(e==eor) switchFlag=1;
  }

  fscanf(d,"%d",&e);
 }

 fprintf(f1, "%d",eof);
 fprintf(f2, "%d",eof);

 fclose(d);
 fclose(f1);
 fclose(f2);
}


Now the contents of the files F1 and F2 will be as follows.

F1: 76 | 5, 15, 20, 29, 31, 37, 41 | 49, 91 | 76 | EOF
F2: 53 | 40, 50 | 61, 81 | 1, 2, 3, 4, 5, 6 | EOF

Mark and Distribute as one procedure
We have passed through the source file twice, 1st time to mark runs and 2nd time to distribute runs. These were explicit mark & distribute operations using two separate sub routines. This can be combined into one subroutine Mark & Distribute. The subroutine for primitive run marking and distribution is as follows. Flag is used to switch between files F1 and F2.
1. Reset(Source)
2. Rewrite(F1)
3. Rewrite(F2)
4. do
    Read frame full of data
    Sort internally using any array sorting algorithm
    If Flag=true
      Copy in F1
      Flag = flase
    Else
      Copy in F2
      Flag = true
    End if
    Mark eor
   Till you reach end of source
5. Stop

Algorithm of subroutine for marking and distributing natural runs is as follows. When we mark end of run, we switch the file and place element. This means that two things will be done at the same time. 1. Marking of runs and 2. Distribution of runs.

1. Reset(Source)
2. Rewrite(F1)
3. Rewrite(F2)
4. Flag=true
5. s.current = Read(source)
6. While (true)
     If(flag==true)
       f=F1;
     Else
       f=F2
     End if
     
     s.next = Read(source)

     If s.next==eof

       Write(s.current,f)
       Mark eor in f
       break

     Else

       If(s.current>s.next)
         Write(s.current, f)
         Mark(eor, f)
         flag=!(flag)
       Else
         Write(s.current, f)
       End if

     End if

     s.current=s.next
   End While
7. Write(F1, eof)
8. Write(F2, eof)
9. Stop

<A sub procedure in C for marking and distributing natural runs>

markRunsAndDistribute()
{
 int runCount=0;
 int current, next;
 int switchFlag=1;

 if( (d = fopen("File0.txt", "r")) == NULL)
 {
  printf("Source file not found.");
  exit(1);
 }

 f1 = fopen("File1.txt","w");
 f2 = fopen("File2.txt", "w");


 fscanf(d,"%d",¤t);

 while(1)
 {
  fscanf(d, "%d", &next);

  if(next==eof)
  {
   if(switchFlag==1)
   {       fprintf(f1, "%d %d %d",current,eor,eof);
    fprintf(f2, "%d",eof);
    runCount++;
    break;
   }
   else
   {
    fprintf(f2, "%d %d %d",current,eor,eof);
    fprintf(f1, "%d",eof);
    runCount++;
    break;
   }
  }
  else
  {
   if(current>next)
   {
    if(switchFlag==1)
    {
     fprintf(f1, "%d %d ", current, eor);
     runCount++;
     switchFlag=2;
    }
    else
    {
     fprintf(f2, "%d %d ", current, eor);
     runCount++;
     switchFlag=1;
    }
   }
   else
   {
    if(switchFlag==1)
    {
        fprintf(f1, "%d ", current);
    }
    else
    {
        fprintf(f2, "%d ", current);
    }

   }
  }

  current = next;
 }

 fclose(d);
 fclose(f1);
 fclose(f2);

 return runCount;
}

Merging Runs
After marking runs in the raw data and distributing it into two files, the next step is to merge them and put them back into a file. We pick an element from F1 and F2 each, place the smaller in F0 and advance in the file of which element is placed. If eor is encountered in a file, then copy the remaining elements from the run of the other file. E.g we are having files F1 and F2 as follows.

F1: 76 | 5, 15, 20, 29, 31, 37, 41 | 49, 91 | 76 | EOF
F2: 53 | 40, 50 | 61, 81 | 1, 2, 3, 4, 5, 6 | EOF

Compare 76 and 53; 53<76; 
write 53 in F0; 
advance in F2; 
eor encountered; 
Copy 56 in F0; 
Mark eor; 
Take next elements 5 and 40; 
5<40;
write 5 in F0;
advance in F1;
15<40;
write 15 in F0;
and so on till eof in any of the file is encountered. After that copy the remaining runs from the other file as it is in F0.

Finally F0 will be having following data.
F0: 53, 76 | 5, 15, 20, 29, 31, 37, 40, 41, 50 | 49, 61, 81, 91 | 1, 2, 3, 4, 5, 6, 76 | EOF

The algorithm for merging runs from files F1 and F2 and placing them in F0 is as follows.
1. runCount=0;
2. Rewrite(F0)
3. Reset(F1)
4. Reset(F2)
5. Read(e1, F1)
6. Read(e2, F2)
7. While (e1 != eof and e2 != eof)
     While (e1 != eor and e2 != eor)
        If (e1<e2)
          Write(e1, F0)
          Read(e1, F1)
        Else
          Write(e2, F0)
          Read(e2, F2)
        End if
     End While
  
     If (e1==eor)
        Copy remaining elements of the run from F2 in F0
     Else
        Copy remaining elements of the run from F1 in F0
     End if

     Write(eor, F0)
     Runcount++

     Read(e1, F1)
     Read(e2, F2)
8. End While
9. If(e1==eof)
     While (e2 != eof)
        Write(e2,F0)
        If(e2==eor) runCount++;
        Read(e2,F2)
     End While
   Else if(e2==eof)
     While (e1 != eof)
        Write(e1,F0)
        If(e1==eor) runCount++;
        Read(e1,F1)
     End While
   End if
10. Mark EOF in S0
11. Return runCount

<A sub procedure in C for merging runs>

mergeRuns()
{
 int e1,e2;
 int runCount=0;

 f1 = fopen("file1.txt","r");
 f2 = fopen("file2.txt","r");
 d = fopen("File0.txt","w");

 fscanf(f1,"%d",&e1);
 fscanf(f2,"%d",&e2);

 while (e1!=eof && e2!=eof)
 {
  while(e1!=eor && e2!=eor)
  {
   if(e1<e2)
   {
    fprintf(d,"%d ",e1);
    fscanf(f1,"%d",&e1);
   }
   else
   {
    fprintf(d,"%d ",e2);
    fscanf(f2,"%d",&e2);
   }
  }

  if(e1==eor)
  {
   while (e2!=eor)
   {
    fprintf(d, "%d ", e2);
    fscanf(f2, "%d", &e2);
   }
  }
  else if(e2==eor)
  {
   while (e1!=eor)
   {
    fprintf(d, "%d ", e1);
    fscanf(f1, "%d", &e1);
   }
  }

  fprintf(d, "%d ", eor);
  runCount++;

  fscanf(f1,"%d",&e1);
  fscanf(f2,"%d",&e2);
 }
 if(e1==eof)
 {
  while (e2!=eof)
  {
   fprintf(d, "%d ", e2);
   if(e2==eor) runCount++;
   fscanf(f2, "%d", &e2);
  }
 }
 else if(e2==eof)
 {
  while (e1!=eof)
  {
   fprintf(d, "%d ", e1);
   if(e1==eor) runCount++;
   fscanf(f1, "%d", &e1);
  }
 }

 fprintf(d,"%d",eof);

 fclose(f1);
 fclose(f2);
 fclose(d);

 return runCount;
}

This process of merge and distribute continues till the number of runs in file F0 become 1. At this time the data is completely sorted.

If we start with two files each having runs = N; then after 1 merge and distribute each file will have N/2 runs. After another merge and distribute each file will have N/4 runs. This continues till the run count becomes 1.

Hence for N runs the number of passes, P = log2N Similarly if we know the passes P then number of runs N will be 2P-1 < N <= 2P.


Finally, a complete program in C to demonstrate file sorting using two source and one destination.
#include<stdio.h>
#include<conio.h>
#include<stdlib.h>

#define eor -1
#define eof -2

FILE *f1, *f2, *d;

markRuns();
mergeRuns();
void distributeRuns();

void main()
{

     markRuns();
     distributeRuns();

     while(mergeRuns()!=1)
     {
        distributeRuns();
     }

     printf("File Sorted....");
     getch();
}


markRuns()
{
 int runCount=0;
 int current, next;

 if( (d = fopen("Source.txt", "r")) == NULL)
 {
  printf("Source file not found.");
  exit(1);
 }

 f1 = fopen("File0.txt","w");

 fscanf(d,"%d",¤t);

 while(1)
 {
  fscanf(d, "%d", &next);

  if(next==eof)
  {
     fprintf(f1, "%d %d %d",current,eor,eof);
     runCount++;
     break;
  }
  else
  {
   if(current>next)
   {
    fprintf(f1, "%d %d ", current, eor);
    runCount++;
   }
   else
   {
    fprintf(f1, "%d ", current);
   }
  }

  current = next;
 }

 fclose(d);
 fclose(f1);
 fclose(f2);

 return runCount;
}

void distributeRuns()
{
 int switchFlag=1;
 int e;

 f1 = fopen("File1.txt", "w");
 f2 = fopen("File2.txt", "w");
 d = fopen("File0.txt", "r");

 fscanf(d,"%d",&e);

 while (e!=eof)
 {
  if(switchFlag==1)
  {
   fprintf(f1, "%d ", e);
   if(e==eor) switchFlag=2;
  }
  else
  {
   fprintf(f2, "%d ", e);
   if(e==eor) switchFlag=1;
  }

  fscanf(d,"%d",&e);
 }

 fprintf(f1, "%d",eof);
 fprintf(f2, "%d",eof);

 fclose(d);
 fclose(f1);
 fclose(f2);
}

mergeRuns()
{
 int e1,e2;
 int runCount=0;

 f1 = fopen("file1.txt","r");
 f2 = fopen("file2.txt","r");
 d = fopen("File0.txt","w");

 fscanf(f1,"%d",&e1);
 fscanf(f2,"%d",&e2);

 while (e1!=eof && e2!=eof)
 {
  while(e1!=eor && e2!=eor)
  {
   if(e1<e2)
   {
    fprintf(d,"%d ",e1);
    fscanf(f1,"%d",&e1);
   }
   else
   {
    fprintf(d,"%d ",e2);
    fscanf(f2,"%d",&e2);
   }
  }

  if(e1==eor)
  {
   while (e2!=eor)
   {
    fprintf(d, "%d ", e2);
    fscanf(f2, "%d", &e2);
   }
  }
  else if(e2==eor)
  {
   while (e1!=eor)
   {
    fprintf(d, "%d ", e1);
    fscanf(f1, "%d", &e1);
   }
  }

  fprintf(d, "%d ", eor);
  runCount++;

  fscanf(f1,"%d",&e1);
  fscanf(f2,"%d",&e2);
 }
 if(e1==eof)
 {
  while (e2!=eof)
  {
   fprintf(d, "%d ", e2);
   if(e2==eor) runCount++;
   fscanf(f2, "%d", &e2);
  }
 }
 else if(e2==eof)
 {
  while (e1!=eof)
  {
   fprintf(d, "%d ", e1);
   if(e1==eor) runCount++;
   fscanf(f1, "%d", &e1);
  }
 }

 fprintf(d,"%d",eof);

 fclose(f1);
 fclose(f2);
 fclose(d);

 return runCount;
}

No comments:

Post a Comment

Your comments are very much valuable for us. Thanks for giving your precious time.

Do you like this article?