Sorting by the Numbers Sorting Part Four

Question Suppose you are given the task of writing an application to sort a big data file. What do you need to know to pick a good solution?  File Size = 1 GB  Record Size = 250 Bytes  Available Memory = ¼ GB

How many Runs? How big is each Run? Total Records to Process  1 billion bytes in the file  250 bytes for each record  = 4 million records in the file Run Size  1GB file  ¼ GB memory  = 4 Runs of 1 million records each

Time to Create the Runs Sorting One Run  Using either Quicksort or Ordered Binary Tree N log 2 N 1million * 20  approximately 20 million comparisons of internal memory locations Sorting Four Runs  80 million internal memory comparisons

Refresher on Merging Files So, to merge 2 files of N random records each, requires 2N compares And, to merge 2 files where the runs were built from a sorted file requires N compares File One File Two File One File Two

Merging the Four Files R1R2 T2 R3 T1 R4R1R2 T2 R3T1 R4 2 million compares 4 million compares 3 million compares 2 million compares 4 million compares

Total Processing Time Time to Create the 4 Runs  80 million comparisons Time to Merge the 4 Runs  8 million comparisons Assuming a File Read takes just 100 times longer than a Memory Read  Total Time = 880 million time units  note, we have omitted the time to read the runs into memory and to write the runs to temp files

Second Example 2 Runs of 2 Million Records each 2 Runs of 2 Million Records each  Internal Sorting N log2 N = 2million * 24 = 48 million compares 96 million to create both runs  File Merging 4 million compares  Total Time 496 million time units 496 million time units

Next in this course So how much time does it take to access the disk?