Question
Write a program that measures the time required to create and then remove a large number of 1-byte files from a single directory. The program should creaqte files with names of the form xNNNNNN where NNNNNN is replaced by a random six-digit number. The file should be created in the random order in which their names are generated, and then deleted in increasing numberical order (i.e., an order that is different from that in which they were created). The number of files (NF) and the directory in which they are to be created should be specifiable on the commaqnd line. Measure the times required for diffferent values of NF (e.g., in the range from 10000 to 20,0000) and for different file systems (e.g. ext2, ext3, and XFS). What patterns do you observe on each file system as NF increases? Do the results change if the files are created in increasing numerical order (x000001, x000002, x000003, and so on) and then be deleted inthe same order? If so, what do you think the reason(s) might be? Again, do the results var across file-system types?
Answer
The source for this one is in benchmarkwrites.c. See runtests.sh for what I did on my system. Here's the results I got what a few filesystems that were readily available to me. Note that with NTFS, even though the FS is on an SSD, the performance was much worse. This, I would guess, is largely due to the driver being quite unoptimized (e.g. NTFS can perform well).
** Tests with ext4 FS **
== Write/Delete Benchmark ==
NF = 100000
randomized = false
time: 1.295 secs
== Write/Delete Benchmark ==
NF = 100000
randomized = true
time: 1.426 secs
** Tests with tmpfs FS **
== Write/Delete Benchmark ==
NF = 100000
randomized = false
time: 0.373 secs
== Write/Delete Benchmark ==
NF = 100000
randomized = true
time: 0.416 secs
** Tests with ntfs4 FS **
== Write/Delete Benchmark ==
NF = 100000
randomized = false
time: 16.803 secs
== Write/Delete Benchmark ==
NF = 100000
randomized = true
time: 16.378 secs