Take the 2-minute tour ×
Unix & Linux Stack Exchange is a question and answer site for users of Linux, FreeBSD and other Un*x-like operating systems.. It's 100% free, no registration required.

The commands below may takes minutes depends on the file size. Is there any more effient method?

sed -i 1d large_file 
share|improve this question

4 Answers 4

up vote 15 down vote accepted

Try ed instead:

ed <<< $'1d\nwq' large_file

If that “large” means about 10 million lines or more, better use tail. Is not able for in-place editing, but its performance makes that lack forgivable:

tail -n +2 large_file > large_file.new

Edit to show some time differences:

(awk code by Jaypal added to have execution times on the same machine (CPU 2.2GHz).)

bash-4.2$ seq 1000000 > bigfile.txt # further file creations skipped

bash-4.2$ time sed -i 1d bigfile.txt
time 0m4.318s

bash-4.2$ time ed -s <<< $'1d\nwq' bigfile.txt
time 0m0.533s

bash-4.2$ time perl -pi -e 'undef$_ if$.==1' bigfile.txt
time 0m0.626s

bash-4.2$ time { tail -n +2 bigfile.txt > bigfile.new && mv -f bigfile.new bigfile.txt; }
time 0m0.034s

bash-4.2$ time { awk 'NR>1 {print}' bigfile.txt > newfile.txt && mv -f newfile.txt bigfile.txt; }
time 0m0.328s
share|improve this answer
    
Thanks. I'll use tail. –  Cheng Nov 29 '11 at 12:08
    
In case of tail, I would rather count the time to do both remove the first line and replace bigfile.txt with bigfile.new. –  rozcietrzewiacz Nov 29 '11 at 14:30
    
@rozcietrzewiacz, your point is correct. Thank you. Updated. –  manatwork Nov 29 '11 at 14:57
    
This is really cool! I did the same with awk and got the following result - [jaypal:~/Temp] seq 1000000 > bigfile.txt [jaypal:~/Temp] time awk 'NR>1 {print}' bigfile.txt >newfile.txt real 0m0.649s user 0m0.601s sys 0m0.033s –  jaypal Nov 29 '11 at 20:29
1  
@Jaypal, I added your code to the list of alternatives. On my machine it was even faster. Strange, I expected awk's performance to be closer to sed's. (Note to myself: never expect – test instead.) –  manatwork Nov 30 '11 at 7:49

There is no way to efficiently remove things from the start of a file. Removing data from the beginning requires re-writing the whole file.

Truncating from the end of a file can be very quick though (the OS only has to adjust the file size information, possibly clearing up now-unused blocks). This is not generally possible when you try to remove from the head of a file.

It could theoretically be "fast" if you removed a whole block/extent exactly, but there are no system calls for that, so you'd have to rely on filesystem-specific semantics (if such exist). (Or having some form of offset inside the first block/extent to mark the real start of file, I guess. Never heard of that either.)

share|improve this answer
    
If the file is very large, I/O overhead is likely to be (possibly much) greater than the CPU overhead required to process end of lines. –  Mat Nov 29 '11 at 10:46
    
You are right. However there could be difference in the way the tools access the file content. The best is not processing line by line when not necessary or at least not reading line by line when not necessary. –  manatwork Nov 29 '11 at 10:56
2  
I'm surprised the difference is so big in your results, and can reproduce it with that file size here. Benefits seem to decrease as the file size increases though (tried with seq 10M, 15s for sed, 5s for ed). Good tips anyway (+1). –  Mat Nov 29 '11 at 11:13
    
Starting with version 3.15, Linux now has an API to collapse parts of a file on some extent based file systems, but at least for ext4 that can only be done on full blocks (usually 4k). –  Stéphane Chazelas 8 hours ago

The most efficient method, don't do it ! If you do, in any case, you need twice the 'large' space on disk, and you waist IOs.

If you are stuck with a large file that you want to read without the 1st line, wait until you need to read it for removing the 1st line. If you need to send the file from stdin to a program, use tail to do it:

tail -n +2 | your_program

When you need to read the file, you can take the opportunity to remove the 1st line, but only if you have the needed space on disk:

tail -n +2 | tee large_file2 | your_program

If you can't read from stdin, use a fifo:

mkfifo large_file_wo_1st_line
tail -n +2 large_file > large_file_wo_1st_line&
your_program -i large_file_wo_1st_line

of even better if you are using bash, take advantage of process substitution:

your_program -i <(tail -n +2 large_file)

If you need seeking in the file, I do not see a better solution than not getting stuck with the file in the 1st place. If this file was generated by stdout:

large_file_generator | tail -n +2 > large_file

Else, there is always the fifo or process substitution solution:

mkfifo large_file_with_1st_file
large_file_generator -o large_file_with_1st_file&
tail -n +2 large_file_with_1st_file > large_file_wo_1st_file

large_file_generator -o >(tail -n 2+ > large_file_wo_1st_file)
share|improve this answer

This is just theorizing, but...

A custom filesystem (implemented using FUSE or a similar mechanism) could expose a directory whose contents are exactly the same as an already existing directory somewhere else, but with files truncated as you wish. The filesystem would translate all the file offsets. Then you wouldn't have to do a time-consuming rewriting of a file.

But given that this idea is very non-trivial, unless you've got tens of terabytes of such files, implementing such filesystem would be too expensive/time-consuming to be practical.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.