Sort Command – Scalability of ‘sort -u’ for Gigantic Files

Drav's awk solution is good, but that means running one sort command per paragraph. To avoid that, you could do:

< file awk -v n=0 '!NF{n++};{print n,$0}' | sort -k1n -k2 | cut -d' ' -f2-

Or you could do the whole thing in perl:

perl -ne 'if (/\S/){push@l,$_}else{print sort@l if@l;@l=();print}
          END{print sort @l if @l}' < file

Note that above, separators are blank lines (for the awk one, lines with only space or tab characters, for the perl one, any horizontal or vertical spacing character) instead of empty lines. If you do want empty lines, you can replace !NF with !length or $0=="", and /\S/ with /./.

Not a POSIX solution, but GNU sort offers the -g option which supports a wider range of number specifications including infinity. From http://www.gnu.org/software/coreutils/manual/html_node/sort-invocation.html -

‘-g’
‘--general-numeric-sort’
‘--sort=general-numeric’

Sort numerically, converting a prefix of each line to a long double-precision
floating point number. See Floating point. Do not report overflow, underflow, or
conversion errors. Use the following collating sequence:

    Lines that do not start with numbers (all considered to be equal).
    NaNs (“Not a Number” values, in IEEE floating point arithmetic) in a
      consistent but machine-dependent order.
    Minus infinity.
    Finite numbers in ascending numeric order (with -0 and +0 equal).
    Plus infinity. 

Use this option only if there is no alternative; it is much slower than
--numeric-sort (-n) and it can lose information when converting to floating
point.

From my own tests it seems that any line beginning with Inf (any combination of upper/lower case) will appear after any numbers.

Failing that I don't think there are any character sequences that are reliably sorted after numbers using sort -n. GNU sort seems to treat all other sequences first as zero, placing the after negative numbers but before positive ones. What you could do, if it is timestamps that are being sorted, is to use the maximum value for a 64 bit timestamp plus one:

 9,223,372,036,854,775,808

This is a few more digits than you started out with!