[Bug libc/11607] strstr function very slow.

[Bug libc/11607] strstr function very slow.

[maxim.kuvyrkov at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=11607

Maxim Kuvyrkov <maxim.kuvyrkov at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|                            |maxim.kuvyrkov at gmail dot
                   |                            |com

--- Comment #5 from Maxim Kuvyrkov <maxim.kuvyrkov at gmail dot com> 2012-06-28 22:22:51 UTC ---
We expect this to be fixed in 2.17.

Maxim's patches [1] fix performance regression for small needles -- which is
what the testcase for this bug exercises.  Ondrej's patches [2] appear to
further improve performance for x86.

[1] http://sourceware.org/ml/libc-alpha/2012-05/msg01228.html
[2] http://sourceware.org/ml/libc-alpha/2012-06/msg00027.html

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[maxim.kuvyrkov at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=11607

Maxim Kuvyrkov <maxim.kuvyrkov at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|drepper.fsp at gmail dot    |maxim.kuvyrkov at gmail dot
                   |com                         |com

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[eblake at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=11607

--- Comment #6 from Eric Blake <eblake at redhat dot com> 2012-06-28 22:40:23 UTC ---
(In reply to comment #5)
> We expect this to be fixed in 2.17.
> 
> Maxim's patches [1] fix performance regression for small needles -- which is
> what the testcase for this bug exercises.  Ondrej's patches [2] appear to
> further improve performance for x86.
> 
> [1] http://sourceware.org/ml/libc-alpha/2012-05/msg01228.html
> [2] http://sourceware.org/ml/libc-alpha/2012-06/msg00027.html

Note that [2] was just a summary of Ondrej's benchmarking; searching for an
actual patch of his turns up this link:

http://sourceware.org/ml/libc-help/2011-11/msg00011.html

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[carlos_odonell at mentor dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=11607

Carlos O'Donell <carlos_odonell at mentor dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |carlos_odonell at mentor
                   |                            |dot com

--- Comment #7 from Carlos O'Donell <carlos_odonell at mentor dot com> 2012-09-04 15:14:02 UTC ---
Maxim,

I thought this had already been checked in and fixed?

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[maxim.kuvyrkov at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=11607

Maxim Kuvyrkov <maxim.kuvyrkov at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #8 from Maxim Kuvyrkov <maxim.kuvyrkov at gmail dot com> 2012-09-04 20:02:10 UTC ---
Hm, I thought I marked this FIXED.  Apparently, I didn't.  Thanks Carlos.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=11607

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |security-

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/11607] strstr function very slow.

[eblake at redhat dot com]

------- Additional Comments From eblake at redhat dot com  2010-05-17 12:13 -------
In glibc 2.7, the strstr() implementation has worst-case quadratic complexity of
O(m*n) (m the length of the haystack, n the length of the string); and gnulib
introduced a test case that demonstrated this quadratic complexity could be used
as a denial of service attack on any program that searches for a large needle
within a large string.  With glibc 2.10, the strstr() has a guaranteed linear
complexity of O(m+n), but that reduced algorithmic complexity comes with a
larger startup cost of factoring the needle, where the shorter the needle is,
the larger proportion of the overall strstr() is spent on that factoring. 
Furthermore, the glibc 2.7 implementation has had years of tweaking to be
cache-line friendly and tuned for various architectures, while my 2.10
implementation is relatively new, and probably still has some room for
improvements (at any rate, I know that when I wrote it, I was catering to large
needles and large haystacks, and was focused on the algorithmic improvement,
rather than the cache-line effects of small needles).

With appropriate benchmarking, it may be worth altering the implementation to be
a hybrid, where short-enough needles are handled with the older brute force
implementation - even if it is quadratic, less startup cost might be better for
short needles even if it results in more comparisons against the haystack for
some of those needles.

Furthermore, strstr() has a lousy interface when compared with regex searching.
 That is, if you repeatedly search for the same needle, the use of re_compile()
is spent only once, while you can reuse that compiled needle across multiple
re_search() calls.  But with strstr(), the cost of factoring the needle is spent
on every call.  Perhaps we should consider exposing some new interfaces that
separate the strstr() factorization startup code from the body that uses the
factored needle for the actual search.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|ebb9 at byu dot net         |eblake at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=11607

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/11607] strstr function very slow.

[jakub at redhat dot com]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ebb9 at byu dot net


http://sourceware.org/bugzilla/show_bug.cgi?id=11607

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/11607] strstr function very slow.

[xoswald at debian dot org]

------- Additional Comments From xoswald at debian dot org  2010-05-17 09:33 -------
in test-strstr.c, just try to match a string a the end of the file test-strstr.txt
I tested with a needle > 16 and < 16, it doesn't change anything.

-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=11607

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/11607] strstr function very slow.

[xoswald at debian dot org]

------- Additional Comments From xoswald at debian dot org  2010-05-17 09:28 -------
Created an attachment (id=4799)
 --> (http://sourceware.org/bugzilla/attachment.cgi?id=4799&action=view)
Data file


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=11607

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libc/11607] strstr function very slow.

[xoswald at debian dot org]

------- Additional Comments From xoswald at debian dot org  2010-05-17 09:28 -------
Created an attachment (id=4798)
 --> (http://sourceware.org/bugzilla/attachment.cgi?id=4798&action=view)
Source code


-- 


http://sourceware.org/bugzilla/show_bug.cgi?id=11607

------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.