Linux Performance Increases

[jwt@example.com (Jim Tittsler)]

I tend to read the linux-kernel mailing list in fits and starts
because of its volume.  I did happen to notice this rather interesting
table that suggests the new kernels do have a slight performance edge.

------------------------------

From: pommnitz@example.com (Joerg Pommnitz)
Date: Tue, 19 Dec 1995 12:14:45 +0100 (CET)
Subject: Re: Linux 1.3.47

> I'd be interested to hear what the results of the TLB invalidation
> changes are for benchmarks (byte benchmark, anyone?).  Also, I didn't
> have time to do the alpha versions of the invalidate code, but it should
> actually be trivial (and might actually have more of an impact on the
> alpha than on the x86 - although the original reason for the
> finer-grained invalidates were for more braindamaged chips like sparc
> and powerpc that _really_ need it). 
> 
> 		Linus

This is a small table I compiled to compare the performance development
of the Linux kernel over the 1.3 series. Especially the xbench tests have
a few gaps, but I think the general trend towards an improved performance
is quite obvious. 
My comments are mostly guesses to explain the changes in the numbers. If 
somebody has better explanations, feel free to correct me.

Hardware:
 CPU   : i486/66
 RAM   : 16M
 SCSI  : NCR 53c810
 HD    : QUANTUM MAVERICK 540S
 video : ATI Mach64
 
Software:
 bm     : BYTE Unix Benchmarks Version 3.6 -- 5/15/91 (as found on tsx-11)
          statically linked ELF binaries compiled with gcc 2.6.3
 xbench : patched version
          dynamically linked ELF binary compiled with gcc 2.7.0 and
	  gcc 2.7.1 snapshots
 X      : XFree86 3.1.2 self compiled with gcc 2.7.0	  
	  
  +---------------+--------+-----+----------------------------------------+
  | OS  Version   | xbench | bm  | Comment                                |
  +---------------+--------+-----+----------------------------------------+
  | Linux-1.2.09  | 149882 | 4.1 | old scheduler, old UNIX domain sockets |
  | Linux-1.2.09k | 151112 | 4.3 | orig. kswap patches applied            |
  | Linux-1.3.08  |        | 3.9 |                                        |
  | Linux-1.3.11+ |        | 4.3 | improved scheduler                     |
  | Linux-1.3.14  |        | 4.7 | improved fork(2)/clone(2) ???          |
  | Linux-1.3.17  |        | 5.0 |                                        |
  | Linux-1.3.18  |        | 5.0 |                                        |
  | Linux-1.3.29  | 121501 | 5.0 | new UNIX domain sockets                |
  | Linux-1.3.30  |        | 4.9 |                                        |
  | Linux-1.3.34  |        | 4.9 |                                        |
  | Linux-1.3.35  | 145499 |     |                                        |
  | Linux-1.3.36  |        | 4.9 |                                        |
  | Linux-1.3.38  |        | 5.7 | kswap patches in ???                   |
  | Linux-1.3.39  | 151284 | 5.5 |                                        |
  | Linux-1.3.41  | 154599 | 5.6 | libc-5.2.13 with native readv/writev   |
  | Linux-1.3.42  | 150920 | 5.7 |                                        |
  | Linux-1.3.43  | 153943 | 5.8 |                                        |  
  | Linux-1.3.45  | 154134 | 5.7 |                                        |
  | Linux-1.3.46  | 152350 | 5.8 |                                        |
  | Linux-1.3.48  | 152328 | 5.6 |                                        |
  +---------------+--------+-----+----------------------------------------+

- -- 
Best regards
    Joerg
- --------------------------------------------------------------------------------
Joerg Pommnitz, ARTe Team, IBM Scientific Center Heidelberg, Germany
   Mail1: pommnitz@example.com
   Mail2: jpo@example.com
   Tel. : Germany (06221) 59 3609

------------------------------