Skip to contentSkip to conversation

lmbench 2.4.20-8(RH9), 2.4.20, 2.6.0

DISCUSSION Huo Zhigang(霍志刚) 2 participants 2 messages Dec 20, 2003

lore
Share & tools
lore
#0 lore raw

Howdy,

Is the network performance of 2.6.0 so bad or my test totally wrong?

Could anyone tell me how to dig into performance
optimization of linux?

=================================================================

L M B E N C H 3 . 0 S U M M A R Y
------------------------------------
(Alpha software, do not distribute)

Basic system parameters
------------------------------------------------------------------------------

Host                 OS Description              Mhz  tlb  cache  mem   scal
                                                     pages line   par   load
                                                           bytes  
--------- ------------- ----------------------- ---- ----- ----- ------ ----
lucent.rd Linux 2.4.20-       i686-pc-linux-gnu 1595    14   128 1.6100    1
lucent.rd  Linux 2.4.20       i686-pc-linux-gnu 1592    66   128 1.6700    1
lucent.rd   Linux 2.6.0       i686-pc-linux-gnu 1572    63   128 1.6600    1

Processor, Processes - times in microseconds - smaller is better
------------------------------------------------------------------------------
Host OS Mhz null null open slct sig sig fork exec sh
call I/O stat clos TCP inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- lucent.rd Linux 2.4.20- 1595 0.99 1.09 3.25 4.54 8.41 1.57 4.25 194. 671. 3273 lucent.rd Linux 2.4.20 1592 1.01 1.11 14.5 15.9 10.4 1.58 5.42 185. 752. 3739
lucent.rd Linux 2.6.0 1572 0.29 0.75 17.9 25.9 10.8 1.07 16.1 336. 1223 5075

                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~??

Basic integer operations - times in nanoseconds - smaller is better
-------------------------------------------------------------------
Host OS intgr intgr intgr intgr intgr
bit add mul div mod
--------- ------------- ------ ------ ------ ------ ------

lucent.rd Linux 2.4.20- 0.3200 0.3200 8.8500   36.8   41.3
lucent.rd  Linux 2.4.20 0.3200 0.3100 8.8400   36.6   40.5
lucent.rd   Linux 2.6.0 0.3200 0.3200 9.0300   37.7   42.4

Basic float operations - times in nanoseconds - smaller is better
-----------------------------------------------------------------

Host                 OS  float  float  float  float
                         add    mul    div    bogo
--------- ------------- ------ ------ ------ ------ 
lucent.rd Linux 2.4.20- 3.1400 4.3900   27.5   27.1
lucent.rd  Linux 2.4.20 3.3800 4.3900   27.1   27.1
lucent.rd   Linux 2.6.0 3.1900 4.4800   27.4   27.4

Basic double operations - times in nanoseconds - smaller is better
------------------------------------------------------------------
Host OS double double double double
add mul div bogo
--------- ------------- ------ ------ ------ ------

lucent.rd Linux 2.4.20- 3.1500 4.4300   27.1   27.1
lucent.rd  Linux 2.4.20 3.1300 4.3700   29.3   27.1
lucent.rd   Linux 2.6.0 3.1900 4.6600   27.4   27.5

Context switching - times in microseconds - smaller is better
-------------------------------------------------------------------------
Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
--------- ------------- ------ ------ ------ ------ ------ ------- -------

lucent.rd Linux 2.4.20- 1.6600 2.0400 2.0900 3.1900   24.4 6.05000    34.6
lucent.rd  Linux 2.4.20 1.7300 0.8800                 14.7 0.10000    24.9
lucent.rd   Linux 2.6.0 1.5600 1.6500 1.9200 4.6500   30.5    10.0    40.4

*Local* Communication latencies in microseconds - smaller is better
---------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----

lucent.rd Linux 2.4.20- 1.660 8.525 13.3              20.7        60.
lucent.rd  Linux 2.4.20 1.730 7.541 13.5              63.8        73.
lucent.rd   Linux 2.6.0 1.560 9.622 73.4             132.6       451.
                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~??
                        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~??

File & VM system latencies in microseconds - smaller is better
-------------------------------------------------------------------------------

Host                 OS   0K File      10K File     Mmap    Prot   Page   100fd
                        Create Delete Create Delete Latency Fault  Fault  selct
--------- ------------- ------ ------ ------ ------ ------- ----- ------- -----
lucent.rd Linux 2.4.20-   31.4 4.1380   66.9 8.2450  1124.0 1.327 2.95680 5.316
lucent.rd  Linux 2.4.20   43.1   15.6   76.2   20.7  1563.0 2.989 3.55410 7.224
lucent.rd   Linux 2.6.0   67.6   28.5  124.1   58.7  1198.0 1.094 3.41040  10.6

*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------------------------
Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read write
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----

lucent.rd Linux 2.4.20- 799. 1348 378.  988.8 1522.1  406.9  418.0 1520 602.8
lucent.rd  Linux 2.4.20 754. 199.       758.3  900.5  304.3  341.9 971. 438.1
lucent.rd   Linux 2.6.0 671. 316. 92.9 1044.1 1472.6  447.5  466.0 1472 647.7

Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
------------------------------------------------------------------

Host                 OS   Mhz   L1 $   L2 $    Main mem    Guesses
--------- -------------   ---   ----   ----    --------    -------
lucent.rd Linux 2.4.20-  1595 1.2980   11.6       135.9
lucent.rd  Linux 2.4.20  1592 1.3650   11.7       136.7
lucent.rd   Linux 2.6.0  1572 1.3000   11.8       144.2

--
"Mom-proof" linux is my dream.

#1 lore raw
Quoted 8 lines · Huo Zhigang(???)

Howdy,

Is the network performance of 2.6.0 so bad or my test totally wrong?

Could anyone tell me how to dig into performance
optimization of linux?

This has come up a few times, but nobody is too worried about it. I have profiles showing that more than double the amount of time is spent in copying data for the same amount of work.

Its worth noting though, that it is "networking" over localhost. I expect the numbers for remote networking are better, but don't have a good network setup to test it.

Do you have any real applications (not benchmarks) that have performance problems?