Re: [2.4.17/18pre] VM and swap - it's really unusable

rwhron@earthlink.net
Mon, 21 Jan 2002 20:44:40 -0500


2.4.18pre2a2 aa vm and low-latency included
2.4.18-pre3
2.4.18-pre3ll low-latency patch
2.4.18-pre3pelb preempt and lockbreak patches

It appears 2.4.18pre2aa2 is the best overall, and the
low-latency patch does better than preempt in most cases.

7 lmbench runs run for each kernel. Below are averages the averages.
Remember that "smaller is better" in the latency tests and "bigger is
better" in the bandwidth tests.

L M B E N C H 2 . 0 S U M M A R Y
------------------------------------

Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
OS Mhz null null open selct sig sig fork exec sh
call I/O stat clos TCP inst hndl proc proc proc
------------------------ ---- ---- ---- ---- ---- ----- ---- ---- ------ ------ ------
Linux 2.4.18pre2aa2 0.43 na 3.18 4.79 61.8 1.44 3.17 711 2839 11156
Linux 2.4.18-pre3 0.43 na 3.34 5.83 38.6 1.51 3.28 908 3656 13139
Linux 2.4.18pre3ll 0.43 na 3.39 4.98 41.5 1.49 3.19 884 3524 12829
Linux 2.4.18pre3pelb 0.43 na 4.46 5.98 41.5 1.57 3.94 919 3620 13155

Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
------------------------ ----- ------ ------ ------ ------ ------ ------
Linux 2.4.18pre2aa2 1.24 21.64 180.3 56.1 203.0 58.0 223.6
Linux 2.4.18-pre3 1.06 25.13 192.5 57.5 206.4 59.4 226.3
Linux 2.4.18pre3ll 1.74 23.71 192.3 57.6 207.7 59.1 225.5
Linux 2.4.18pre3pelb 2.55 25.36 187.3 61.8 207.2 62.2 227.0

*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
----------------------- ----- ----- ----- ----- ----- ----- ----- -----
Linux 2.4.18pre2aa2 2.04 10.65 20.67 na na 64.99 na 269.7
Linux 2.4.18-pre3 2.04 9.846 16.45 na na 73.41 na 286.7
Linux 2.4.18pre3ll 2.04 9.763 18.7 na na 78 na 293.5
Linux 2.4.18pre3pelb 2.04 16.38 17.95 na na 67.98 na 312.5

File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
OS 0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
------------------------ ------ ------ ------ ------ ------- ----- -----
Linux 2.4.18pre2aa2 149.5 166.9 610 257.1 2683 1.194 5.6
Linux 2.4.18-pre3 148.6 172.7 637.3 257.7 5308 1.084 6.7
Linux 2.4.18pre3ll 153.1 177.9 636.9 274.7 3447 1.003 6.9
Linux 2.4.18pre3pelb 154.6 182.6 649.0 270.3 2753 0.618 7.0

*Local* Communication bandwidths in MB/s - bigger is better
-----------------------------------------------------------
OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read write
------------------------ ----- ----- ----- ------ ------ ------ ------ ----- -----
Linux 2.4.18pre2aa2 69.9 50.0 45.4 62.5 237.6 59.2 60.4 237.5 85.0
Linux 2.4.18-pre3 68.7 49.5 44.2 60.7 237.5 59.2 60.4 237.5 85.0
Linux 2.4.18pre3ll 65.0 45.0 50.4 59.8 237.3 59.2 60.4 237.4 85.1
Linux 2.4.18pre3pelb 68.5 41.9 44.2 61.7 237.4 59.2 60.4 237.3 85.1

Memory latencies in nanoseconds - smaller is better
---------------------------------------------------
OS Mhz L1 $ L2 $ Main mem
------------------------- ---- ----- ------ --------
Linux 2.4.18pre2aa2 501 4.2 193 262
Linux 2.4.18-pre3 501 4.2 191.7 261.9
Linux 2.4.18pre3ll 501 4.2 194.8 261.9
Linux 2.4.18pre3pelb 501 4.2 191.8 262.3

dbench
------
2.4.18pre2aa2 Throughput 22.2399 MB/sec 64 procs
2.4.18pre3ll Throughput 12.563 MB/sec 64 procs
2.4.18-pre3 Throughput 12.0448 MB/sec 64 procs
2.4.18pre3pelb Throughput 11.9764 MB/sec 64 procs

2.4.18pre2aa2 Throughput 14.902 MB/sec 128 procs
2.4.18pre3ll Throughput 7.93162 MB/sec 128 procs
2.4.18pre3pelb Throughput 7.76489 MB/sec 128 procs
2.4.18-pre3 Throughput 7.72214 MB/sec 128 procs

2.4.18pre2aa2 Throughput 8.75662 MB/sec 192 procs
2.4.18pre3ll Throughput 7.83356 MB/sec 192 procs
2.4.18-pre3 Throughput 7.32843 MB/sec 192 procs
2.4.18pre3pelb Throughput 6.95136 MB/sec 192 procs

BYTE UNIX Benchmarks (Version 4.1.0)
2.4.18pre3ll 2.4.18pre3pelb 2.4.18pre3 2.4.18pre2aa2
Execl Throughput 329.7 305.0 311.1 391.4
File Read 1024 bufsize 2000 maxb 164316.0 152915.0 171358.0 169157.0
File Write 1024 bufsize 2000 max 41825.0 38918.0 43153.0 44459.0
File Copy 1024 bufsize 2000 maxb 29279.0 28279.0 31533.0 31036.0
File Read 256 bufsize 500 maxblo 116575.0 102763.0 118511.0 116442.0
File Write 256 bufsize 500 maxbl 16163.0 15563.0 18021.0 17119.0
File Copy 256 bufsize 500 maxblo 12812.0 11072.0 13496.0 13932.0
File Read 4096 bufsize 8000 maxb 152353.0 150358.0 158244.0 150669.0
File Write 4096 bufsize 8000 max 62735.0 63425.0 66933.0 64480.0
File Copy 4096 bufsize 8000 maxb 40332.0 41387.0 43400.0 41999.0
Pipe Throughput 319240.9 301399.6 384233.8 387119.6
Pipe-based Context Switching 95662.9 46867.6 103815.3 112401.9
Process Creation 1182.9 1088.6 1182.1 1378.7
System Call Overhead 303256.1 224553.0 335236.6 342022.3
Shell Scripts (1 concurrent) 634.1 624.7 615.0 692.7
Shell Scripts (8 concurrent) 86.0 84.3 84.0 97.7
Shell Scripts (16 concurrent) 43.0 42.0 42.0 49.0
Shell Scripts (32 concurrent) 21.0 21.0 21.0 24.0
Shell Scripts (64 concurrent) 10.0 10.0 10.0 12.0
Shell Scripts (96 concurrent) 6.6 6.3 6.3 8.3
C Compiler Throughput 229.4 222.0 224.7 265.9
Dc: sqrt(2) to 99 decimal places 15045.9 13881.1 14874.8 20420.1
Recursion Test--Tower of Hanoi 14154.3 14152.5 14154.8 14151.6

More results at http://home.earthlink.net/~rwhron/kernel/k6-2-475.html

-- 
Randy Hron

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/