From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mailserv2.iuinc.com (IDENT:qmailr@mailserv2.iuinc.com [206.245.164.55]) by puffin.external.hp.com (8.9.3/8.9.3) with SMTP id KAA13147 for ; Mon, 22 Jan 2001 10:27:46 -0700 Received: from bame.riverrock.org (really [127.0.0.1]) by bame.riverrock.org via in.smtpd with esmtp id (Debian Smail3.2.0.102) for ; Mon, 22 Jan 2001 10:28:22 -0700 (MST) Message-Id: From: bame@riverrock.org Reply-To: bame@riverrock.org Errors-To: bame@riverrock.org To: parisc-linux@thepuffingroup.com Date: Mon, 22 Jan 2001 10:28:18 -0700 Sender: bame@bame.riverrock.org Subject: [parisc-linux] First benchmark results List-ID: Executive summary: HP-UX is shown out-of-the-box, with no parameter adjustments which could help it show better in these benchmarks. HP-UX, a mature much-tuned kernel and Linux, a not fully functional and never tuned on HP hardware kernel, are competitive in a great many areas. HP-UX is faster in some areas, and Linux is faster than HP-UX in fewer areas. This portends a bright future for Linux on parisc! HPUX, is MUCH faster than Linux at fork/exec and pipe and unix/tcp socket bandwidths. Note: Linux fork/exec got a lot faster with the recent cache-flushing changes. Linux wins big on stat, open/close, file create/delete (sometimes > 100:1). Linux beats HPUX in the 2-process context-switch trials, but not in the 8 or 16-process trials. The 100MHz gecko is roughly 8x slower than the 400MHz C3000 on compute-bound benchmarks, re-emphasizing that MHz is not an entirely trustworthy performance predictor. Caveats: HPUX was not optimized for this benchmark. Changing the fs_async flag might make a huge improvement in the file-system-related performance. The "TCP conn" data should not be trusted at all. LOTS OF DATA IS OBSCURED and possibly MISREPRESENTED by this summary. The hoards of data upon which this summary is based is available at http://puffin.external.hp.com/~bame/lmbench-2beta1/results. c3k-hpux is an active in-use HPUX box with variable loads runing all the standard daemons therefore some benchmark results are slow. Lmbench 2beta1 by its name is clearly "beta" software (though it seems to work fine). More info about lmbench: http://www.bitmover.com/lmbench Some test programs on Linux still require manual intervention. Key: c3k-j09-3 C3000 Pre-cache-flushing (Jan 9) kernel 32-bit Linux c3k-j19-3 C3000 Post-cache-flushing (Jan 19) kernel 32-bit Linux 712-100 100MHz 9000/712 (gecko) Jan 19 Linux c3k-hpux C3000 HPUX 10.20 (*** an active in-use system ***) L M B E N C H 2 . 0 S U M M A R Y ------------------------------------ (Alpha software, do not distribute) Basic system parameters ---------------------------------------------------- Host OS Description Mhz --------- ------------- ----------------------- ---- c3k-j19-6 Linux 2.4.0-t parisc64-linux-gnu 400 712-100 Linux 2.4.0-t parisc-linux-gnu 100 c3k-j09-3 Linux 2.4.0-t parisc-linux-gnu 400 c3k-j19-3 Linux 2.4.0-t parisc-linux-gnu 400 c3k-hpux HP-UX B.10.20 hppa2.0-hp-hpux10.20 400 c3k-hpux HP-UX B.10.20 hppa2.0-hp-hpux10.20 400 c3k-hpux HP-UX B.10.20 hppa2.0-hp-hpux10.20 400 c3k-hpux HP-UX B.10.20 hppa2.0-hp-hpux10.20 400 Processor, Processes - times in microseconds - smaller is better ---------------------------------------------------------------- Host OS Mhz null null open selct sig sig fork exec sh call I/O stat clos TCP inst hndl proc proc proc --------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ---- c3k-j19-6 Linux 2.4.0-t 400 0.85 1.48 8.74 10 90 2.44 18 2513 13K 61K 712-100 Linux 2.4.0-t 100 3.07 5.10 44 47 249 12 72 5651 41K 229K c3k-j09-3 Linux 2.4.0-t 400 0.77 1.34 8.35 8.96 63 2.31 14 11K 115K 433K c3k-j19-3 Linux 2.4.0-t 400 0.76 1.32 8.12 8.88 61 2.26 14 1674 11K 53K c3k-hpux HP-UX B.10.20 400 0.34 21 25 55 3.42 12 699 1873 9950 c3k-hpux HP-UX B.10.20 400 0.33 21 25 54 3.42 12 699 1876 9946 c3k-hpux HP-UX B.10.20 400 0.33 21 25 55 3.42 12 698 1866 6678 c3k-hpux HP-UX B.10.20 400 0.33 21 25 55 3.42 12 700 1871 6715 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 --------- ------------- ----- ------ ------ ------ ------ ------- ------- c3k-j19-6 Linux 2.4.0-t 1.870 1.8800 1.5800 25 84 29 97 712-100 Linux 2.4.0-t 23 156 617 175 625 179 627 c3k-j09-3 Linux 2.4.0-t 1.420 1.4400 1.2400 22 83 27 95 c3k-j19-3 Linux 2.4.0-t 1.360 1.3000 1.1500 22 82 27 95 c3k-hpux HP-UX B.10.20 3.760 3.6500 273 119 288 92 184 c3k-hpux HP-UX B.10.20 3.760 3.6800 3.7500 3.7000 32 6.31000 60 c3k-hpux HP-UX B.10.20 3.760 3.6300 3.5100 3.5600 13 6.84000 64 c3k-hpux HP-UX B.10.20 4.320 3.7700 3.6300 3.7000 10 5.92000 51 *Local* Communication latencies in microseconds - smaller is better ------------------------------------------------------------------- Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP ctxsw UNIX UDP TCP conn --------- ------------- ----- ----- ---- ----- ----- ----- ----- ---- c3k-j19-6 Linux 2.4.0-t 1.870 10 19 29 144 38 218 208 712-100 Linux 2.4.0-t 23 43 84 153 260 1032 c3k-j09-3 Linux 2.4.0-t 1.420 21 49 41 36 c3k-j19-3 Linux 2.4.0-t 1.360 9.501 19 28 36 8.2M c3k-hpux HP-UX B.10.20 3.760 20 22 42 93 39 103 2.0M c3k-hpux HP-UX B.10.20 3.760 19 22 41 93 39 101 2.0M c3k-hpux HP-UX B.10.20 3.760 19 21 41 93 39 101 2.0M c3k-hpux HP-UX B.10.20 4.320 19 21 41 93 39 102 2.0M File & VM system latencies in microseconds - smaller is better -------------------------------------------------------------- Host OS 0K File 10K File Mmap Prot Page Create Delete Create Delete Latency Fault Fault --------- ------------- ------ ------ ------ ------ ------- ----- ----- c3k-j19-6 Linux 2.4.0-t 11 2.9577 35 7.4666 1944 20 712-100 Linux 2.4.0-t 78 11 184 20 3546 42 c3k-j09-3 Linux 2.4.0-t 9.0001 1.0000 170 3.0000 174.2860 937 c3k-j19-3 Linux 2.4.0-t 9.0001 1.0000 28 3.0000 1423 c3k-hpux HP-UX B.10.20 1123 598 1694 598 191.8620 6.480 5504 c3k-hpux HP-UX B.10.20 1123 598 1754 598 181.1960 6.880 5738 c3k-hpux HP-UX B.10.20 1136 598 1724 598 184.6730 4.232 5418 c3k-hpux HP-UX B.10.20 1123 598 1724 598 185.1890 4.300 5475 *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 --------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- ----- c3k-j19-6 Linux 2.4.0-t 47 49 39 73 514 196 201 505 432 712-100 Linux 2.4.0-t 7.35 6.61 6.13 12 69 34 26 69 52 c3k-j09-3 Linux 2.4.0-t 47 49 41 76 509 195 199 509 430 c3k-j19-3 Linux 2.4.0-t 48 49 42 76 513 193 200 504 430 c3k-hpux HP-UX B.10.20 406 528 298 490 391 164 163 503 452 c3k-hpux HP-UX B.10.20 402 537 315 506 509 207 208 504 467 c3k-hpux HP-UX B.10.20 407 537 312 508 509 207 206 494 467 c3k-hpux HP-UX B.10.20 406 538 311 505 509 205 209 499 467 Memory latencies in nanoseconds - smaller is better (WARNING - may not be correct, check graphs) --------------------------------------------------- Host OS Mhz L1 $ L2 $ Main mem Guesses --------- ------------- ---- ----- ------ -------- ------- c3k-j19-6 Linux 2.4.0-t 400 7.503 7.5030 184 No L1 cache? 712-100 Linux 2.4.0-t 100 22 22 223 No L1 cache? c3k-j09-3 Linux 2.4.0-t 400