Jean-Marc Zucconi benchmarked FreeBSD 4.9-R, 5.1-R, 5.2-R and DragonFly on a multiprocessor system, and posted the large quantity of data this generated. It’s mashed into a table here:
(updated with better results for 5.2; ACPI was generating so many interrupts it was slowing his system.)
(see end of entry for his notes)
| FreeBSD-4.9 | FreeBSD-5.1 | FreeBSD-5.2 | DragonFly | |
| Dhrystone 2 without register variables | 660050.1 lps | 650148.4 lps | 661362.9 lps | 667712.8 lps |
| Dhrystone 2 using register variables | 660062.3 lps | 650258.6 lps | 661430.8 lps | 665312.0 lps |
| Arithmetic Test (type = arithoh) | 1518178.8 lps | 1527450.3 lps | 1055234.6 lps | 1513511.1 lps |
| Arithmetic Test (type = register) | 89009.9 lps | 91551.8 lps | 91672.9 lps | 87657.5 lps |
| Arithmetic Test (type = short) | 81709.0 lps | 91413.9 lps | 88086.6 lps | 81487.4 lps |
| Arithmetic Test (type = int) | 89030.8 lps | 91550.6 lps | 91662.9 lps | 88331.6 lps |
| Arithmetic Test (type = long) | 88999.0 lps | 91544.6 lps | 91682.2 lps | 89063.4 lps |
| Arithmetic Test (type = float) | 88148.0 lps | 105020.8 lps | 105154.2 lps | 87531.7 lps |
| Arithmetic Test (type = double) | 88148.2 lps | 104999.4 lps | 105242.2 lps | 87891.6 lps |
| System Call Overhead Test | 107461.9 lps | 82630.7 lps | 82300.9 lps | 89537.9 lps |
| Pipe Throughput Test | 133588.2 lps | 86297.9 lps | 78012.8 lps | 106999.4 lps |
| Pipe-based Context Switching Test | 38436.5 lps | 23821.9 lps | 23268.0 lps | 18203.4 lps |
| Process Creation Test | 2736.3 lps | 1574.6 lps | 1517.7 lps | 1569.6 lps |
| Execl Throughput Test | 1772.9 lps | 1521.2 lps | 1371.7 lps | 1497.6 lps |
| File Read (10 seconds) | 179477.0 KBps | 167848.0 KBps | 181930.0 KBps | 179972.0 KBps | File Write (10 seconds) | 18633.0 KBps | 12800.0 KBps | 7800.0 KBps | 18433.0 KBps |
| File Copy (10 seconds) | 16298.0 KBps | 11944.0 KBps | 11905.0 KBps | 16271.0 KBps |
| File Read (30 seconds) | 179598.0 KBps | 169150.0 KBps | 182251.0 KBps | 177467.0 KBps | File Write (30 seconds) | 18399.0 KBps | 12589.0 KBps | 7688.0 KBps | 18277.0 KBps |
| File Copy (30 seconds) | 15297.0 KBps | 11468.0 KBps | 4375.0 KBps | 15308.0 KBps |
| C Compiler Test | 428.1 lpm | 258.4 lpm | 241.2 lpm | 407.1 lpm |
| Shell scripts (1 concurrent) | 756.0 lpm | 567.9 lpm | 572.2 lpm | 628.6 lpm |
| Shell scripts (2 concurrent) | 507.3 lpm | 346.6 lpm | 324.9 lpm | 385.6 lpm |
| Shell scripts (4 concurrent) | 271.0 lpm | 183.0 lpm | 163.9 lpm | 208.3 lpm |
| Shell scripts (8 concurrent) | 137.7 lpm | 93.0 lpm | 82.0 lpm | 107.0 lpm |
| Dc: sqrt(2) to 99 decimal places | 83900.0 lpm | 52378.5 lpm | 54949.6 lpm | 25753.6 lpm |
| Recursion Test–Tower of Hanoi | 7920.6 lps | 8162.1 lps | 8382.6 lps | 7825.9 lps |
| INDEX VALUES | ——— FreeBSD-4.9 ——– | ——— FreeBSD-5.1 ——– | ——— FreeBSD-5.2 ——– | ———- DragonFly ——— | ||||||||
| TEST | BASELINE | RESULT | INDEX | BASELINE | RESULT | INDEX | BASELINE | RESULT | INDEX | BASELINE | RESULT | INDEX |
| Arithmetic Test (type = double) | 2541.7 | 88148.2 | 34.7 | 2541.7 | 104999.4 | 41.3 | 2541.7 | 105242.2 | 41.4 | 2541.7 | 87891.6 | 34.6 |
| Dhrystone 2 without register variables | 22366.3 | 660050.1 | 29.5 | 22366.3 | 650148.4 | 29.1 | 22366.3 | 536903.0 | 24.0 | 22366.3 | 667712.8 | 29.9 |
| Execl Throughput Test | 16.5 | 1772.9 | 107.4 | 16.5 | 1521.2 | 92.2 | 16.5 | 1371.7 | 83.1 | 16.5 | 1497.6 | 90.8 |
| File Copy (30 seconds) | 179.0 | 15297.0 | 85.5 | 179.0 | 11468.0 | 64.1 | 179.0 | 11811.0 | 66.0 | 179.0 | 15308.0 | 85.5 |
| Pipe-based Context Switching Test | 1318.5 | 38436.5 | 29.2 | 1318.5 | 23821.9 | 18.1 | 1318.5 | 23268.0 | 17.6 | 1318.5 | 18203.4 | 13.8 |
| Shell scripts (8 concurrent) | 4.0 | 137.7 | 34.4 | 4.0 | 93.0 | 23.2 | 4.0 | 82.0 | 20.5 | 4.0 | 107.0 | 26.8 |
| ========= | ========= | ========= | ========= | |||||||||
| SUM of 6 items | 320.7 | 268.0 | 258.2 | 281.3 | ||||||||
| AVERAGE | 53.4 | 44.7 | 43.0 | 46.9 | ||||||||
| “Finally to test if there is a problem with the compiler I used the lorenz benchmark (it simply does raw math computations, and does not use much memory)”: | |||
| Lorenz on 5.1 | |||
| 4.9binary: | 15.77 real | 15.73 user | 0.01 sys |
| 5.1binary: | 16.52 real | 16.51 user | 0.00 sys |
| 5.2binary: | 16.36 real | 16.32 user | 0.02 sys |
| Lorenz on 5.2 | |||
| 4.9binary: | 19.35 real | 18.12 user | 1.14 sys |
| 5.1binary: | 27.65 real | 26.05 user | 1.48 sys |
| 5.2binary: | 29.61 real | 27.60 user | 1.87 sys |
| “The 5.1 results indicate that there is no problem with the compiler generated code, but I don’t know how to interpret the 5.2 results: I would expect also a bad behavior with the 4.9 binary (all are static) Then I booted a uniprocessor kernel and reran the latest test:” | |||
| Lorenz on 5.2 UP | |||
| 4.9binary: | 15.80 real | 15.71 user | 0.02 sys |
| 5.1binary: | 22.57 real | 22.45 user | 0.03 sys |
| 5.2binary: | 24.16 real | 24.05 user | 0.03 sys |
Jean-Marc Zucconi wrote:
- I used the same kernel configuration file for the 4 systems. In particular I removed ‘cpu 486’, maxusers was set to 0, and I don’t have INVARIANTS or WITNESS options.
- the kernels and the worlds were compiled with CFLAGS=-O -pipe.
- /etc/malloc.conf is always linked to aj
- I/O tests were done on the same filesystem (IDE UDMA33 in all cases)
- bytebench binaries were compiled static. I used the same binaries
on FreeBSD 4.9-R and DragonFly; they were recompiled on
FreeBSD-5.[12] as the compiler is not the same.
Umm, 5.2 has changed from static to dynamic by default now. Do you think, it’s because of dynamic that hurt the performance big time?
It’s possible already answered in the mailing list/newsgroup. I am on the different computer right now, so this computer doesn’t has any mail/newsgroup client.