Hi all,
Harald Koerfgen wrote:
> This SCSI driver still has rudimentary DMA support and will certainly fail
> under the following conditions:
>
> o Start address not longword aligned
> o End address not longword aligned
> o Number oy bytes to be transferred not a multiple of four
>
> It seems as if the higher layers handle this nicely for us because I
> haven't seen any problem so far with disk drives. Nevertheless: USE AT
> YOUR OWN RISK! Using this SCSI driver can damage your hardware!
>
> Support for Ultrix disklabels is experimental and mounting Ultrix
> partitions can destroy them. The UFS layer, when mounting Ultrix
> partitions, complains "ufs_read_super: fs needs fsck" and mounts them
> readonly. Be careful! Do not do this unless you know what you are doing.
>
The SCSI driver works fine, although the kernel doesn't recognise my NetBSD
disklabels. I'll look into that later. For now i've fdisk-ed one disk
with a standard PC partition table, and that works fine.
I've done some benchmarks on my DECStation, running Linux, NetBSD, and
compared them with Linux on a P90.
Here are the results:
BYTE UNIX Benchmarks (Version 3.11)
DECStation DECStation PC
5000/260 5000/260 Pentium-90
Linux 2.1.131 NetBSD 1.3.3 Linux 2.0.x
-------------------------------------------------------------------------------
Dhrystone 2 without register variables 104437.7 lps 119300.7 lps 111587.5 lps
Dhrystone 2 using register variables 104897.9 lps 119531.3 lps 111653.7 lps
Arithmetic Test (type = arithoh) 207193.5 lps 223145.1 lps 406588.3 lps
Arithmetic Test (type = register) 10190.5 lps 10981.4 lps 13537.8 lps
Arithmetic Test (type = short) 9961.7 lps 10777.0 lps 11885.6 lps
Arithmetic Test (type = int) 10193.4 lps 10980.4 lps 13539.5 lps
Arithmetic Test (type = long) 10207.1 lps 10980.0 lps 13541.4 lps
Arithmetic Test (type = float) 23665.2 lps 25481.7 lps 15642.3 lps
Arithmetic Test (type = double) 18489.3 lps 19916.5 lps 15655.7 lps
System Call Overhead Test 63271.7 lps 42755.5 lps 67654.3 lps
Pipe Throughput Test 42013.7 lps 14912.1 lps 49972.9 lps
Pipe-based Context Switching Test 16076.7 lps 6792.1 lps 16265.3 lps
Process Creation Test 38.7 lps 180.8 lps 538.7 lps
Execl Throughput Test 11.6 lps 32.4 lps 83.0 lps
File Read (10 seconds) 188359.0 KBps 71164.0 KBps 183065.0 KBps
File Write (10 seconds) 8936.0 KBps 2400.0 KBps 11898.0 KBps
File Copy (10 seconds) 5855.0 KBps 2299.0 KBps 2366.0 KBps
File Read (30 seconds) 184830.0 KBps 76089.0 KBps 185199.0 KBps
File Write (30 seconds) 11827.0 KBps 2133.0 KBps 18243.0 KBps
File Copy (30 seconds) 2823.0 KBps 2175.0 KBps 2210.0 KBps
C Compiler Test 32.4 lpm 39.2 lpm 97.3 lpm
Shell scripts (1 concurrent) 34.2 lpm 81.3 lpm 170.0 lpm
Shell scripts (2 concurrent) 18.4 lpm 45.3 lpm 89.0 lpm
Shell scripts (4 concurrent) 9.2 lpm 24.0 lpm 45.3 lpm
Shell scripts (8 concurrent) 5.0 lpm 12.0 lpm 23.0 lpm
Dc: sqrt(2) to 99 decimal places 1912.6 lpm ---- 3959.8 lpm
Recursion Test--Tower of Hanoi 1373.0 lps 1528.5 lps 2087.7 lps
-------------------------------------------------------------------------------
Bonnie results:
(DECStation 5000/260 only)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
NetBSD 100 1958 74.8 2057 18.2 1070 12.0 2244 91.6 2799 17.1 57.6 3.8
Linux 100 934 77.4 8120 91.3 1629 43.0 624 55.2 19759 92.1 1983.2 99.5
As you can see, the bonnie results show clearly that the currect SCSI driver
uses a lot CPU, but I assume that will improve in the future.
Good work Harald!
Regards,
Karel.
--
Karel van Houten
----------------------------------------------------------
The box said "Requires Windows 95 or better."
I can't understand why it won't work on my Linux computer.
----------------------------------------------------------
|