Hi all ! I got message about new linux hpt drivers from HighPoint Technologies. I probed it and compared with GPL kernel driver. This results maybe are interested for other users. HW === MB : MSI6163Pro (Intel BX chipset) CPU : PIII750@1000 (FSB 100->133, stable) RAM : 2x 256 MiB SIMM, 133 MHz ATARaid : PCI card with HPT370 chip, BIOS 2.3 HDD : 2x ST360020A (Seagate 60 GB, 5400 rpm) as RAID0 SW === linux RH7.2 + some updates kernel 2.4.18-0.21 (from RH Rawhide) self compiled On this RAID0 array I have two partitions - one with ext3, second with fat32. I probed throughput with this commands : Writing : ----------- time dd if=/dev/zero of=/mnt/.../test.xxx bs=1048576 count=1024 This command writes on array file with 1GiB size. Reading : ------------- time dd if=/mnt/.../test.xxx of=/dev/null This command reads the same file (1 GiB too). Test was made in single mode (init 1). Driver "HPT" is "source" driver from HPT version 2.3 compiled for used kernel. Driver "GPL" is build-in driver from kernel version 0.0.1 (I hope). Measured time is result "time.." command and described real | user | sys portion. Test was made for VFAT (fat32), EXT3 and EXT2 (partition manualy mounted with ".. -t ext2..." option. Maesurment was made minimaly three times for eliminating write cache etc. Results are minimal maesured times. RESULTS : ========= a) EXT3 filesystem a1) GPL driver a1w) writing 36.88 | 0.00 | 10.76 a1r) reading 18.88 | 1.77 | 5.19 a2) HPT driver a2w) writing 46.23 | 0.01 | 9.92 a2r) reading 30.13 | 1.92 | 4.32 b) EXT2 filesystem b1) GPL driver b1w) writing 35.98 | 0.02 | 8.30 b1r) reading 18.74 | 1.81 | 5.10 b2) HPT driver b2w) writing 39.10 | 0.01 | 6.39 b2r) reading 28.05 | 1.85 | 4.36 c) VFAT filesystem c1) GPL driver c1w) writing 36.51 | 0.01 | 9.76 c1r) reading 33.08 | 1.94 | 13.88 c2) HPT driver c2w) writing 37.94 | 0.01 | 6.42 c2r) reading 27.93 | 1.80 | 4.31 Recapitulation : ============ Driver for Riva chip on VGA cards made better with driver from HW manufacturer. This fact is expectable - manufacturer has detailed information about own product. But this fact don't hold for HPT chip. Drivers from HighPoint Technologies with binary module (nonGPL) is very slow. Diferences between nonGPL and GPL driver is about 8 / 20 % (writing on ext2 / ext3 filesystems) and even 23 / 38 % (reading ext2/3) !! Work with vfat filesystem is comparable - GPL is about 4 % faster on writing, nonGPL is about 16 % faster on reading. Driver HPT is little better for less system load - especialy on ext2 writing and distinct on both vfat work. Worse results maybe cause bad hdd settings. Works hdd with DMA ? With multiple IO ? I don't know. And here is no way to set up it - due hpt driver compilation was eliminate kernel support for HPT chips whereas with GPL driver I can set up parameteres wia /dev/hde and hdg devices. I thing that this results can be useful for another user ATARaid on HPT chips. If someone compares ext2(3) and another filesystems (e.g. RaiserFS ..) or made similar comparsion for new chips (e.g. HPT372), I will very grateful. Miroslav BENES system administrator TENEZ a.s. Chotebor Czech republic www.tenez.cz
