[ Last edited by JonePeng on 2005-8-11 at 01:20 ]作者: Wengier 时间: 2005-8-11 01:38 本論壇采用UTF-8編碼,就是為了能更好得讓繁體中文、簡體中文同時顯示,使得大家能更好的交流。確實,本論壇,包括最上方的字樣等,均是Traditional Chinese。作者: johnsonlam 时间: 2005-8-11 14:55
Quote:
Originally posted by JonePeng at 2005-8-11 01:03 AM:
呵呵,繁体字没问题,此论坛是繁体字界面的。欢迎香港的DOS迷朋友!
>>The UltraDMA drivers "read" what speed is set for each disk. Such
settings are made by the BIOS before any drivers are loaded and are
SUPPOSED to be the BEST setting that the chipset and disks can run.
>>Also, XDMA uses virtually the SAME I-O logic as UDMA2
but XDMA's run-time logic is 64 bytes LESS, due to changes in "CHS"
code, etc. This can ALTER where input buffers are in memory.
>>An UltraDMA RULE is that I-O may not cross 64K boundaries, as chips
do NOT increment address bits 16-31 during DMA; the 1994 UltraDMA spec
IS an "El Cheapo" from Intel! So, DOS I-O buffers must be ALIGNED
for UltraDMA, in ways the program developers did NOT expect!
>> Have your friend re-arrange how he loads his system with XDMA
(load more drivers into upper-memory, etc.), so input buffers
HOPEFULLY are in one 64K memory area, and direct-DMA input
can occur.
>>Otherwise, a "misaligned" input buffer forces XDMA to use its XMS "local buffer" for input, followed by an XMS move (up to 30% slower on some CPUs).
Output-overlap is always "buffered" and is unaffected by alignment.
>>Note also that XDMA resides in 1280 bytes, UDMA2 in 1344 bytes, and
UDMA in 640 bytes (V7.0 UDMA or below were 608), so performance may
change as each driver may give DIFFERENT buffer "alignment"! Also
由 於 3 個 驅 動 的 大 小 不 一 ﹐ 那 個 緩 充 區 亦 未 必 一 樣 (對 正)
even on my 400-MHz system, I get speed-test results from 109 to 122
MB/sec. Small CPU/controller/disk clock variations DO cause speed
changes from run to run. Tell your friend not-to-worry about such
things; the UltraDMA drivers ARE doing the BEST they can! Even on
buffered input, they are at LEAST 5 times quicker than a BIOS doing
"PIO mode"! Best wishes! Jack R. Ellis