|
From: Alain <al...@po...> - 2005-06-29 15:56:00
|
Hi Eric... I am bringing back this talk to the list. I believe more people can beneffit from it there, and some can help too... >>>>rem dos=umb >>>why not? :-). plus dosdata=umb ... >>This is what makes raw speed slow. Very slow. > > I do not think so. It will be "slow if CERTAIN drivers are loaded > to UMB", so you can still use UMBs for the OTHER drivers and for > DOSDATA. Of course you have to find WHICH drivers... Does raw > speed get slow with UMBPCI only? Or even with EMM386? How slow is > very slow compared to normal? Like "UDMA loaded to UMB is even > slower than not using UDMA at all"? I made dozens of tests (as understanding grows with testing, it doesn't mean that I tested all:). It is the same with UMBPCI or EMM386. Slow is dropping from 36 to 2.4 Mb/s. Loading UMDA low, or the /L switch makes no difference. In fact DOS=HIGH also makes it slow too, but only around 10 to 20%. >>Even more: once a week I force a warm reboot so that >>everything is fresh and verified. > So your software is not overly stable? Or even the hardware? Software is ok, it usualy runs for months. Hardware is not stable at all and when it starts getting old (specialy the coolers), those extra reboots are very nice. Another issue that I noticed is that CodeBase makes a mess of some indexed that are too much modified (100% modified registers in 2 days, 5000 registers and 11 tags). I once had a user complaining about speed, his index file had grown from 1.5 to 200 Mb, the program was running non stop for 2 years! >>>calccach >>>lbacache %errorlevel% tuna tunw >>>for example "lbacache 40 tuna tunw" creates a 10 mb cache. >>That is a goog idea, I will test it... > > Even more: You could make calccach available for others :-)). Ok, itś GPL anyway (used parts of MEM), but I I believe it is of little interest no anyone... >>>Alternatively, calccach could just start the cache itself. >>I opted for the safer way, if the program dos not run (i really don't >>trust my users) I can use a default value. > > That would only happen if the users modify their autoexec, and in > that case, nothing would help anyway. And: If it does not run, it > does not have an errorlevel either. "Users can do the unexpected" Alain |