jeff@drexel.UUCP (10/01/87)
We just received our upgrade of SCO Xenix to version 2.2 today, and I noticed in the manual that they now have a built in device driver for a ramdisk. However, at the end of the manual page, they say that performance can actually be WORSE, because the Xenix memory management can make better use of the memory. Our configuration is a Tandy 3000HD, with a 40 Meg drive and 3.5 Meg of ram. The users (average 2-6) use either SCO Professional (spreadsheet) or Open Systems (accounting package). I was thinking about having them buy 4 sets of 256k chips (for about $100) to populate the rest of one of the memory expansion boards, allocating about 1/2 Meg for the ramdisk, and then copying some of the executable to the ramdisk. Would this give me better performance than increasing the memory and just letting the OS have it all? I would appreciate whatever opinions or hard data people in the net have. Jeff White Drexel University - ECE Dept. rutgers!liberty!drexel!jeff
chkg@ptsfa.UUCP (10/02/87)
In article <522@drexel.UUCP> jeff@drexel.UUCP (Jeff White) writes: > > We just received our upgrade of SCO Xenix to version 2.2 today, and I >noticed in th
miket@ncoast.UUCP (Michael Taht) (10/06/87)
In article <522@drexel.UUCP> jeff@drexel.UUCP (Jeff White) writes: > > We just received our upgrade of SCO Xenix to version 2.2 today, and I >noticed in the manual that they now have a built in device driver for a >ramdisk. However, at the end of the manual page, they say that performance >can actually be WORSE, because the Xenix memory management can make better use >of the memory. I find that amusing. I've always thought that Xenix memory managment was something to avoid! A heavily loaded Xenix system lurches when it starts swapping, I mean, lurches, because it has to swap out whole segments in order to swap-If your system is swapping heavily (which I doubt) then don't use the ram disk. In other words, the best way to get increased performance is Not To Use Xenix memory management! I haven't even cracked my 2.2 doc. I believe, though, that a ramdisk emulating /usr/tmp and tmp would speed the performance of the C compiler greatly. (MSC, under Xenix is 7x slower than DOS). Does the Ram disk support multiple mount points, or do I have to create several? > Our configuration is a Tandy 3000HD, with a 40 Meg drive and 3.5 Meg of >ram. The users (average 2-6) use either SCO Professional (spreadsheet) or >Open Systems (accounting package). I was thinking about having them buy >4 sets of 256k chips (for about $100) to populate the rest of one of the >memory expansion boards, allocating about 1/2 Meg for the ramdisk, and then >copying some of the executable to the ramdisk. > Would this give me better performance than increasing the memory and >just letting the OS have it all? I would appreciate whatever opinions or >hard data people in the net have. You didn't specify whether or not you had an 80287. My tests on xenix with stuff like the savage benchmark, saysthat math is 100 to 1000 times slower without a FPU than with. Subjective tests: I had a spreadsheet that took 27 seconds to recalculate. After the FPU Strike <RET> and wait for the screen update (:>. This will also improve the speed of your accounting package as well. As a 287 is in approximately (weeeeeel) the sam price range the performance increase will be tremendous! Michael Taht "What in hell is TCP/IP? A disease?"
davidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr) (10/12/87)
In article <522@drexel.UUCP> jeff@drexel.UUCP (Jeff White) writes: | | We just received our upgrade of SCO Xenix to version 2.2 today, and I |noticed in the manual that they now have a built in device driver for a |ramdisk. However, at the end of the manual page, they say that performance |can actually be WORSE, because the Xenix memory management can make better use |of the memory. The important thing is to have enough memory to avoid swapping. If you have enough for user programs, and have physical memory left over, then you can make use of the RAMdisk to speed up the initial execution of a program. Note that if the program is already in memory running for another user, the code will be shared (as far as I can tell). Loading a large program when there is a copy in memory is demonstrably faster. I would check to see if you have programs which are not used continuously and which are used a lot. That sounds like utilities to me. I don't have enough memory to use a large RAMdisk, so I can't really do a lot of testing. -- bill davidsen (wedu@ge-crd.arpa) {uunet | philabs | seismo}!steinmetz!crdos1!davidsen "Stupidity, like virtue, is its own reward" -me
rosso@sco.COM (Ross Oliver) (10/22/87)
In article <4820@ncoast.UUCP> miket@ncoast.UUCP (Michael Taht) writes: >In article <522@drexel.UUCP> jeff@drexel.UUCP (Jeff White) writes: >> >> We just received our upgrade of SCO Xenix to version 2.2 today, and I >>noticed in the manual that they now have a built in device driver for a >>ramdisk. However, at the end of the manual page, they say that performance >>can actually be WORSE, because the Xenix memory management can make better use >>of the memory. > ... >In other words, the best way to get increased performance is Not To Use >Xenix memory management! >I haven't even cracked my 2.2 doc. >I believe, though, that a ramdisk emulating /usr/tmp and tmp would speed >the performance of the C compiler greatly. (MSC, under Xenix is 7x slower >than DOS). Does the Ram disk support multiple mount points, or do I >have to create several? RAM disks are treated just like physical disks, so you would need to create one RAM disk for each filesystem. However, the reason for the warning in the doc is that XENIX already uses RAM buffers to buffer disk I/O. When a disk block is "written" by a process, it really goes into a RAM buffer to be written later. When a disk block is requested, the kernel checks to see if it is already in a buffer. If so, no access to the disk is needed. Therefore, for all but a few specialized uses, RAM disks do not significantly increase performance. Moreover, when a RAM disk is in use, there is less memory available for user processes, so swapping is more likely and will occur more often. Ross Oliver SCO Tech Support {uunet, ucbvax!ucscc, decvax!microsoft}!sco!rosso My opinions are my own, my employer denies my existence...