Jihad needs scientists
in [Design]
|
From: nonsense on 24 Feb 2007 18:08 Ken Smith wrote: > In article <erpb68$8qk_001(a)s934.apx1.sbo.ma.dialup.rcn.com>, > <jmfbahciv(a)aol.com> wrote: > >>In article <ermu1l$rph$2(a)blue.rahul.net>, >> kensmith(a)green.rahul.net (Ken Smith) wrote: >> >>>In article <ermmhd$8qk_001(a)s774.apx1.sbo.ma.dialup.rcn.com>, >>><jmfbahciv(a)aol.com> wrote: >>> >>>>In article <erhn0i$em5$4(a)blue.rahul.net>, >>>> kensmith(a)green.rahul.net (Ken Smith) wrote: >>> >>>[.....] >>> >>>>>Sure you can. If the computer is running a printer server, you can >>>>>predict the right order for the files to be read by the server. If there >>>>>is a task constantly running to take sliding billion point FFTs, you know >>>>>what is best for the FFT part. Just because the human may change >>>>>something it doesn't mean they change everything. >>>> >>>>All of this is single-task, single user thinking. >>> >>>No, it isn't. It is taking a practical example of a way that real >>>multiuser systems actually run. >> >>I know of plenty OSes and how they actually ran. We even made >>them go. > > > Then you should know that I am correct in what I am saying about the real > usage. > > >>>It is very common for a small fraction of >>>the tasks to need the large fraction of the memory. This is just the way >>>it is in the real world. >> >>That all depends on the computer site and who the users are. > > > Everything "depends", but 99.9999% of the cases are like that. There are > very few where the jobs are evenly spread. > > > >>>The computer that is doing the work of posting this is a multiuser >>>machine. It has me on it using only several tens of kilobytes. >> >><GAG> That's too much. > > > That's what I am using. "pico" is smallish and there is a little overhead > from "bash". > > >>>> Computer usage >>>>by the general population requires more than this. You keep >>>>looking at the load balance from a naive user POV. >>> >>>No, you are just making stuff up because you've been shown to be wrong >>>about the real world of computers today. >> >>Keep thinking like that and you'll never learn something. > > > From you! You have been shown to be wrong on this subject. > > > > >>>I'm using a company that sells computer time like a timesharing company. >>>They also sell modem access, and data storage. This is the modern >>>business model for this sort of company. >> >>And that is one business. > > > There are a great many like it now. There are also a lot of internet ISP > companies. They have the same sort of usage profile. > > > [....] > >>>You only think that because you haven't stopped to think about what I >>>wrote. We were discussing the case where swapping had to happen. There >>>is no point in asking at this point if it needs to happen because the >>>argument has already shown that it must. There is more data to be >>>maintained than will fit into the RAM of the machine. There is no >>>question about swapping being needed. The discussion is about the >>>advantages of having the code specific to the large memory usage make >>>informed choices about swapping. >> >>You are not talking about swapping; you are talking about the >>working set of pages. You do NOT have to swap code if the >>storage disk is as fast as the swapping disk. > > > What the devil are you talking about? You were sort of making sense until > you got to this. The "swapping" under discussion is between the swap > volume and the physical RAM. The swap volume can never be anything like > as fast as the RAM. A VM system makes it appear that there is more RAM > than is physically there by using the swap volume. > > Do you think that computers still use drum storage or mumble tanks for the > memory. It could just be her shorthand but she still talks about "core" which I remember well, and differing speeds of hard drives, diskpacks, and so on. I wonder if she is still using an 80ms full sized hard drive on her home system. That being said, a great deal of what she has been writing attaches to really elementary computer and OS design which, offhand, reading both of you going at it, she seems to understand better. It seems to me you're a level or few away from the sots of internals she worked with during her career. Most of those essentials haven't changed all that much. AFAIK the linux systems we're running continue to organize the hard drives much as early Unix organized tape magnetic storage. Certainly that was true as recently as 5-7 years ago, but I haven't messed with Linux on those levels in some time so that *might* have changed though I see no reason why it should have. (That and a buck will get you a cup of coffee.)
From: The Ghost In The Machine on 24 Feb 2007 13:08 In sci.physics, MassiveProng <MassiveProng(a)thebarattheendoftheuniverse.org> wrote on Sat, 24 Feb 2007 09:23:06 -0800 <92t0u2580p6fbatoie3ipui683hgkbefce(a)4ax.com>: > On Sat, 24 Feb 07 13:44:56 GMT, jmfbahciv(a)aol.com Gave us: > >>Using your method, a restore would have to start with Backup Tape >>#1, then #2, then #3, ....finishing with the last backup tape made. >>If you have been doing this for three years, you have 1000 tapes >>to restore. > > Nobody said ANYTHING about tape based magnetic storage. > > You are one old dip, I'll say that. Tape-based storage is still used, though nowadays the "tapes" are a far different structure than the old 1/2" reels commonly portrayed in old movies, or even the 1/4" cartridge units some may be familiar with. Today's units are weird-looking and designed to be used with automated storage systems ("jukeboxes"). -- #191, ewill3(a)earthlink.net If your CPU can't stand the heat, get another fan. -- Posted via a free Usenet account from http://www.teranews.com
From: jmfbahciv on 25 Feb 2007 07:24 In article <eos0u259h8fld2e3k6mhot9h6kobif44dj(a)4ax.com>, MassiveProng <MassiveProng(a)thebarattheendoftheuniverse.org> wrote: >On Sat, 24 Feb 07 13:24:40 GMT, jmfbahciv(a)aol.com Gave us: > >> >>Note that you do not have to swap code if it's reentrant; you can >>overwrite it in the RAM if the code segment resides on a device >>whose retrieval speed is equiavlent to memory fetches. OTOH, >>you do have to preserve data. > > > > RULES: > > 2 500MB code blocks that are in the form of a FILE (FileA, FileB) >that has to be processed in 500MB RAM. > > You're solution: > > Run (we call this the name of the executable) both together on the >machine at the same time. > > The REAL solution: > > A simple batch file. > >run FileA > >run FileB You are assessing this from the user POV. There is a lot of background code executing behind the user mode runs of fileA.exe and fileB.exe. Consider the number and locations of instructions needed to do the I/O in behalf of those two user EXE files. /BAH
From: jmfbahciv on 25 Feb 2007 07:27 In article <erpqdl$c02$6(a)blue.rahul.net>, kensmith(a)green.rahul.net (Ken Smith) wrote: >In article <erpeao$8qk_001(a)s934.apx1.sbo.ma.dialup.rcn.com>, > <jmfbahciv(a)aol.com> wrote: >>In article <ermuqc$rph$3(a)blue.rahul.net>, >> kensmith(a)green.rahul.net (Ken Smith) wrote: >>>In article <ermlh8$8ss_012(a)s774.apx1.sbo.ma.dialup.rcn.com>, >>> <jmfbahciv(a)aol.com> wrote: >>>>In article <erhnfn$em5$5(a)blue.rahul.net>, >>>> kensmith(a)green.rahul.net (Ken Smith) wrote: >>>>>In article <erhd4t$8qk_001(a)s916.apx1.sbo.ma.dialup.rcn.com>, >>>>> <jmfbahciv(a)aol.com> wrote: >>>>>>In article <i70nt25k4ubuvllr029cun9ebu1e1bng0a(a)4ax.com>, >>>>>> MassiveProng <MassiveProng(a)thebarattheendoftheuniverse.org> wrote: >>>>>>>On Mon, 19 Feb 07 13:29:06 GMT, jmfbahciv(a)aol.com Gave us: >>>>>>> >>>>>>>>Not at all. OSes were handling the above problems in the 60s. >>>>>>>>The reason virtual memory was invented was to solve the above >>>>>>>>problem. >>>>>>>> >>>>>>> The swapping, in this case, CAUSES the interference. >>>>>> >>>>>>That has to do with severe memory management problems. >>>>> >>>>>No, it is just part of the overhead of doing VM. >>>> >>>>Nope. VM doesn't have to swap. Swapping is done to make >>>>room so a memory request by another task can be serviced. >>> >>>You said "nope" and then confirmed that my statement was correct. >>>Swapping needs to happen if you need more virtual RAM than there is real >>>RAM. To be able to swap, the code for doing the swapping must always be >>>in the real RAM. As a result, there is code overhead in having a VM >>>system. There is also a speed overhead when swapping happens. The OS >>>uses some amount of CPU time on the taks switching needed to make VM do >>>the swap. >> >>VM isn't swapping. VM allows the OS to manage smaller chunks >>of memory rather than segments. > >That is completely and totally worng. "Virtual memory" means quite >literally "memory that is not real". No. It is memory whose addressing is larger than available physical memory. >MVT360 managed memory on a IIRC 4K >page basis. This certainly did not qualify as a virtual memory system. > > >>>>I don't know why but people often confuse virtual memory >>>>addressing with swapping. The two are separate. >>> >>>No, you are confused on the issue of needing to swap. >> >>No, I'm not. You are. > >It seems that your confusion runs much deeper than that. You seem not to >understand what "virtual" means. I know how OSes consider it. I was there when it was first implemented on our architectures. JMF, my other half, did the work. <snip> /BAH
From: jmfbahciv on 25 Feb 2007 07:32
In article <erpnd5$c02$2(a)blue.rahul.net>, kensmith(a)green.rahul.net (Ken Smith) wrote: >In article <erpfgo$8qk_003(a)s934.apx1.sbo.ma.dialup.rcn.com>, > <jmfbahciv(a)aol.com> wrote: >>In article <ur3vt2tl5lg3ujbtu14tgksbjd6s35h51j(a)4ax.com>, >> MassiveProng <MassiveProng(a)thebarattheendoftheuniverse.org> wrote: >>>On Fri, 23 Feb 2007 14:59:04 +0000 (UTC), kensmith(a)green.rahul.net >>>(Ken Smith) Gave us: >>> >>>>No great amount of care is needed. I've done that sort of restore a few >>>>times with no great trouble. Since files are stored with the modification >>>>date, a copy command that checks dates does the hard part. >>>> >>> Batch (read DOS type batch file) driven backup routines worked >>>flawlessly for me for YEARS, and only backed up what was needed, and >>>never overwrote a newer file with an older file. >> >>Using your method, a restore would have to start with Backup Tape >>#1, then #2, then #3, ....finishing with the last backup tape made. >>If you have been doing this for three years, you have 1000 tapes >>to restore. > >No, if the copy checks the dates, you can load the backups in any order. >What you do in practice is mount the complete backup and then the newest >incremental. You then mount the previous incremental and then the one >before that. This is one way to do a full restore. Note that it may also restore the cause of the problem. > >If your software is any good, it will let you know when you can stop. All >that's needed is to record the dates on all of the files. A fairly simple >script can tell you if you have more to do. Software cannot tell you if the file you want is no longer on storage. > > >>Another method was to do a full backup save each day. This will >>work until you find that you lost a source file sometime in the >>last 12 years. Now how do you find the last save of that file? > >This is not a problem in practice if the copy is smart about dates. AFAIK, only our system had enough dates stored in each file's RIB (retrieval information block) that could do this. > >The usual practice is to do a full backup every so often and incremental >ones in between. Yes but only for static storage. This will not cover data that is transaction-based. /BAH |