rtp1@tank.uchicago.edu (raymond thomas pierrehumbert) (12/08/89)
I bought my DN10000 with the 300megabyte drive, figuring that this would do me for a long time, until I managed to set up some optical secondary storage, or until the price of big, fast disks came down. Allright, so I am used to my various Macintosh setups, where I have labored for years without filling up my 80 meg of disk storage. Thought if I ever got 300 meg, I'da died an gone to heaven. Well, nobody expects Unix to be as compact as the Mac os, which after all doesn't do as much as Unix (though in some ways it does a lot more. Wish I had that 256K toolbox in rom on the 10000!). But Domain OS is a real disk hog. I installed BSD_medium, and found that the OS files take up an amount pushing 150 megabytes (even after deleting the games directory)--and yes, I did install using hard links. With the "medium" installation you don't even really get all the Unix you need, e.g. you don't get support for Unix "plot" under DM. And now I find that with 10.2, in order to get the X developer tools, I will need to do a "large" installation. This means I will really need to spring mucho $$$ to get an expensive Apollo 700MB disk to stay in business. Now, I don't want to get into the tired "apollo vs. sun" thing. I've had enough with the interminable and unproductive "mac vs ibm" thing (but alright, since you asked, since you twisted my arm, yes the mac is better than ibm and apollo is generally better than Sun, except that the sparcstation is a great little machine at a great price and I wish Apollo had a PrismStation, but no let's not start a thread on this, it was just because I heard you insisting to know ...:)). But the fact is people have Sun 3/50's with 70MB scsi disks running SunOS as stand-alone systems (i.e. no links to file servers anywhere), and still have room left over for their data files. I think it is probably easier for me to get more disk space than to hack around with custom installations (I am the prime user and sysadmin all in one of this 10000, and tend to think of it as a kind of expensive personal computer). For me, I like minst just fine. But is there some intrinsic reason why the OS on the DN10000 has to be so big? What am I getting for all that space? It is a real nuisance, because not only do you need more disk space, but you have to spend time backing it all up. Does anybody have any bright ideas as to what I could delete to free up significant amounts of space? What is all that stuff in the sau10 directory for? Is all of it really ] needed? What is really needed is a minst script tailored for an installation such as mine, which is strictly a single-node operation. Something that gives you what you need without all the fluff. Failing that, Apollo ought to offer a cheaper line of 700MB drives. I'd much rather be spending $13000 for a second processor than for another disk drive.
krowitz%richter@UMIX.CC.UMICH.EDU (David Krowitz) (12/08/89)
Hmm ... something doesn't sound right here ... I've loaded the "large" installation of Aegis plus BSD onto my DN560 with a 150MB disk. Granted, there was only 6MB left after booting and starting *all* the servers (rgyd, llbd, glbd, sendmail, tcpd, and 3 or 4 of my own). This was for SR10.2, including the X windows load with hard links. I can think of the following things to check: 1) MINST seems to load all of the M68K sau directories from the tape into the /install directory even though I only asked for one or two of the /sau directories. It may be doing something similar for the DN10000. 2) While helping someone else here at MIT who has a DN10000 load software, I noticed that the /sau10 directory contains a couple of subdirectories full of microcode and diagnostic code. A lot of the code is for the 40 and 80 plane graphics subsystem. One or the other set (or both, if you have the regular 8-plane graphics system) can be removed from the system. 3) When we loaded SR10.1p onto this person's DN10K, not everything that was supposed to be hard linked actually got linked. Some of the stuff was, in fact, copies rather than links. I don't remember which directories were involved, off hand, but you might poke around your system. We loaded both Aegis and BSD onto his system (with a 350MB disk like yours) and had 180MB left over. If your just loading BSD, you should have more like 200MB left. 4) Remember that under SR10.1p, every process running on the system takes up 5MB of disk space for virtual memory for the stack (0.5MB on M68K machines). If you have 10 servers running, which is not unusual, you have 50MB of disk space tied up for stack space. SR10.2 fixes this. Stack space is allocated dynamically as it is used. Elminating the extra /sau directories from my DN560 saved me about 20-25MB. If I remember correctly, the 40/80 plane graphics code and diagnostics were about 10 or 20MB, and the stuff that failed to get hard linked was another 20MB or so. -- David Krowitz krowitz@richter.mit.edu (18.83.0.109) krowitz%richter.mit.edu@eddie.mit.edu krowitz%richter.mit.edu@mitvma.bitnet (in order of decreasing preference)
vinoski@apollo.HP.COM (Stephen Vinoski) (12/10/89)
In article <6627@tank.uchicago.edu> rtp1@tank.uchicago.edu (raymond thomas pierrehumbert) writes: > Does anybody have any bright ideas as to >what I could delete to free up significant amounts of space? What >is all that stuff in the sau10 directory for? Is all of it really ] >needed? Most of the files in the /sau10 directory and subdirectories have to do with diagnostics. Apollo machines usually only have diagnostics which run under the Diagnostic EXecutive (DEX), but due to the enormous complexity of testing a machine like the DN10000, we chose to use scan testing techniques to provide high fault coverage and good fault isolation. Scan tests execute quickly, but require large amounts of input data and expected output data. Most of the files in the /sau10/scan directory are scan test data files which are used by the scan diagnostic executive (called the SCAn Test MANager, or SCATMAN) to test the machine. These data files are highly compressed but some of them are still pretty large. One good way to free up some disk space on a DN10000 which does not have Visualization System graphics (an X-Bus graphics board) installed would be to delete the /sau10/scan/graphics directory - that will give you ~10 megabytes back. HOWEVER, I strongly suggest that you avoid deleting any of the OTHER /sau10/scan subdirectories /sau10/scan/cpu /sau10/scan/memory /sau10/scan/utility if your DN10000 is a stand-alone system; without them, SCATMAN will be unable to test and diagnose the machine. Note that the DEX diagnostics were designed to be run only after the SCATMAN tests pass, since those diagnostics require hardware which is verified by SCATMAN. If you have multiple DN10000's on a network, only one of them really needs to have the SCATMAN and DEX diagnostics installed, and the rest can be tested by booting the diagnostics from the disk on that node. Of course, as soon as you purchase and install the X-Bus graphics board, you will have to reinstall that /sau10/scan/graphics directory... ;-) Hope this helps. -steve | Steve Vinoski | Hewlett-Packard Apollo Div. | ARPA: vinoski@apollo.com | | (508)256-6600 x5904 | Chelmsford, MA 01824 | UUCP: ...!apollo!vinoski | | "My second wife isn't even born yet." |
ross@cancol.oz (Ross Johnson) (12/11/89)
In article <8912081415.AA12414@richter.mit.edu>, krowitz%richter@UMIX.CC.UMICH.EDU (David Krowitz) writes: > Hmm ... something doesn't sound right here ... I've loaded the "large" > 3) When we loaded SR10.1p onto this person's DN10K, not everything > that was supposed to be hard linked actually got linked. Some of > the stuff was, in fact, copies rather than links. I don't remember > which directories were involved, off hand, but you might poke > around your system. We loaded both Aegis and BSD onto his system > This happened to me too (SR10.1p). I think it was most of /bsd4.3/usr/bin. Our 10000 originally came preloaded, but has since been invol'ed and reloaded off tape. I mention that because I asked the local Apollo engineer to check his 10000 which he found was ok. I presume his came preloaded too. > 4) Remember that under SR10.1p, every process running on the system > takes up 5MB of disk space for virtual memory for the stack > (0.5MB on M68K machines). If you have 10 servers running, which > ???? Unless 'df' is lying to me ('du' does!), each process only?! grabs 500k (SR10.1p) - same as for the m68k machines. That is, as a minimum. +----------------------+---+ | Ross Johnson |:-)| ACSnet: ross@cancol.oz.au | Info. Sciences & Eng.|___| ARPA: ross%cancol.oz.au@uunet.uu.net | University of Canberra | UUCP: {uunet,ukc}!munnari!cancol.oz.au!ross | P.O. Box 1 | CSNET: ross%cancol.oz@australia | Belconnen A.C.T. 2616 | JANET: ross%au.oz.cancol@EAN-RELAY | AUSTRALIA | BITNET: ross%cancol.oz.au@relay.cs.net +--------------------------+
lori@hacgate.UUCP (Lori Barfield) (12/12/89)
Discussing OS disk hogging, David Krowitz writes: >> 4) Remember that under SR10.1p, every process running on the system >> takes up 5MB of disk space for virtual memory for the stack >> (0.5MB on M68K machines). [...] In article <267@cancol.oz> ross@cancol.oz (Ross Johnson) writes: >Unless 'df' is lying to me ('du' does!), each process only?! grabs 500k >(SR10.1p) - same as for the m68k machines. That is, as a minimum. I remember hearing that firing up a diskless node under SR10 soaks up 6MB of disk, a lot of it for swap. Perhaps this is related. It certainly is a problem, although not exclusive to the 10k. ...lori (an SR10 wanna-be anyway)
rtp1@tank.uchicago.edu (raymond thomas pierrehumbert) (12/12/89)
I have received many interesting comments on this, and provide a digestion of some of the more critical ones: (1) My size estimates seem large. This is true. I did a crude estimate originally, based on df, and came up with the almost 150mb figure. I did a better estimate by taking a du, and subtracting the "install" directory and my "users" directory,which is where all our own stuff resides (except for Kermit, C and Fortran compilers). Result is 120MB, not 150. (2) The stuff in sau10 is sophisticated scan diagnostics, which is large by its very nature. Well, at least it's nice to know what I'm getting. Certainly, I'm all for good hardware diagnostics. Downtime, and MyTime are a lot more expensive than disk space (3) You could free up some space in sau10 by eliminating the scan/graphics diagnostics for 40 and 80 plane controllers, if you don't have them. The rest of the stuff is really needed. (4) Comparing the Sun 3/50's with Prism was apples and oranges. Point well taken, Indeed, the 3/50's were running an old Sun OS, and if they ever upgrade, they'll be in trouble. Further, object code for RISC architectures like Prism is always larger than for CISC architectures. Asking around, I have found that the OS for Sun 4's, Sparcs, and for SG Iris's are not much smaller than 10.1p So, all seems to be right with the world-- except I wish disk space were cheaper. Seems we are perhaps beset by the "creeping featurism" that the Unix gods used to inveigh against. I do wonder sometime about just what has happened to Unix. I need my DN10000, and I need my MacII's, but the fact is that the 1MB operating system on the MacII does about 80% of what an OS ought to, and am not a little disconcerted that getting the remaining 20% takes 119MB. This, of course, is not an Apollo-Specific comment.
krowitz%richter@UMIX.CC.UMICH.EDU (David Krowitz) (12/12/89)
Yeah, firing up a diskless node would start about 6 or 7 processes on the diskless node. Under SR10.1, each process had 500Kb of stack space plus whatever else they used in the way of common storage, static arrays, and the like. It's better under SR10.2. The stack space allocation is done differently. -- David Krowitz krowitz@richter.mit.edu (18.83.0.109) krowitz%richter.mit.edu@eddie.mit.edu krowitz%richter.mit.edu@mitvma.bitnet (in order of decreasing preference)
ALBRECHT@INTELLICORP.COM (Steve Albrecht) (12/14/89)
RE: Why is DOMAIN/OS so big? I was able to install a full BSD plus links to SYS5 and AEGIS on another node in 49MB. I happily have 98MB free on my little 160MB hard disk. I do not know why/if OS 10.1 requires lots more space on a DN10000 (I did this on a DN3500). (::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::) ) Steve Albrecht - IntelliCorp, Inc. - Knowledge Systems Product Development ( ( "Opinions expressed here are my own, if anyone's, and not my employer's." ) ) DDS albrecht@intellicorp.com : COMPUSERVE 73657,1342 ( ( UUCP ...!sun!intellicorp.com!albrecht : public bbs (415)969-5643 ) ) or ...!sun!icmv!albrecht : "c"omment to sysop ( (::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::) -------