richard@locus.com (Richard M. Mathews) (09/10/90)
huntting@boulder.Colorado.EDU (Brad Huntting) writes: >>Halt! Who goes there? >>Is anyone out here? >I do... What I want to know is what is this newsgroup for? And who created >it? It is my understanding that this group is for discussions of Unix on large machines (how large is large?). I'd be interested in starting things off by finding out what large machines and what versions of Unix you have out there. For what do you use these systems? We have AIX/370 running on 370s including a 3090. We have been working with IBM to develop AIX on these machines. Since we use VM to run lots of small virtual machines in different configurations, perhaps we don't count as users of "large Unix" (except when helping a customer who IS using a large system). As a supplier of such systems, we are, however, very interested in the opinions and needs of such users. What other large systems are out there? What sorts of problems do you have that you feel are unique to users of large systems? Richard M. Mathews Locus Computing Corporation richard@locus.com lcc!richard@seas.ucla.edu ...!{uunet|ucla-se|turnkey}!lcc!richard
wojcik@crl.dec.com (Theodore Wojcik) (09/10/90)
In article <richard.652960541@fafnir.la.locus.com>, richard@locus.com (Richard M. Mathews) writes: |> huntting@boulder.Colorado.EDU (Brad Huntting) writes: |> |>..... |> It is my understanding that this group is for discussions of Unix on large |> machines (how large is large?). I'd be interested in starting things off |> by finding out what large machines and what versions of Unix you have out |> there. For what do you use these systems? |> |>..... |> |> What other large systems are out there? What sorts of problems do you |> have that you feel are unique to users of large systems? |> My feeling is that the problems associated with large clusters of workstations are as difficult and perhaps more complex than those associated with large single machines. The subjects discussed here are likely to be interesting to both . I'd like to suggest that the key is "large installation" and keep away from what exactly constitutes one. When you have one, you know it. As the original proposal stated: comp.unix.large Unix on mainframes and in large networks ^^^^^^^^^^^^^^^^^^^^^ I guess the point I'm trying to make is that there isn't any single attribute that makes an installation "large". By their very nature, mainframes usually have big user communities and a big farm of disks. By the same token, large clusters of workstations generally have big user communities and a disk server with a big farm of disks. It's my feeling that the two styles of computing have more in things in commmon than not. FWIW. |> Richard M. Mathews |> Locus Computing Corporation |> richard@locus.com |> lcc!richard@seas.ucla.edu |> ...!{uunet|ucla-se|turnkey}!lcc!richard -- Ted Wojcik, (wojcik@crl.dec.com), Systems Manager Digital Equipment Corporation Cambridge Research Lab 1 Kendall Sq. Bldg. 700 Flr. 2 Cambridge, MA 02139, USA (617)621-6652
fwp1@CC.MsState.Edu (Frank Peters) (09/11/90)
In article <richard.652960541@fafnir.la.locus.com> richard@locus.com (Richard M. Mathews) writes:
It is my understanding that this group is for discussions of Unix on large
machines (how large is large?). I'd be interested in starting things off
by finding out what large machines and what versions of Unix you have out
there. For what do you use these systems?
We have AIX/370 running on 370s including a 3090. We have been working
with IBM to develop AIX on these machines. Since we use VM to run lots
of small virtual machines in different configurations, perhaps we don't
count as users of "large Unix" (except when helping a customer who IS
using a large system). As a supplier of such systems, we are, however,
very interested in the opinions and needs of such users.
What other large systems are out there? What sorts of problems do you
have that you feel are unique to users of large systems?
Well, we are developing a different kind of 'large system' that has
its own unique complexities.
We are in the process of migrating from a mainframe environment to a
distributed environment of UNIX servers and workstations. Though none
of these systems taken individually are what I would call large
(though I wouldn't call a Sun 4/490 small either), taken together they
add up to more power and complexity than any mainframe I'm familiar
with.
There are a great many issues that are addressed in typical mainframe
operating systems (and perhaps by big iron unices like UTS and
AIX/370) that are all but ignored in the typical mini and workstation
implementations of UNIX.
I would hope that discussion of how varius users have addressed the
following issues would be appropriate to this group.
(1) Operator driven UNIX. The concept of a machine room operator is
ingrained in most mainfram operating systems. The concept of a
large installation in which many of the day to day administration
tasks are performed by operators in a machine room rather than by
the user or the system administrator with root priviledges is
relatively foreign to most UNIX implementations. There are
really two sub issues involved here.
a) We generally have two operators on duty in our machine room
during peak hours. Either of these operators must perform
tasks such as answering requests, killing runaway logins and
the like. Requiring each of these operators to log in under a
separate userid leads to an unacceptable amount of wasted
time. On the other hand, a single operator userid is a
definite security problem. We are currently attempting to
establish an operator userid that can only be logged in on the
system console in our machine room.
b) We would like to delegate many tasks such as tape control,
backup, printer control and such to our operators. At the
same time we don't want to share the root password. There are
a few systems out there to allow the delegation of tasks to
certain users. All of these, of course, have security issues
involved that must be considered.
(2) Tape device management. Another capability supported in
mainframe operating systems is the ability to gain exclusive
control of a tape device, request that a tape be mounted, and
release the device when the user is finished with it. We would
like to have this capability in our UNIX environment as well.
(3) Load balancing. In a single box balancing the load among several
CPUs is relatively straitforward (at least in concept). When
your CPUs are spread across a dozen or more machines how do you
avoid the situation of one machine being sunk to its knees while
another is nearly idle. When you add multiple classes of
processor (is a 4/490 at 50% more loaded than a sparstation at
30%?) or multiple types (how do the above two compare to a
decstation 3100 at 40%?) this issue can become a nightmare.
(4) Userid management. Most UNIX boxes come with instructions about
which several files should be edited to add a user to the system.
We are developing programs to manage the addition of userids in a
relatively bullet proof way so that non-technical personnel can
add new users. While there are programs to do that around very
few address the large system issues such as password file locking
and batch additions of large groups of users like a class roll.
(5) Accounting. We have many projects of one sort and another which
require accounting of resources. On our mainframe we have an
account number which can be used for this purpose. The same
userid can be under many accounts. The UNIX accounting system
(at least the one on a Sun running SunOS 4.1) doesn't seem to
have an equivalent concept. As far as we can determine no
accounting information by group is available so that the newgrp
command gives us no help. It looks as if we will have to have
multiple userids per user in this case. The thought of that is
repulsive both from a personal and administrative viewpoint.
I'm sure there are many other issues out there waiting for us that we
haven't thought of yet. Any discussion or suggestions on the above
would be most welcome.
Regards,
FWP
--
--
Frank Peters Internet: fwp1@CC.MsState.Edu Bitnet: FWP1@MsState
Phone: (601)325-2942 FAX: (601)325-8921
ikluft@uts.amdahl.com (Ian Kluft) (09/11/90)
In article <1990Sep10.153946.7269@crl.dec.com> wojcik@crl.dec.com writes: >I guess the point I'm trying to make is that there isn't any single attribute >that makes an installation "large". By their very nature, mainframes >usually have big user communities and a big farm of disks. By the same token, >large clusters of workstations generally have big user communities and a disk >server with a big farm of disks. It's my feeling that the two styles of >computing have more in things in commmon than not. FWIW. Point well taken. I think we have established enough direction to start discussions in the topic area. -- Ian Kluft ----------------------------- # Another flying fanatic UTS Systems Software \ |--*--| / # PP-ASEL Amdahl Corporation C - 172 /\___/\ Skyhawk # Member AOPA, ACM, UPE Santa Clara, CA o o o #include <std-disclaimer>
richard@locus.com (Richard M. Mathews) (09/11/90)
fwp1@CC.MsState.Edu (Frank Peters) writes: >Well, we are developing a different kind of 'large system' that has >its own unique complexities. Good point -- large networks are also supposed to be part of this newsgroup. Actually, we have both kinds of "large" systems. As I said before, we have many AIX guests running on our 3090. We also have a handful of other 370s, each with a number of AIX guests. Finally we have many PS/2s running AIX. Groups of these are connected via TCF, and we use good old fashioned telnet, rlogin, NFS, etc. to connect the clusters. >(3) Load balancing. In a single box balancing the load among several > CPUs is relatively straitforward (at least in concept). When > your CPUs are spread across a dozen or more machines how do you > avoid the situation of one machine being sunk to its knees while > another is nearly idle. When you add multiple classes of > processor (is a 4/490 at 50% more loaded than a sparstation at > 30%?) or multiple types (how do the above two compare to a > decstation 3100 at 40%?) this issue can become a nightmare. TCF allows processes to migrate between machines, and I know there are others developing similar capabilities. I can send a signal to a process to request that it move to a new site (by default, to the site from which the signal was sent). A load leveling daemon could be written (but one does not come with TCF) which automatically moves processes around in response to varying load. A difficulty is deciding which processes to move -- it would be a shame to waste time moving an I/O bound process which is currently accessing local data. In our environment I have found it quite sufficient to be able to manually move things when the load goes up. Richard M. Mathews Locus Computing Corporation richard@locus.com lcc!richard@seas.ucla.edu ...!{uunet|ucla-se|turnkey}!lcc!richard
russell@ccu1.aukuni.ac.nz (Russell J Fulton;ccc032u) (09/11/90)
What is a large Unix system? --- Probably anything with over 100 regular users. This eliminates individual workstations but should include clusters of workstations run from a server as well as systems serving lots of ascii terminals. We operate a SGI 4D/240S with is accessed by terminals on the network, either PCs or terminals connected to terminal servers. We usually have about 50 users on the system with about 1000 registered users. We also have a SG PI (Personal Iris, diskless) and a SUN 4/330) both of which mount most of the 240S's disk via NFS. I think this group could provide a useful forum for the discussion of issues relating to maintaing and using large systems. This is not an area in which UNIX excells. For instance we are currently having problems with the 240S crashing and not leaving any evidence as to the reason. It is very frustrating for us and our local support agents because we can't even decide if the problem is hardware or software. A workstaion crashing every now and again is a nuisance for us its disasterous. --
diamond@tkou02.enet.dec.com (diamond@tkovoa) (09/11/90)
Well, if this group is as large as the world, you could try saying "hello"
to it. ;-)
> Who goes there?
Monsters. ;-)
Large ones.
--
Norman Diamond, Nihon DEC diamond@tkovoa.enet.dec.com
(tkou02 is scheduled for demolition)
We steer like a sports car: I use opinions; the company uses the rack.
ggs@ulysses.att.com (Griff Smith) (09/11/90)
In article <FWP1.90Sep10140422@ra.CC.MsState.Edu>, fwp1@CC.MsState.Edu (Frank Peters) writes: > (2) Tape device management. Another capability supported in > mainframe operating systems is the ability to gain exclusive > control of a tape device, request that a tape be mounted, and > release the device when the user is finished with it. We would > like to have this capability in our UNIX environment as well. > Frank Peters Internet: fwp1@CC.MsState.Edu Bitnet: FWP1@MsState > Phone: (601)325-2942 FAX: (601)325-8921 Part of this is available now, at least for academic sites. I have a resource manager that we use to control access to tape drives on our systems. The operator request part will probably come soon. The down side: academic only, SunOS and BSD only, source not available yet because of licensing problems (sigh), not ported to System V yet. The up side: it's free if you qualify. See the proceedings of the 1989 Baltimore Summer USENIX convention for details. -- Griff Smith AT&T (Bell Laboratories), Murray Hill Phone: 1-201-582-7736 UUCP: {most AT&T sites}!ulysses!ggs Internet: ggs@ulysses.att.com
anselmo-ed@cs.yale.edu (Ed Anselmo) (09/12/90)
>>>>> On 10 Sep 90 19:04:22 GMT, fwp1@CC.MsState.Edu (Frank Peters) said:
Frank> b) We would like to delegate many tasks such as tape control,
Frank> backup, printer control and such to our operators. At the
Frank> same time we don't want to share the root password. There are
Frank> a few systems out there to allow the delegation of tasks to
Frank> certain users. All of these, of course, have security issues
Frank> involved that must be considered.
We have a setuid-root program that allows most of the above to be done
without having to be logged in as a super-user. Users in group wizard
can kill runaway processes (among other things....):
Menu for wizard.
0. Exit this Menu.
1. Control Printer Queues.
2. Remove Job(s) From Printer Queues.
3. Reboot System.
4. Halt System.
5. Terminate A Process.
6. Write To All Users Logged On This Machine.
7. Set Date & Time.
8. Alter Priority of Process.
9. Rebuild UserDataBase Alias Files.
10. Remove IPC Resources.
(The last option was added to remove IPC resources that ill-mannered
Linda programs started leaving around).
The similar "operator" program allows members of group "operator" to
do backups from a regular account.
Both programs log every action performed by the user.
Frank> (4) Userid management. Most UNIX boxes come with instructions about
Frank> which several files should be edited to add a user to the system.
Frank> We are developing programs to manage the addition of userids in a
Frank> relatively bullet proof way so that non-technical personnel can
Frank> add new users. While there are programs to do that around very
Frank> few address the large system issues such as password file locking
Frank> and batch additions of large groups of users like a class roll.
Yale CS uses the all-singing, all-dancing "User Database Program"
(udb) which tracks users, uids, mailboxs, mailing lists, machines,
serial numbers (among other things). Through a series of programs and
Shell Scripts from Hell, it's used to build and delete accounts
(assigning unique uids, and keeping them consistent across machines),
and rebuild the sendmail aliases files.
It has also managed to keep several generations of Yale undergraduate
summer programmers entertained for months on end.
anselmo[371] % xdb
Yale Data Base access program (xdb).
Version 1.4 (Exp) of 89/10/02 15:42:52 by long.
Type '?' at any prompt for help.
Trying eli.cs.yale.edu...[Connected]...[OK]
Establishing identity...[OK]
The Database Daemon welcomes anselmo-ed@bigbird
Figuring out who you are...[OK]
Checking for wizardhood...[Wizard]
I welcome Wizard anselmo-ed
Loading entities: distribution...entity...field...machine...mailing-list...person...program...pseudo-user...[Done]
wizard> sh anselmo
** person anselmo-ed
Fullname: Ed Anselmo
Status: staff
Expiration: 1999
Birthday: 4/25/59
Work-address: 51 Prospect St. (AKW) Room 012
Work-phone: 432-6428
Room-number: 012
Home-phone: 469-2562
Capability: arpanet, database
Workstation: bigbird
Group: facility
ID Number: 118
Mailbox: 'anselmo@ra, 'anselmo@yale-rt-alaska
wizard> sh machine bigbird
** machine bigbird
Fullname: bigbird.cf.cs.yale.edu
Description: alaska client
Operating-system: sun os
Host-id: 51005683
Component/Make/Model/Ser-Num: cpu sun 4/60fgx-12-p4 935f2634
yaleid: 066099
Install-date: 9/89
Location: 012
Primary-user: anselmo-ed
principle-investigator: facility
Owner: facility
Grant: overhead
wizard>
--
Ed Anselmo anselmo-ed@cs.yale.edu {harvard,cmcl2}!yale!anselmo-ed
russell@ccu1.aukuni.ac.nz (Russell J Fulton;ccc032u) (09/12/90)
I agree that the issues Frank Peter raised all need discussion and that this group seems a good place to start. I would like to comment on a couple of point he raises. fwp1@CC.MsState.Edu (Frank Peters) writes: >(2) Tape device management. Another capability supported in > mainframe operating systems is the ability to gain exclusive > control of a tape device, request that a tape be mounted, and > release the device when the user is finished with it. We would > like to have this capability in our UNIX environment as well. This is very important to us particularly as it affect our backup operations at te moment the operators leave backups running when they leave at midnight and the tapes are vuneralble to overwriting until they are unmounted in the morning. We change the protection on all the device drivers but its a messy solution. I have seen reference to systems that control access to devices but have never had time to follow them up. I would be interested in hearing from people who have used such systems. >(4) Userid management. Most UNIX boxes come with instructions about > which several files should be edited to add a user to the system. > We are developing programs to manage the addition of userids in a > relatively bullet proof way so that non-technical personnel can > add new users. While there are programs to do that around very > few address the large system issues such as password file locking > and batch additions of large groups of users like a class roll. This is one area where we may have something to offer. We are currently converting out user data base management program from our vax cluster to our new UNIX system. It is called ZUMP (Zeno Users Management Program, Zeno was a student computing system we implemented on a DEC 10 about 10 years ago, the name has stuck.) Firstly a little background, we are cronically short of staff and hence we try and distribute as much of the work for administration of our systems out to the deptartment that use them. When we got the DEC 10 it was our first interacive machine for student users. Which meant that each of about 5000 students would need individual usercode, diskspace allocation, etc. Also with that many student at least 10 will forget their password on any given day :-) anyway it all adds up to a lot of administration. What we did was to write a program that managed a hierarchical data base of users. At the top we had the computer centre, under that we had depts and under dept we had classes. Each class had a supervisor. The program allowed resources, process time, connect time, etc. to be passed down the tree. It also had the concept of capabilities that could also be passed on to users below. The most important capability was that of supervision. Thus we created the departments and gave them supervisor capabilities and resources and it was then up to the departments to create the classes and share out the resources as they saw fit. The departmental supervisor could creat a class and delegate the supervision of that class to somebody else. Somebody with supervisor status could move resources around for people under them and perform tasks such as setting passwords. When the DEC 10 was replaced by a vax cluster we move zump to the vaxes and exteneded it to take files from the administration computer to automatically create userids for whole classes. It was also tranlated into C at this time. We recently replaced our IBM VM system (which is used for research) with a SGI 4D/240S UNIX system and we decided to start using zump for controlling administration on it. We hope to have all user administration handled by zump next year. We now have a basic version going that tracks usage and will allow supervisors to change password. There are issues that still need resolving, inparticular we require users of our research systems to sign a statement saying that they have read the 'Computer Centre Regs' and agree to abide by them. Thus we cannot completely distribute the creation of usercodes to the departments. What we are thinking of doing is having zump create a skeleton user and add another capability which turns it into a real user. Thus the department does the donkey work of entering the details and fowards the signed form to us. We then activate the userid with a single command. We have been using zump in various guises for nearly 10 years and have found it very satistfactory. We would be happy to distribute the source to anybody who wanted it but I suggest that we wait until the UNIX version has stabalised. Has anybody else tried anything like this? --
matt@locus.com (matt) (09/12/90)
In response to Frank Peters' article: Frank asks about the following system management problems on large Unix systems. > (1a) Operator Ids. > (1b) Operator root privs. > (2) Tape Management > (3) Load Balancing > (4) User Management > (5) Accounting The OSF has recently sent out a request for designs for a system management tool package that addresses some of the above concerns. SCO's ODT on smaller unix machines has a menu driven system operations facility that helps a beginning user access the various unix utilities to do some of the above functions. I am developing a system management tool for AIX on large IBM iron that will allow full accountability and root privileges with restricted access to certain functions. It will cover the following areas of system administration: Inventory Control Change Management Problem Management System Management Network accounting and management Performance accounting and management System maintenance User account management over a distributed network System accounting and change logging The interface is through a menu with command line access to any function or sub-function along with complete customizability and portability. If anyone is interested or has any comments or observations I would be delighted to hear. I am also interested in anything along these lines that is being used that can be integrated with our software to increase its usability or functionality. If there are other system management functions that you would like to see added to the above list, or if there is a unix operating system that you would like to see this available for please respond directly to me at the following address. Matt Campbell (213)337-5900 matt@locus.com lcc!matt@seas.ucla.edu {randvax,ucbvax}!ucla-se!lcc!matt
wnp@iiasa.AT (wolf paul) (09/12/90)
In article <26107@cs.yale.edu> anselmo-ed@cs.yale.edu (Ed Anselmo) writes:
)We have a setuid-root program that allows most of the above to be done
)without having to be logged in as a super-user. Users in group wizard
)can kill runaway processes (among other things....):
Is this available??
)The similar "operator" program allows members of group "operator" to
)do backups from a regular account.
And this?
)(udb) which tracks users, uids, mailboxs, mailing lists, machines,
)serial numbers (among other things). Through a series of programs and
)Shell Scripts from Hell, it's used to build and delete accounts
)(assigning unique uids, and keeping them consistent across machines),
)and rebuild the sendmail aliases files.
)It has also managed to keep several generations of Yale undergraduate
)summer programmers entertained for months on end.
And this? the entertainment value alone would warrant making it
available, apart from its usefulness.
--
Wolf N. Paul, IIASA, A - 2361 Laxenburg, Austria, Europe
PHONE: +43-2236-71521-465 FAX: +43-2236-71313 UUCP: uunet!iiasa.at!wnp
INTERNET: wnp%iiasa.at@uunet.uu.net BITNET: tuvie!iiasa!wnp@awiuni01.BITNET
* * * * Kurt Waldheim for President (of Mars, of course!) * * * *
dricejb@drilex.UUCP (Craig Jackson drilex1) (09/12/90)
In article <7brK02ZFc7RI01@amdahl.uts.amdahl.com> ikluft@uts.amdahl.com (Ian Kluft) writes: >In article <1990Sep10.153946.7269@crl.dec.com> wojcik@crl.dec.com writes: >>I guess the point I'm trying to make is that there isn't any single attribute >>that makes an installation "large". By their very nature, mainframes >>usually have big user communities and a big farm of disks. By the same token, >>large clusters of workstations generally have big user communities and a disk >>server with a big farm of disks. It's my feeling that the two styles of >>computing have more in things in commmon than not. FWIW. > >Point well taken. I think we have established enough direction to start >discussions in the topic area. wojcik does have a good point, but I think in the area of disks there still is a boundary. Except for things like the Auspex, few file servers really support connecting more than about 10 Gigs of disk. (Also excepting Amdahls, 3090s, etc running Unix.) This is a real issue for us--we have a user who would like to keep about 20 GB online at any one time, with convenient access to another 100 GB of archives. Right now, even on our mainframe he can only keep about 10 GB up at a time--he pages to tape. (It's data about international trade. I don't know what the current paging scheme is, but at one time import data was available on even-numbered days, and export data on odd-numbered days.) Note that these data don't split real well across multiple boxes--any give update job will probably go through the whole thing, and any given access job may access throughout the database. Given his druthers, the customer would like to think of it as a single file... Sure, it's necessary to break it up for several reasons, but none of them are natural from the applications standpoint. -- Craig Jackson dricejb@drilex.dri.mgh.com {bbn,axiom,redsox,atexnet,ka3ovk}!drilex!{dricej,dricejb}
ces@milton.u.washington.edu (Christopher e Stefan) (09/16/90)
our site (*.u.washington.edu) runs several localy developed utillities that address some of the problems voiced here, specificly: 1. an account creation system tied into unique usernames and student numbers 2. an accounting system tied into the student number database providing load leveling when users exceed their cpu and connect time quotas 3. TMS, a tape management system, providing exclusive access to tape drives and check-in/check-out from the tape library 4. a mail forwarding system and, undoubtedly several more utillities useful to sysadmins at large sites that I either: a) can't remember, or b) have never heard of becasue I'm a user not an administrator. If you are interested in obtaining more information about the programs used at the University of Washington, or just exchanging information I would strongly getting in contact with either: Ken Lowe (ken@cac.washington.edu) or Ken Case (kc@cac.washington.edu) who are the people responsible for administrating the systems in the U.Washington.EDU domain, and who are also responsible for writing/maintaining most of the locally developed UNIX systems software. I hope this is of some help. :-) -- Christopher e Stefan. | INTERNET: ces@u.washington.EDU | (206) 526-9817 "Natural causes, | UUCP: ... !beaver!blake!ces | 4746 21st Ave NE people just explode." | BITNET: ces%milton@UWAVM | Seattle WA, 98105