First OpenVMS Boot On IA64 300
vaxzilla writes "At 3:31pm EST on Friday, January 31st, 2003,
OpenVMS for the Intel IA64
architecture
successfully booted and ran a DIR command.
The Intel Itanium family of processors is the third architecture supported
by OpenVMS in its
25 year
history. Originally it ran on Digital Equipment Corporation VAX
systems; in the early 1990s, support was added for the DEC Alpha
processors. Following the acquisition of DEC by Compaq, and more recently
Compaq by HP, the Itanium and Itanium2 port of OpenVMS is now being
undertaken by HP. Congratulations on a job well done to the folks at
ZK03 in Nashua, NH!"
Wow!!! (Score:5, Funny)
This is truly a breakthrough. Intel is waay ahead in computing than companies like Nike or Coca Cola.
Re:Wow!!! (Score:2)
Re:Wow!!! (Score:2)
The more things change (Score:2, Funny)
Because it was PR0N.
or it contained (Score:2)
Modern VMS applications? (Score:3, Interesting)
Re:Modern VMS applications? (Score:4, Interesting)
Re:Modern VMS applications? (Score:3, Interesting)
VMS -> Unix is like Unix -> NT.
Of course since fewer people are using VMS there are fewer data points.
Re:Modern VMS applications? (Score:3, Insightful)
You do know that Unix hails from 1970 and VMS from 1978 don't you? It always amazes me when Unix kiddies don't seem to realize that VMS is actually more modern.
People use VMS when scalability and reliability matter. It's perhaps 15 years ahead of Unix for that (i.e. VMS clusters 15 years ago are where Unix clusters are now). You can do useful stuff like add a node to a cluster, migrate the applications onto it, shut down the original node, and the users won't even notice a gap in application availability. Add to that real ACLs and a versionning, journalled filesystem (things that modern Unix has only gotten in the last few years), and very fine-grained tunability, for example you can set the working set size per process and configure the system to assign different priorities to programs or users at different times. And DECnet is smart enough to authenticate user at the packet level, inherently more secure than TCP/IP.
Essentially, VMS died because DEC was run by engineers who thought that a good product would sell itself, whereas Sun et al were smart enough to hire marketers by the boatload.
Re:Modern VMS applications? (Score:2)
Re:Modern VMS applications? (Score:2)
DCL (the CLI for VMS) didn't need any more special keys than bash does. At the time I used it, its biggest weakness was an extraordinarily lame way of implementing interprocess pipelining. It was so slow that you would re-write utilities to be specificially interprocess aware and talk through mailboxes (which had nothing to do with e-mail) rather than let the shell handle the commands as filters, a la *nix.
The priviledge model on the other hand, was beautiful. You could keep know-it-all snots just out of college (like me) out of trouble by giving them only what they needed to do for the job. Unix's model of God and Peon pales by comparison.
And the clustering. Jeebus, you've not even seen clustering until you've seen VMS. We were doing more with clusters and shared resources in 1989 than most systems can do today.
Digital had the marketing sense of a camel's rear end, but they hired damned good engineers. Just goes to show that the best technology doesn't always win. Something to chew over when you contemplate the fate of the Unices versus the marketing monster from Redmond.
Re:Modern VMS applications? (Score:2)
Go to Eurex [eurexchange.com], for example. They are the largest electronic financial derivatives exchange in the world and their core systems run OpenVMS. SWIFT (the money transfer people) still do a lot with VMS as do many other people.
I have been looking at the problem of rewriting exchanges onto modern, cheaper hardware platforms with other operating systems. It really isn't easy.
Re:Modern VMS applications? (Score:2)
Pretty much anyone serious about process control or mission critical stuff uses VMS. UNIX simply cannot compete with the levels of reliability those systems routinely achieved. Uptime measured in years is normal. Unscheduled downtime is due to hardware failure - PERIOD.
Optimized for Fourier wave analysis???? (Score:3, Informative)
Regarding your claim that VMS is optimized for Fourier wave analysis, I can't believe that this is unique today. The main impetus behind VAX BSD was ARPA's desire to have a Unix system that would handle the memory demands of computer graphics. We made use of various Unix systems at Pixar and Pixar's predecessors, where there were similar sorts of problems (texture rendering rather than FFT) and Linux is now the darling of those places.
This is a VM and cache issue, not really rocket science these days.
Thanks
Bruce
Reasons to use VMS (Score:5, Insightful)
Re: Reasons to use VMS (Score:5, Interesting)
> but is OpenVMS really good for anything _new_ today?
The answer to your question is cultural rather than technical. VMS is a superb OS, but it is now viewed as déclassé in most circles, so it only has a thin slice of mindshare. That's not really any more a reflection on it than the thin slice of mindshare given to some very excellent programming languages.
I more than half wish the OSS revolution had centered around VMS rather than UNIX. There's not the slightest reason we couldn't be doing all the things we do under VMS... except the "price architecture". Put a free+open version on x86 and Linux might have some hot competition.
Re: Reasons to use VMS (Score:2)
VMS will never be forgotten. It may be complex kind of user friendly, but the system was a work of art. Hopefully, parts of it will find the mainstream and live forever.
Re: Reasons to use VMS (Score:2, Insightful)
Back in the 80's I had access to our campus VMS machine, and to the Unix box. I think the big difference was that the VMS machine very much restricted what I could do, the Unix machine didn't. So, I spent more time on the Unix machine. This probably has two root causes: VMS gave the admin more ability to control users, and VMS was more expensive to run than Unix in terms of hardware and whatnot (so the Unix admin didn't care as much).
As mentioned elsewhere, this is certainly a "culture" issue. But VMS seemed to enable a culture of control, while Unix enabled more anarchy. OSS software falls out of anarchy.
Re:VMS is the worst OS ever. (Score:5, Informative)
The fact that VMS HAS options which allow extremely fine-grained selection of user privs is a positive thing about the OS. VMS also had all kinds of login security years (break-in detection and evasion) before other systems, and was designated "trusted" quite early on.
VMS could be mismanaged so that it would crash, if ALL logging options were enabled. But that doesn't make it bad for it to have had so many different logging options.
Diskquotas weren't even enabled by default when I was using VMS. You *could* enable them (and obviously your silly sysadmins both enabled them and put very low limits on you), but you never had to.
VMS is a very flexible tool, and tools can be made to do lots of things, some good, some bad.
By the way, even now there aren't that many systems with the availability and redundancy VMS clusters had in 1985 (automatic failover from one machine to another, separate shared disk controllers, etc. etc.).
Finally VAX/VMS virtual memory worked better than any other such system I've seen. You could actually let things page and they didn't slow down much, since the paging was so intelligent.
*sigh* anyway, that was all a long time ago. I haven't used VMS professionally since 1992 or so...
Re:VMS is the worst OS ever. (Score:2)
Re:VMS is the worst OS ever. (Score:2)
VMS is a modern OS. It had features decades ago that are considered 'bleeding edge' to the Windows & Unix world.
Re:VMS is the worst OS ever. (Score:2)
VMS was also a very good application platform with incredible stability and a wonderfully rich set of APIs.
The only criticism I would offer is to do with the DCL shell which, although fairly feature rich, was hardly rational or intuitive in design. Lexical functions for example - did *anybody* ever learn to handle the context-sensitive calls (like the ones to do with queue information) without having the manual open beside them?
More fundamentally, DCL lacked command-line pipes - which are hard to do without once you have tasted Unix. When VMS became OpenVMS a POSIX shell was added, but it was very sluggish in processing large files through pipelines. I once heard an explanation something along the lines that the POSIX shell command line pipes in OpenVMS v6 were implemented using terminal IO rather than buffered IO and this slowed things down. I don't know what the truth is about that.
Having said all that there were some things VMS enjoyed that Unix doesn't. I have often lamented the absence in Unix of VMS logical name tables which in VMS allow you to define a "logical" device which is actually an ordered searchlist of directories. This allows any application *transparently* to make use of a PATH-like variable for any and all file accesses. Its a very useful thing to have and if you have never used it you may be surprised to learn of all the places where it comes in useful. The most obvious example would probably be where you want a hierarchy of config files (specific in turn to site, host, user, pwd).
Re: VMS is the worst OS ever. (Score:4, Insightful)
Which just reinforces the parent rant. On a PC, that 100MB would cost ten cents. Maybe instead of rationing disk space, the sysadmins could save more money for the company by scavenging abandoned half-full cups of coffee in the break room and pouring them back into the coffee pots.
Re: VMS is the worst OS ever. (Score:2)
> > And if he was speaking about 15 years ago, 2 MB might have been a very generous disk quota. I know of UNIX shops where you feel lucky to get 100MB even today.
> Which just reinforces the parent rant. On a PC, that 100MB would cost ten cents.
Actually, there was a study about a decade ago that showed that the average PC cost a company $15,000-$20,000 per year to own and operate: $5,000 for the official costs and another $10,000-$15,000 for the run-down-the-hall-and-bother-an-expert cost.
Some companies might reasonably conclude that it's a better bargain to give you less disk space on a server. Especially if your work doesn't require a lot of space in your personal directory.
Re: VMS is the worst OS ever. (Score:2)
What if the cheap IDE disk in the secretaries PC fries? all her data is lost... If a single SCSI disk as part of a raidset on a high end server dies, then the disk is just replaced and the array rebuilds itself. And ofcourse if all the important data is stored in one place it can be backed up all at once too, just incase worse things happen to the drives.. It would be far too costly to go around everyone`s workstations taking backups every day.
When it comes to portability, a given employee can login anywhere in the office, or even in remote offices if necessary, and access all their files, doing this with 500+ desktop machines is incredibly difficult, due to the frequency with which they may get unplugged or rebooted etc..
However, the admins should take into account what business the company does, and provide adequate space for each employee, so that assuming they do their work and dont try to save loads of downloaded porn, they shouldnt have a problem.. Ofcourse ever less efficient fileformats will just keep making this more difficult.
Re: VMS is the worst OS ever. (Score:2)
VMS is a fully featured os, it has a kernel, virtual memory, etc etc etc... DOS is by comparison extremely simplistic... if you want a modern comparison install dos windows and linux on a single machine and compare the speed of your calculation, dos will be quicker because it can divert 100% of the processing power to your task... Unless ofcourse you have to use the crappy dos disk drivers.
The VMS system also most likely had a lot of other users all trying to do things at once.
But this all goes to show the inefficiency present in modern os designs, some is essential for the task (multiuser, protected memory, multitasking etc) and many things are not.
Re:VMS is the worst OS ever. (Score:2, Informative)
The fact that we can lock the system down as much as we do is why it hasn't been replaced by more "fashonable" OS's
Re:Reasons to use VMS (Score:2)
Re:Reasons to use VMS (Score:2)
Re:Reasons to use VMS (Score:4, Interesting)
Re:Reasons to use VMS (Score:5, Informative)
Re:Reasons to use VMS (Score:4, Informative)
And since a VMS cluster can be fully upgraded automaticaly without any downtime to the cluster as a whole, the system can be continuely upgraded with no downtime to the users.
OpenVMS's clustering is the reason why most VMS users think that it's so cool. Think about it - 15 years of uptime. That's insane.
Re: Reasons to use VMS (Score:2)
Re: Reasons to use VMS (Score:3, Informative)
Use the right hardware, and yes, it does. (I believe that OpenVMS also supports removing a processor by automatically moving any processes running on that processor to a new processor with no data loss. I forget if you need to tell OpenVMS to "turn off" the processor or if you can just pop it out. I think certain combos of VMS and hardware allow true hotswapping, where the CPU can just be plucked out with no loss to the OS.)
With clustering, you can add new computers to the cluster. Therefore, the cluster does not run on continuously 15-year old CPUs. You take a computer out of the cluster, upgrade the computer, and reinsert. (Or, more likely, get a new, more powerful computer, add it to the cluster, and retire a computer that's due to be removed.)
Therefore, the cluster itself stays running 24/7 with no downtime, and the individual computers in it can be upgraded over time.
So no, an OpenVMS cluster with an uptime of 15 years is not running on 15-year-old hardware or running a 15-year-old OS - it can be continuely upgraded throughout its life. And, generally, it is.
Re: Reasons to use VMS (Score:2, Informative)
Certain models of VAX hardware were fault-tolerant and had more than one CPU (usually three); the CPUs would "vote" by running the same code and comparing results. If two CPUs agreed and one didn't, the one that didn't was marked faulty, and a message was sent to the operator.
The future was ten years ago; now it's just in syndication.
Open? (Score:2)
Re:Open? (Score:2, Informative)
Re:Open? (Score:4, Interesting)
In those days, there was a lot of fuzz among customers about the need to buy "open" systems and not "proprietary" ones, (meaning that VMS was proprietary and Solaris or HP-UX were open). That's why Digital felt POSIX branding was a good thing.
Then the customers bought lots of M$ Windows stuff instead; so much for wanting "open" systems!
Another fact: VMS came with source code from the start! On microfiche. Not so that you could recompile the OS, but rather learn about it, check bugs, etc.
Re:Open? (Score:4, Interesting)
Much of VMS was written in BLISS-32 whose back-end produced fantastic code - sure it was code no human being would have written - but damn good code just the same.
Being able to pull out the microfiche and check out the BLISS source was often useful when learning to program deep into the OS.
Not to mention the DECUS meetings where you could talk to the developers. I can remember the meeting in LA when at a small session DEC and the guys from MIT revealed the 782 - assymmetric multi-processing. It was exciting stuff. DEC had some really good engineers.
Remember - the VAX was about the ultimate CISC processor. Memory was scarce in those days - having 64MB of RAM was a big deal! The processor was very efficient in the use of memory.
Re:Open? (Score:5, Informative)
Core OpenVMS
http://www.openvms.compaq.com/ [compaq.com]
OpenVMS Future Release Contents, Schedules
http://www.openvms.compaq.com/openvms/roadmap/ope
OpenVMS and Core Layered Product Documentation
http://www.openvms.compaq.com/doc/ [compaq.com]
http://www.openvms.compaq.com:8000/ [compaq.com]
http://www.openvms.compaq.com/commercial/ [compaq.com]
Core OpenVMS Support Search Engine URLs, FTP Patch Area http://askq.compaq.com/ [compaq.com]
http://ftp.digital.com.au/pub/ecoinfo/ [digital.com.au]
ftp://ftp.service.digital.com/public/vms/vax/... [digital.com]
ftp://ftp.service.digital.com/public/vms/axp/... [digital.com]
The OpenVMS Freeware
http://www.openvms.compaq.com/freeware/ [compaq.com]
Encompas
http://www.encompassus.org/ [encompassus.org]
Tech Help OpenVMS
http://askq.compaq.com/ [compaq.com]
http://www.openvms.compaq.com/wizard/ [compaq.com]
ftp://ftp.service.digital.com/public/vms/vax/... [digital.com]
In other news: (Score:2, Funny)
Typo (Score:2)
oh god, I actually remember that...
Re:In other news: (Score:2)
LOAD "$",8
LIST
Take THAT OpenVMS!
WOW! (Score:5, Funny)
No, wait... what the hell does this matter? We're shutting the few remaining vaxes at work off soon...isn't everyone?
- A.P.
Re:nope... (Score:2)
Very cool.. wish I still cared about VMS (Score:4, Interesting)
Compared to the various dialects of unix, the VMS environment was so much friendlier and forgiving... I'm only now realizing how much my hands were in mittens using it. I'd still prefer a system that wasn't so case-sensitive.
The chief engineers behind VMS then went to work at Micro$oft to develop NT, so some of the legacy is still there: expensive process starts, but a nice memory model to work with.
Strengths:
Linkers in the early 80s that were easy to cross languages in a single project
A powerful set of run-time libraries, including some excellent flatfile databases
A scripting language that had access to a nice library of "lexical" functions.
But like I said, I wish I still cared. While we still have Alphas around running openVMS at the office, I haven't logged onto one for about three years. Somewhere, I have a huge library of shell routines, login scripts, and ancient forms-oriented code.
Oh Alpha, where art though? (Score:3, Insightful)
Re:Oh Alpha, where art though? (Score:2)
Bruce
Re:Oh Alpha, where art though? (Score:3, Interesting)
Remember the iAPX 432? A message-passing architecture implemented in hardware, with every function living in its own privilege ring. You could do that machine right today. And you wouldn't have to use ADA to do it. Someone should make more new silicon in this direction. IA64 goes a little bit in this direction, but it's too easy for operating systems and compilers to not use its capabilities in the name of portability. With the '432, there was never any possibility of the OS running on anything else.
Bruce
Re:Oh Alpha, where art though? (Score:3, Interesting)
With the original VAX architecture, there were four privilege rings. You had the usual user space and kernel space. You had exec space which was used by RMS (Record Management System) and allowed it to do complicated things like cross user buffer sychronisation, securely but outside the kernel. It was also used by some of the Digital database systems. The other mode, supervisor, was for the command intrepreter. VMS fully used all these special facilityies which made it non-trivial to port to other architecture (they tried initially to go to MIPS, but that failed).
Best part of VMS? (Score:5, Informative)
if you have foo.txt and you save another foo.txt in the same directory, you get foo.txt;2 !
damn, i wish Windows had that.
-c
Re:Best part of VMS? (Score:3, Interesting)
Re:Best part of VMS? (Score:2)
Forget NT/XP, I want it back on Linux now - I really miss it, especially whenever I screw up. Of course I have no chance adding versioning to WInodws, but I know that someone could at least do that to Linux.
Regrettably, I don't thin k it will happen. A lot of Unix people who briefly touched VMS looked upon this as an inconvenience. VMS had its problems compared to Unix, but this was't one of them.
Re:Best part of VMS? (Score:2)
With VMS you could edit something a few times and you would have those versions hanging around until you or your system manager did a purge. This was really useful. With CVS/RCS, you must think about it first.
Re:Best part of VMS? (Score:2)
You can do everything under Unix, but because one developer does things one way, it doesn't mean to say that other apps would work with it. OpenVMS provides the support to everything, and it doesn't matter which language it is written in.
Re:Best part of VMS? (Score:2)
Instead, try 'Versioning File System for Linux' on Google. You might hit some interesting link.
Now all we need... (Score:2)
VMS can bite me (Score:2)
Re:VMS can bite me (Score:2)
Actually LS is an abreviation for LSE brings up Digital's Language Sentsitive Editor (LSE) which was based on TPU. TPU was like a much higher level EMACS largely programmed in its own language. LSE made programming very easy with builtin language and procedure call skeletons as well as excellent folding. LSE was quite expensive when originally distributed (you got TPU bundled) for non-educational users, so many people missed out on it.
The beauty of VMS (Score:4, Informative)
One of the questions that comes up all the time is: How enthusiastic is our support for UNIX? Unix was written on our machines and for our machines many years ago. Today, much of UNIX being done is done on our machines. Ten percent of our VAXs are going for UNIX use. UNIX is a simple language, easy to understand, easy to get started with. It's great for students, great for somewhat casual users, and it's great for interchanging programs between different machines. And so, because of its popularity in these markets, we support it. We have good UNIX on VAX and good UNIX on PDP-11s. It is our belief, however, that serious professional users will run out of things they can do with UNIX. They'll want a real system and will end up doing VMS when they get to be serious about programming. With UNIX, if you're looking for something, you can easily and quickly check that small manual and find out that it's not there. With VMS, no matter what you look for -- it's literally a five-foot shelf of documentation -- if you look long enough it's there. That's the difference - - the beauty of UNIX is it's simple; and the beauty of VMS is that it's all there.
Ken Olsen, Chmn&CEO, DEC, 1984
Re:The beauty of VMS (Score:2)
Having done much programming on both UNIX and VMS, I can say that this is completely true. VMS is much better documented. All parts fit together better, and the documentation is very clear. And VMS has more features that UNIX is missing, such as handling asynchronous event, doing parallell async I/O on dozens of disks simultaneously, scaling to thousands of network connections, etc. If you do some types of programming, UNIX certainly feels like a toy system. (It is becoming gradually better, though.)
Unix / VMS comparison (Score:2)
OpenVMS boots. (Score:3, Informative)
They're even funding a port of GCC (Score:3, Informative)
Announcement on GNAT for ia64/OpenVMS [gnu.org] on 14Mar2002
This is great and it's the right thing to do!
Laurent
Re:If only ... (Score:4, Funny)
TOAST!
Re:If only ... (Score:2, Funny)
Re:If only ... (Score:3, Funny)
Re:If only ... (Score:2)
Re:If only ... (Score:3, Informative)
Funny? Perhaps it was missed sarcasm on my part. While we're at it, maybe we could also use an Athlon processor to keep our beer cold while we watch forthcoming "Itanium 2 powers your eBusiness" commercials?
You must have C3 or Motorola on your desktop. Tell me this: Does your processor underclock itself when it detects it's overheating due to a failed CPU fan?
Re:If only ... (Score:4, Funny)
What, your Athlon processor has disappointed you?
Re:If only ... (Score:3, Interesting)
I've seen an Athlon physically melt the solder connections of a motherboard (mine, sadly). Something must've caused the fan to seize momentarily -- then the building I was in had a fire drill -- nothing wrong at all, just a test of the system... I was away about 5 minutes. Walk back to my computer, and notice the odd smell. Turn off the power immediately; open the case... The heat from the Athlon had melted the CPU fan into the CPU heatsink. The (copper) heatsink was quite discolored due to heat. I made the mistake of brushing my hand against the heatsink after I powered the thing off... (but before noticing that it was the melted fan that smelled so awful). It took weeks for the burn to heal; the act of pulling my hand away jarred the case slightly, and a couple of the toroidal cores slid out of their holes (the solder had melted) And, what's more: There were all these surface-mount components oozing downward (with gravity).
Absolutely no overclocking or tweaking was involved.
The thing I couldn't help but think was how close the fire 'drill' had come to being an actual fire... At least everyone was out of the building... All thanks to the nice, cool-running nature of the AMD Athlon.
Interesting note: The AMD Athlon pulls ~75 W of power. The average soldering iron pulls 15-30W. Pentiums aren't much better than the Athlon... Is it any wonder that many people are using these tornado cases, or liquid cooling? That's a lot of heat to be dissipating!
Interesting note II: I replaced that first Athlon with a second one.
Interesting note III: The second Athlon died (not by heat, however) under a year later.
Interesting note IV: I have yet another Athlon, which is getting close to the 6-month mark... Here's hoping it makes it to a full year!!!
Interesting note V: My next computer is going to be a Mac.
Re:If only ... (Score:2)
My room gets pretty warm when my PC is on.
Re:64 bit architecture: illusionary performance (Score:2)
Re:64 bit architecture: illusionary performance (Score:2)
Re:64 bit architecture: illusionary performance (Score:2)
Re:64 bit architecture: illusionary performance (Score:5, Offtopic)
4GB is still a lot by any standard, but the problem is that the kernel, for example, needs to have some of its structures appear in the process's address space. Shared libraries also need to be in each process's address space, even if it's in memory only once. Even better, you need to leave room for the heap to grow as well as for user-loaded entities (like dlopen()'ed shared objects). In practice, I understand the default restriction for Linux is somewhere in the neighborhood of 1.5GB per process, though you can increase that to 3GB with some libc tweaks. There are a few kernel patches to raise that limit further to 3.5GB, but that's the absolute ceiling (and you are sacrificing what may be important things, like SysV shared memory segments).
Even now, it's not uncommon to see gaming machines with 1GB RAM or more. Even small servers will often have 4GB RAM. In a few years, the number of "high-memory" systems is only going to increase as advances in technology continue to drive down the cost of RAM and drive up the requirements of software. This is especially the case as databases become more important and commonplace in the business world. Everyone uses them now, but we can expect to see them used more often, and in more diverse places, than in the past.
There is also a hope among many of us that Intel and AMD will use this opportunity to create good chips, not just cheap ones. They have the opportunity to fix a lot of the stupid design decisions Intel made 20 years ago and put together a modern, clean system. Unfortunately, it doesn't look like this is going to happen (I don't think anyone at AMD has had an original idea in their entire lives, and Itanium is being designed by committee), but this is the hope. In short, we may not see any immediate performance gains, but in the long term, the design of the chips would enable faster improvements than we're getting with IA32s right now.
Re:64 bit architecture: illusionary performance (Score:3, Informative)
Note that this is completely transparent to the process. The OS is responsible for setting up paging. Note also that on a server you typically run a whole lot of processes anyways (e.g. a whole bunch of web server processes) so for most server applications I don't see it as a major limitation (big databases may be one problem)
Note also that theoretically that you could have system where the process could access more than 4GB of linear address space, if you set aside some region of linear address space for this purpose, and had a system call to update where this was going to. This is an ugly, ugly kludge, requires OS modification, and reminds of DOS style memory management but could be done (and would probably be cheaper than moving off of X86...)
Re:64 bit architecture: illusionary performance (Score:2)
The "64GB" options in the kernel just enable the page address extension (PAE), which is a way of addressing up to 64GB for the entire OS. So you can have 16 processes using 4GB each, and you'll be using all 64GB, but you can't have 8 processes using 8GB each.
It's not impossible to write an OS which lets each process access up to 64GB RAM, but it would be nontrivial. It would require many different segments and probably a lot of page faults. And it would be a lot of work, unless I'm missing something very obvious.
FMTYEWTK About Intel PIII 36-Bit Addressing (Score:2)
OK, here's the story:
The 386 processor lets you refer to memory with a 32-bit address. This is good and useful and lets you specify any byte in a 4 GB range. This is continued up through the latest Xeon. This is known as a flat memory model in that you only have one address for each memory byte. The alternative, used in the 8086 / 186 / 286 "real mode" and many other (mostly older) CPU designs, is segmented addresses, where each memory location is addressed by a segment and an offset within the segment. In some cases you can optimise out the segment address by assuming it is the same as your "current" segment, however that may be defined. Segmented memory addressing is seen nowadays as needlessly complex for general computing.
So as to flat addressing: when you refer to a (in our case 32-bit) memory address, this address is usually not the physical address but a virtual address, which is looked up by the CPU automatically via page tables. These page tables and other associated machinery are set up and managed by the OS.
Now, note that when Intel added 64 GB support to the Pentium III, they didn't use classic "segmented addressing", because that would be really inconvenient for application programmers (and compiler writers) who are accustomed to the assumption that you can hold a memory pointer in a single CPU register - and general-purpose registers on x86 are all 32-bit. Instead, they instituted a 36-bit address space through page tables - the OS basically sets up a new type of page table scheme, so 32-bit virtual addresses in an application can refer to anything in a 64-GB range of physical memory.
So from the application standpoint, it still has a maximum of 4 GB of address space, less a certain amount of overhead for the OS. More than 4 GB at a time would require larger than 32-bit addresses, which would be a major compatibility hurdle.
So how is it useful to have 64 GB of memory if an application still only gets 4 GB at a time? Because if you have multiple processes, they can each have their own page tables which don't necessarily overlap. Indeed, the OS always keeps page tables separate between processes - that's how memory protection works, and why you can't access another process's memory without explicit arrangement.
If your application needs more than 4 GB of memory, but you don't want to invest in a real 64-bit CPU, you basically have to break up your app into multiple processes each of which can allocate its own pool of non-overlapping memory, and have the processes communicate as necessary via shared memory, message passing, the filesystem, or similar.
The other way to do it is to have the application do its own overlay management. The idea here is that the app knows it needs to access memory not in the current page table setup, so it issues a call to the OS to "swap out" the page tables regarding a particular region of its virtual address space in favor of another swath of physical memory. I believe this alternative is not currently implemented in Linux, but it may be soon. I say this on the basis of Jeff Dike's proposed extensions to support multiple address spaces per process - he wants this for User-Mode Linux, but I think it could potentially be useful for overlays as well.
Re:64 bit architecture: illusionary performance (Score:5, Informative)
64 bit does not mean a thing.
It means something important to anybody who ever has to receive a CAT scan or a nMRI scan... VMS/VAX systems run nMRI and CAT scanners... They use 64 bit architecture during Fourier analysis...
99.99999999999999% of software today does NOT run on it
probably because 99% of software today is used for text and graphics processing; not for mission critical apps. that's kind of like saying that 99% of all driving accidents happen within 25 miles of home... well, geeze, 99% of all driving period occurs within 25 miles of home...
performance difference in mhz between 32 bit and 64 bit processors (especially in the north bridge) makes any performance gained by using 64 bit architecture negligible
I disagree with you. The difference between being able to handle 2^32 and 2^64 is worlds apart in performance. I suggest that you compare 16 bit computers, which didn't support true-color, full motion multimedia, and compare to 32 bit computers. They both support text editing; however, one supports WYSIWYG better than the other...
FYI, my day job involves running MRI scans on a VMS/VAX Gyroscan Intera workstation... This 64 bit architecture is the hottest stuff around, for somebody who works with a VMS/VAX workstation... here's why: MRI scanners work just like any other printer/scanner device, in terms of device drivers, and general operation. The difficulty is, because MRI looks at differential angular momentum of hydrogen atoms to obtain it's pictures, it's got to calculate a Fourier wave analysis on each atom it vibrates. Being able to run an algorithm with 64 bits means less data manipulation, higher resolution, faster scan times, and increased diagnostic imaging power to the medical doctors.
Anyhow, for those interested, there currently seems to be a big migration from VMS/VAX/Alpha solutions to Windows/Intel compatibility (for obvious reasons). Philips has introduced an InteraNT product into their Intera Gyroscan line, which runs the MRI scanner on a Windows NT platform, instead of the traditional VMS/VAX platform which they've been using for some time...
As usual, great tool for the server companies, crap for everyone else in the world.
This is slashdot... they cover stuff which is great for server companies, hospital radiology departments, nuclear power facilities, astronautical engineering groups, etc. etc. That's why we love it...
Re:64 bit architecture: illusionary performance (Score:2)
Guess what, 100% of Alpha VMS software runs on 64-bit hardware. Because Alpha has been 64 bit for about ten years. 64-bittness is nothing new here...
Re:64 bit architecture: illusionary performance (Score:2)
Vector math isn't called 64-bit (Score:3, Informative)
Would it help with graphics applications?
No, graphics would use vector operations, which use 64-bit vectors but are not called "64-bit" operations. A "64-bit" operation is typically defined as one that uses a 64-bit number or a 64-bit pointer, not a vector of four 16-bit numbers. Current 32-bit processors are perfectly capable of performing operations on vectors of 16-bit numbers through such instruction set extensions as 3DNow! and AltiVec.
Re:No congratulations for selling out... (Score:4, Insightful)
Innovations didn't save DEC from its stupid managment.
Well, it sux but it's not going to kill them any time soon. I wish it would kill them but USA is about Intel and MS and other mediocraties.
HP wants to be in bed with Intel. HP needs to keep OpenVMS only alive enough to avoid jilting its inherited customer base. The same is true with porting HP-UX to the Itanic. Linux on the Itanic is HP's real server "solution". They are getting into Linux clustering in a major way.
Re:No congratulations for selling out... (Score:2)
Re:No congratulations for selling out... (Score:2)
Re:But all their hard work was lost! (Score:3, Informative)
Hello!??! This is VMS we are talking about. The filesystem is not corrupted on power failure.
Re:But all their hard work was lost! (Score:2)
Too many people have been ruined by cheap PC's that have to be rebooted; I still remember the amazement I felt when I learned that mainframes (which is what OpenVMS runs on anyway), just dont' crash-- ever. It took me months to get used to the fact that you didn't have to reboot the thing-- ever, for any reason whatsoever. Entire CPU upgrades, OS upgrades, entirely new OS kernels / system builds -- nothing brought it down.
Re:open source? (Score:2)
Re:OpenVMS (Score:2)
And also it has a privilege "grant yourself any privilege
Seems just as reasonable as having to login as a member of the wheel group to be able to su to root. Better to be running around without all those privileges you could grant yourself.
Re:OpenVMS (Score:2)
And anyone who has the priv is automatically reconned to have SYSTEM.
There is a major advantage to be able to have an account where you can turn on privs when you need them but run with them turned off by default. What you do is to grant the exact privs needed by specific applications, the least privillege principle.
Of course never having used a secure O/S, you probably think UNIX is a great model with only 2 priv levels, nothing and absofuckinglutelyeverything.
Re:OpenVMS (Score:2)
VMS had Mandatory Access Controls, and information labelling but that was in Security Enhanced VMS only. The problems was getting export licenses. It was possible, but hard work. There was an experimental Orange book 'A' system (with a VM) produced but this couldn't even be sold outside the government without problems.
ACL and privileges were relatively easy to handle. There was only one manual you needed to look at and it was explained clearly there. It was quite useable by anyone who took the time to look at the docs.
Isolation between processes a problem under VMS? No, I think you may be confused there. With fine grained control over quotas, it was difficult to trash other processes. Indeed this was the main problem with VMS, building a new process was so expensive because of all the protection built around it.
As regards the subjects/objects - this really wasn't a problem. You could group your subjects into a class and grant or revoke access via class. The number of objects isn't really avoidable, because directories could be protected but directory access isn't worth a damn because most file systems allow you to open directly via node number.
RSTS/E and TSX security wasn't even slighly comparable. RSTS definitely didn't have a VM. Essentially it was just a multithreaded basic interpreter. OS 370 was certainly not up to much (starnge but IBM really didn't 'get' security' in those days) and Cray also not (it was designed as a fats number cruncher). Actually security on the Cray was a problem for many people ivolved classified/non-classified research because it was very difficult to securely partition the system.
Re:OpenVMS (Score:2)
Fortunately it does not appear that anyone takes much notice of your opinion.
VMS had very tight isolation between the processes, each of which ran in a completely separate address space.
When people start holding up VM as an example of how to structure an O/S for anything you know they are blowing smoke. VM was so disfunctional that users could not even share files without major aggravation. This led to account sharing on a massive scale at CERN.
Re:OpenVMS (Score:2)
Managing a complex system like OpenVMS does not require that you are smart, it just requires that you read the manual "Guide to OpenVMS Security".
Are you trolling or something?
Re:Just curious (Score:2)
Re:VMS on the desktop; a stable PC OS... (Score:2)
Don't bet on it. I would be surprised if this port actually works, much less is officially supported, on anything other than HP systems with special firmware. Now maybe HP could, given some vision, turn VMS into a seriously competitive OS for commodity IA-64 hardware (especially if the free hobbyist license was still available). But current management seems to show little interest in their own original technology, not just DEC/Compaq stuff like VMS and the Alpha, but the PA-RISC architecture, the HP 3000 series (a midrange platform even older than VMS), and the famous test equipment division which was spun off. They see more profit in selling printers and functioning as an OEM for Microsoft and Intel. VMS is being kept around because, for a while longer at least, they will still be able to sell enough systems at inflated prices to locked-in VMS customers to justify keeping it around. Once most of them have switched to IBM or Sun machines (or maybe even PCs running some future more robust iteration of Linux) it will probably go the way of MPE.
If they sold a version of VMS that would work on open hardware it would accelerate the erosion of the user base for their hardware, and since the idea that the number of VMS users might actually increase as a result is probably completely incomprehensible to them, I doubt they would risk this.
Re:25 years.... (Score:2)
Re:VMS trivia question (Score:2)
Re:VMS trivia question (Score:2)
I saw him just a couple of weeks ago when I was in ZK, visiting a friend for lunch.
I'm a former system manager (sysadmin to Unix folks) that worked in the VMS Development Group.
Re:Good Ol' Open VMS (Score:2)
What I would like is the distributed lock manager to run undr Linux. Then I could have a nice cluster.