Path: utzoo!attcan!uunet!portal!cup.portal.com!doug-merritt From: doug-merritt@cup.portal.com Newsgroups: comp.sys.amiga Subject: Re: Amiga UNIX Message-ID: <6469@cup.portal.com> Date: 13 Jun 88 15:16:00 GMT References: <211@laic.UUCP> <3663@cbmvax.UUCP> <1872@sugar.UUCP> <134@ssdis Organization: The Portal System (TM) Lines: 81 XPortal-User-Id: 1.1001.4407 Peter da Silva wrote: >Comparing a PDP-11 running Berkeley UNIX (3BSD) with a VAX running Berkeley >UNIX (4BSD). The PDP-11 gave a lot better response time, and supported more >users, with less real memory than the VAX. And *this* was with an overlaid >kernel in the '11! This was true with earlier versions, too. Even *Version 6* Unix (as modified at Berkeley) could support 40 users on a PDP-11/70 with not-too- unreasonable response time. Or 25 users with *fast* response time. I saw 60 users on it many times (students at the end of the quarter), but the response time got ridiculous by that point. Still, I've never seen a VAX 11/780 running 40 users. I have seen it running 20. Slowly. To be fair, however, this is not altogether because a swapping Unix is faster than a paging Unix..there are several factors. Probably the biggest is that, when you've got plenty of memory, you use "too much" memory. People start writing stuff that's much bigger than it really needs to be, and that kills you when it needs to be paged. Look at the size of GNU software for a really good set of examples of this phenomena. Lots of features, but also remarkably large. Another factor is that the PDP 11 has a more efficient instruction set. The same program compiles smaller on the PDP 11 than on the VAX (you could say the PDP 11 is more RISC-like than the VAX), and that means there's less of it to swap on the 11 than to page on the VAX. And a final factor: I haven't looked at benchmarks, but some instructions at least run faster on the PDP than on the VAX...purely from an architectural point of view I would expect the PDP 11/70 to win most non-floating point benchmarks against the VAX 11/780. David Albrecht wrote: > Apples and Oranges. It has been a long time, but last time I used a PDP 11 > a process maxed out a 128K. Supporting many users each in a 128K > hunk is a lot easier than a bunch of users running in arbitrary size areas. > You can almost 'swap' a 128K hunk and get reasonable response for a fair > set of users. Kind of limits you to wenie processes though don't it? Yeah, the address space supported 64K of text and 64K of data. For some purposes that 128K is "wenie" indeed (e.g Franz Lisp, Macsyma, etc). On the other hand, it's *plenty* for the usual types of processes. For instance, it was in that environment that C shell and VI were developed. Never mind whether you like those utilities; the point is that they were large and complex, and their functionality has not increased very much since they moved to larger address spaces. At the time, the source code to each of these programs was larger than that of the PDP 11 Unix kernel. Both of these programs took up essentially *all* of the 64K of text space available to them, which I will grant limited the features Bill *could* add to them (as he lamented). But they are beyond the "wenie" stage even so. You can do an awful lot in 128K. Granted it's a real pain (as the 80286 is these days) when you *are* working on some huge project, but the other 99% of the software will fit just fine. (Hmmm...I'm really playing devil's advocate here. Usually I take the "we need more memory!" side of it.) Also, for the reasons you point out, it is often true that swapping *is* faster than paging. Peter said: >As a side comment, the 128K limit *is* the reason paging wasn't implemented on >the '11... even though the hardware supported it. As I recall there was a problem also with restarting certain instructions (i.e. you couldn't restart them after a page fault in their middle). This can sometimes be worked around via smart memory layout, but as you way, for 128K it's not really worth it. It's easier to tune paging to be fast anyway. All of the above is aimed simply getting some perspective on the issues. The PDP 11 was actually a better machine overall than the VAX. Problem is that address spaces larger than 128K are essential for *some* applications, and paging is very convenient for many of those same applications. Hence the VAX. It would've been better a better machine, though, if they'd just increased the size of the address space, rather than adding all those sexy but inefficient instructions. Doug -- Doug Merritt ucbvax!sun.com!cup.portal.com!doug-merritt or ucbvax!eris!doug (doug@eris.berkeley.edu) or ucbvax!unisoft!certes!doug