polder@cs.vu.nl (Polderman Paul) (05/20/88)
Benchmarks show that a 68020 is at least as fast as a VAX-11. But if I start e.g. 6 C-compilers on a UNIX-system running on a 68020, the performance collapses completely (even when no swapping is necessary), where on a VAX the system-performance stays reasonable. I wonder why this is so, when the CPU-speeds are roughly the same. I asked several people, but none of them seems to be able to give me a clear answer. -- Paul Polderman (polder@cs.vu.nl)
guy@gorodish.Sun.COM (Guy Harris) (05/24/88)
> Benchmarks show that a 68020 is at least as fast as a VAX-11. > But if I start e.g. 6 C-compilers on a UNIX-system running on a 68020, > the performance collapses completely (even when no swapping is necessary), > where on a VAX the system-performance stays reasonable. > I wonder why this is so, when the CPU-speeds are roughly the same. I don't see that you can make a generic statement about "UNIX systems running on a 68020" here. (For that matter, you can't make generic statements about "a VAX-11; I presume you have some particular VAX, such as a 780, in mind.) 68020 UNIX machines may share the same CPU, but may run at different clock speeds, or have different memory management units, or.... Which 68020 machine did you try this on? It is conceivable that if you start lots of processes running on a Sun-3, for example, the performance may degrade because the Sun-3 MMU has only 8 hardware contexts, and the context switching overhead may increase substantially when you have more than 7 or 8 processes running simultaneously because the kernel actually has to load and unload translations from the MMU rather than just switching the context number.
joel@intelisc.UUCP (Joel Clark) (05/25/88)
In article <54190@sun.uucp> guy@gorodish.Sun.COM (Guy Harris) writes: >> Benchmarks show that a 68020 is at least as fast as a VAX-11. >> But if I start e.g. 6 C-compilers on a UNIX-system running on a 68020, >> the performance collapses completely (even when no swapping is necessary), >> where on a VAX the system-performance stays reasonable. >> I wonder why this is so, when the CPU-speeds are roughly the same. > >I don't see that you can make a generic statement about "UNIX systems running >on a 68020" here. (For that matter, you can't make generic statements about "a >VAX-11; I presume you have some particular VAX, such as a 780, in mind.) 68020 >UNIX machines may share the same CPU, but may run at different clock speeds, or >have different memory management units, or.... > I once had the pleasure of porting a compiler to about 20 different vendors Unix 68020 boxes. The porting changes were limited to different spellings for many of the system DEFINEs and different signal handling. The resultant compile time speed varied from 7 lines a minute to 3500 lines a minute. When the machine that produced 7 lines a minute had brk() and sbrk() rewritten it produced 100 lines a minute. So there is a great variance between 680X0 boxes. However when any of these boxes were compared to a Vax 11/750 or Vax 11/780, as the number of compiles executing increased the relative speed degradation (per box) was always greater on the 680X0's. I believe the Vax is much better design for 30 to 200 users then any 680X0 box I've seen. Joel Clark Intel Scientific Computers joel@intelisc.uucp Beaverton OR. 97201 {tektronix}!ogcvax!intelisc!joel (503) 629-7732
greg@xios.XIOS.UUCP (Greg Franks) (05/31/88)
In article <1273@ark.cs.vu.nl> polder@cs.vu.nl (Paul Polderman) writes: >Benchmarks show that a 68020 is at least as fast as a VAX-11. >But if I start e.g. 6 C-compilers on a UNIX-system running on a 68020, >the performance collapses completely (even when no swapping is necessary), >where on a VAX the system-performance stays reasonable. >I wonder why this is so, when the CPU-speeds are roughly the same. > I'll bet that the I/O system has a higher throughput and is a lot faster on a VAX than on your 68020 system. Another possibility is context switching: the VAX may be able to support a higher number of runnable processes (eg, doesn't have to reload page tables as often etc). -- Greg Franks XIOS Systems Corporation, 1600 Carling Avenue, utzoo!dciem!nrcaer!xios!greg Ottawa, Ontario, Canada, K1Z 8R8. (613) 725-5411. "Those who stand in the middle of the road get hit by trucks coming from both directions." Evelyn C. Leeper.