Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!ernie.Berkeley.EDU!jas
From: jas@ernie.Berkeley.EDU (Jim Shankland)
Newsgroups: comp.databases
Subject: Re: Single Server Bottlenecks (was RE:ORACLE REL 6, INGRES REL 6)
Message-ID: <29144@ucbvax.BERKELEY.EDU>
Date: 11 May 89 18:50:30 GMT
References: <3176@tank.uchicago.edu>
Sender: usenet@ucbvax.BERKELEY.EDU
Reply-To: jas@ernie.Berkeley.EDU.UUCP (Jim Shankland)
Distribution: usa
Organization: University of California, Berkeley
Lines: 36

In article <3176@tank.uchicago.edu> cs_bob@gsbacd.uchicago.edu writes
[about bottlenecks in singler-server DBMS architectures, describing
INGRES 6.1 running under VAX/VMS on a "small interactive" system.
As a single process, the INGRES server gets inadequate CPU, since
it is competing for time-slices on an even footing with various
non-DBMS users, even though it is serving multiple users, and thus
should get a bigger slice of CPU.  Bumping the priority of the INGRES
server doesn't work, because then the non-DBMS users starve when the
server becomes CPU-bound.]

That sounds like a deficiency in the OS, rather than with the idea
of a single-server architecture (although I suppose one could argue
that a single-server architecture is inappropriate for such a badly
crippled OS).

I barely know VAX/VMS.  Are you really saying it provides no way to
grant a process a multiple of the resources (CPU, memory, etc.)
available to other processes without potentially starving the other
processes when the high-priority one goes completely CPU-bound?
Even UNIX, which has been roundly criticized at times for being
deficient in this area, can manage that with no problem:
if i "nice" a process to its highest possible (numerically lowest)
priority, and that process goes into a tight loop, other processes
will continue to get *some* CPU.

For my money, where the single-server architecture really starts
to hurt is in a multi-CPU environment.  Then you just have to
go to multiple servers to get reasonable CPU usage.  Even with multiple
servers, it may be difficult to exploit intra-query parallelism.
(Not impossible; in fact, a recent posting claimed this will soon
be in INGRES.)

Jim Shankland
jas@ernie.berkeley.edu

"Blame it on the lies that killed us, blame it on the truth that ran us down"