Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!sdd.hp.com!hplabs!hpda!hpcuhc!dhepner
From: dhepner@hpcuhc.cup.hp.com (Dan Hepner)
Newsgroups: comp.databases
Subject: Re: Need information on DBMS architecture
Message-ID: <2060013@hpcuhc.cup.hp.com>
Date: 17 Apr 91 21:26:25 GMT
References: <5274@goanna.cs.rmit.oz.au>
Organization: Hewlett Packard, Cupertino
Lines: 25

From: hao@wren.cs.rmit.OZ.AU (Hao Hong Nguyen)

>As my knowledge, on an Unix system, the kernel and the
>front-end can communicate through socket, pipe or shared memory.
>Is there any other DBMS architecture ? and how how different compenents of a
>DBMS communicate to each other ?
>
>The reason I ask for this is that my group is implementing a research
>DBMS and find out that the communication overhead is quite
>siginificant.

The bulk of the cost of a shared memory front-end to back-end IPC is 
in the context switch, which in unoptimized implementations includes 
going through a general purpose scheduler algorithm, which is expensive
in this context.

HP-UX, and maybe others, offers a system call to two-process DBMS 
implementations which passes a single pointer into shared memory, 
and in as cheap a way as we could do, switches immediately to the 
partner process.

Using this, we find that communication overhead is not all that
significant.

Dan Hepner