l4-hurd
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Linus replies. Re: Computer: Can We Make Operating Systems Reliable


From: Jonathan S. Shapiro
Subject: Re: Linus replies. Re: Computer: Can We Make Operating Systems Reliable and Secure?// SAS
Date: Thu, 11 May 2006 21:34:47 -0400

On Thu, 2006-05-11 at 23:27 +0200, Bas Wijnen wrote:
> Yes, these things can be enforced with a virtual machine.  But I don't
> actually see why a virtual machine would be better than a real machine,
> considering the performance penalty a virtual machine gives.  I have no
> numbers, but I don't think it's comparable to the performance you lose because
> you can't share more flexible than per page.

There are two separate things here, and it is good not to confuse them.

The cost of a safe programming language, compiled in the traditional
way, is between 5% and 10%.

The cost of a language virtual machine **for long-running programs** is
about 100% (i.e. 2x). The cost of a language virtual machine for
short-running programs is nearly infinite.

One factoid from my IBM days:

There was a time when the idea of running java inside a database system
seemed attractive. Then they ran some numbers.

A typical transaction requires 150,000 instructions for the database
server to perform.

A typical Java startup requires 1,500,000 bytecodes (about 10
instructions each, so 15,000,000 instructions) **before the first
bytecode of the main procedure is executed**.

Conclusion: adding server-side java to databases is a great way to
convince people that they need faster hardware!


shap





reply via email to

[Prev in Thread] Current Thread [Next in Thread]