[Info-vax] Intel x86-64 Processor Design Security Vulnerability?

[Stephen Hoffman]

On 2018-01-03 18:45:16 +0000, Simon Clubley said:

> What is interesting is that people put down microkernels because of the 
> additional overheads they incur, but now monolithic kernels are likely 
> to have most, if not all, of those overheads on Intel 64-bit processors 
> without getting the benefits that come with microkernels.

it's all in the trade-offs and the compromises.  On what problems 
you're actually building your operating system to solve.  Beyond 
maintaining and increasing the revenues among the commercial operating 
system product offerings, of course.  Encryption overhead is not 
(comparatively) as performance-significant as it once was.  Some of the 
recent microkernel designs such as L4 do far better around reducing the 
effects of message-passing overhead.   But the same things that make a 
single address space kernel problematic from various perspectives are 
also the sorts of things that makes the design as effective and as 
common as it can be and is.  And it's all a moving target, too; we have 
far larger numbers of cores, mixed sorts of cores, cache design changes 
and cache size increases, far larger memory sizes, massively faster 
storage, and a whole host of other fundamental changes to computing.  
Etc.

Some light reading...
https://blog.cloudflare.com/how-expensive-is-crypto-anyway/
http://l4hq.org/docs/performance.php
http://www.microkernel.info
http://www.cs.toronto.edu/%7Edemke/469F.06/Lectures/Lecture2.pdf
http://www.cs.toronto.edu/%7Edemke/469F.06/Lectures/Lecture3.pdf
http://srl.cs.jhu.edu/courses/600.439/ExperienceMicrokernelBasedOS.pdf
https://www.kernel.org/doc/ols/2007/ols2007v1-pages-251-262.pdf
http://sture.ch/vms/Usenix_VMS-on-Mach.pdf

The L4 latencies on now-ancient Itanium 2 configurations and the 
even-more-ancient Alpha configurations are surprisingly fast, too.

I don't see VSI doing much in the way of innovation with the OpenVMS 
kernel, though.  They've a port to finish and products to establish in 
the market, and a customer base and a staff and revenues to build.  
Maybe call back in five or ten years, or whenever VSI can acquire the 
staff and the cycles and the revenues to allow them to really figure 
out what they want to do "next".




-- 
Pure Personal Opinion | HoffmanLabs LLC