[Info-vax] VSI: "Official 8.4-1H1 Launch"

[johnwallace4 at yahoo.co.uk]

On Saturday, 6 June 2015 20:00:54 UTC+1, JF Mezei  wrote:
> >> I wonder what kind of jitter/latency this will cause for real time
> >> applications ?
> 
> 
> When working on the early boot sequence/stages, I doubt that performance
> really matters.  And if most operating systems can run just fin on a VM,
> it means that the VM gives it adequate CPY time slots.
> 
> Also, remember that VMS ran on the All Mighty Microvax II, where
> performance wasn't measure in instructions per second, but seconds per
> instruction :-)  Pretty sure a VM can offer better performance overall
> than an MV II.

In throughput terms the HYPErvisors are allegedly adequate (KVM's 
conceptually not unlike a HYPErvisor), with caveats (eg below).

In terms of interrupt response or process context switch latency or
even just normal IO latency, which may well matter to RT applications?
You've got something more substantial than assumptions? 

I haven't played with KVM but my personal experience with VMware's ESX
HYPErvisor left me unwilling to consider it for even 'soft' realtime
stuff. Other HYPErvisors are available but were not tried.

For example, ESX seemed to have a "timesharing" effect with disk IO. 
The achievable IO rate per virtual machine seemed to be capped at
real-IO-max divided by the number of virtual machines active, *even
if the other virtual machines weren't doing any disk IO*. So if your
disk could do 100MB/s, and you've got two virtual machines active,
neither of them will see more than 50MB/s. No explanation was
available.

In your typical datacentre, no one would notice, and if they did,
they'd not care.

Proper realtime doesn't work like that. Proper computers don't
actually work like that either.