[Info-vax] OpenVMS servers and clusters as a cloud service

[Kerry Main]

> -----Original Message-----
> From: Info-vax [mailto:info-vax-bounces at rbnsn.com] On Behalf Of
> Johnny Billquist via Info-vax
> Sent: January 2, 2018 5:52 AM
> To: info-vax at rbnsn.com
> Cc: Johnny Billquist <bqt at softjar.se>
> Subject: Re: [Info-vax] OpenVMS servers and clusters as a cloud service
> 
> On 2018-01-02 09:16, Richard Maher wrote:
> > On 02-Jan-18 1:45 PM, DaveFroble wrote:
> >> Richard Maher wrote:
> >>> Exactly. There is *no* common system disk save a Desired State
> image
> >>> copy. And, at least for IaaS, VMS Clusters are dead. The DLM is dead
> >>> unless you can tolerate the irony of licensing Oracle's.
> >>
> >> I don't agree.  The DLM is alive and well, and if VSI uses a
> >> recommendation I sent to them, it would have numeric range locking.
> >>
> >
> > You're missing the point :-( The DLM is/was amazing and lives on in
> VMS
> > Clusters but that pathetically limited (geographically and scalability
> > wise) architecture has been resigned to niche applications.
> >
> > If VMS wants to grow it needs World Wide Session State services and
> lock
> > managers.
> >
> > 32 byte lock value blocks are a fucking joke today!
> 
> If you want to grow to Google scale, you cannot have locking in the
> first place.
> 
>    Johnny
> 

Google scale is all about active-passive scaling horizontally models with syncing being done at the App/DB layer - not the OS layer.  Having stated this, I would be willing to bet that Google's next gen designs are no where similar to the highly distributed model they designed 10-15 years ago and that is now deployed today.

Fwiw, OpenVMS can also be deployed in an active-passive shared nothing horizontal model using large numbers of standalone servers with syncing done at the App/DB layer as well. Yes, there may be some specific optimizations missing or not as performant as they might be, but its really not a technical limitation for the OpenVMS architecture itself.

Course, a number of major OpenVMS enhancements are cooking (X86-64, new file system, new TCPIP stack, new pricing models etc.), so comparing todays OpenVMS world to the OpenVMS world that will be available in 2020 is not really much of a comparison.

Re: cluster models - Both shared disk and shared nothing compute models have pro's and con's. 

I like the analogy that compares the shared nothing model (Windows, Linux, OpenVMS) to a dragster and the shared everything model (Linux/GFS, OpenVMS, Z/OS) to a Ferrari.  In a quarter mile race on a track, the dragster will win hands down every time. In a race on normal streets, the Ferrari will win every time.

😊

Active-active shared disk cluster models using cluster communications based on technologies like RoCEv2 / InfiniBand definitely have a place in the future - especially since it is LAN / WAN networks which are by far the biggest latency in the overall solution latency in most compute models in todays world. 

Next Gen designs are all about moving data closer to the compute engines and those relying on multiple disparate severs for communication over LAN networks with all of its huge latencies (compared to tightly coupled, more centralized models) is in for some tough times ahead.

Reference:
< http://highscalability.com/blog/2015/10/12/making-the-case-for-building-scalable-stateful-services-in-t.html>
" For a long time now stateless services have been the royal road to scalability. Nearly every treatise on scalability declares statelessness as the best practices approved method for building scalable systems. A stateless architecture is easy to scale horizontally and only requires simple round-robin load balancing.

What’s not to love? Perhaps the increased latency from the roundtrips to the database. Or maybe the complexity of the caching layer required to hide database latency problems. Or even the troublesome consistency issues."

Regards,

Kerry Main
Kerry dot main at starkgaming dot com