[Info-vax] "The Machine"

[IanD]

On Tuesday, September 1, 2015 at 1:08:44 AM UTC+10, Stephen Hoffman wrote:
> On 2015-08-31 14:33:15 +0000, <removed> said:
> 
> > On Sunday, August 30, 2015 at 4:28:31 PM UTC-4, IanD wrote:
> > 
> > [snip]
> >> 
> >> While we are at it, why limit VMS clusters to VMS? The clustering 
> >> component once ported to x86-64 could be potentially licensed to run on 
> >> other OS's, in time.>[snip]
> > 
> > It was my understanding that everything about VMS clustering that could 
> > be patented was patented in the late 1980s.  Likewise IBM's parallel 
> > sysplex, and the Oracle code for shared locking.  I believe that 
> > patents from those years expire 20 years from filing date or 17 years 
> > from issue date; if these haven't already expired, they will soon.  
> > Licensing cluster code seems an unlikely source of long-term, possibly 
> > even of short-term, revenue.
> 
> Ignoring the legal aspects and whether there are any software patents 
> lurking here...
> 
> The OpenVMS clustering code was ported to C, and was later open-sourced 
> by Compaq as the cluster single system image (SSI) bits, and went... 
> nowhere.
> 
> Some URLs from that era...
> 
> <http://www.prnewswire.com/news-releases/compaq-and-osdn-create-clustering-foundry-for-open-source-community-74216477.html> 
> 
> <http://www.openssi.org/ssi-intro.pdf> (lots of links in this PDF)
> <http://compaq.com/alphaserver/download/Giga_IdeaByte_TruCluster_vs_Sun_Cluster.pdf> 
> 
> <http://opencf.org/home.html>
> 

I did start looking into OpenSSI some weeks ago but when I saw the last update date I stopped bothering

In fact, I stopped bothering looking at a number of things because they were outdated as far as VMS was concerned :-(

> Ignoring that OpenVMS-style clustering doesn't scale up to where modern 
> clusters are operating now, OpenVMS-style clustering is clearly not 
> interesting to the market.
> 

True

However, there is also a vast lack of knowledge out there about OpenVMS clustering too

I work with so called architects know nothing about OpenVMS clusters, nothing at all actually

Oracle RAC is about as far as their knowledge stretches when it comes to clustering. Application clusters is what they think a cluster is

Mind you, where I work is telco centric. I have friends working in the academic side of things and clusters to them are a totally different beast, geared towards high speed collaborative computation. Redundancy and availability of a compute engine is down the list of importance when compared to raw data crunching ability. Horses for courses

> Y'all might ponder why this is, and whether that perception can be changed.
> 

Not unless it's from the ground up. This is why I said VMS must redefine what clustering means. I think a name change, to something like 'Reliable Cluster' (groan, a little too cliche I know) or something else. But yes, a name change alone is not enough, a technological revamp must go hand in hand also and it must bring real measured customer value

When I talk with Oracle RAC folk or the architects where I work, their eye's gloss over as soon as you mention the word 'cluster' because they immediate flip to their defined brain image of what cluster means and so unless you can differentiate a VMS cluster in their minds, forget it. The only times they have to deal with a VMS cluster is when the systems move DC's and they immediately want to try and force it onto VMware and when they discover it's not Oracle or linux or windows, it's put into the 'exceptions systems' camp

> Ponder what different or alternative approaches might be used in its stead.
> 
> Then go see what software is actually being used out there.   What are 
> its advantages and disadvantages.   Get some experience with it.
> 
> Then figure out whether or how you might market OpenVMS and its 
> clustering against that.  If you can.
> 

I rub shoulders with the Oracle folk all the time. Oracle financials, Sibel etc rule the roost here

Reliable clusters I think still have a place though

Breaking up VMS clusters so that you can zone them might be another idea

So a cluster zone might consist of what we know as the traditional VMS cluster, with reliable replication and data consistency.

Another zone might be a compute zone where you drop the need for a synchronized cluster and allow just the computational power of the systems to run just as one large distributed cpu/gpu compute engine

Another could be a disk cluster where it's just distributed storage structured for high i/o writes as in data capture rather than for interrogating etc.  

In other words splitting up a cluster functionality so one can cherry-pick what bit's of clustering you really need/want and using those bits instead of a one size fits all

The biggest driver behind all of this where I work is cost cost cost. It's how VMware grew, they tapped the need to consolidate all those machines onto a few, which saved on cost and power, which has risen vastly in cost

Rather than suffer long periods of downtime, this place will build a whole new system at huge expense, another reason why VMware get's such a look in. For VMS to tap this market, VMS clusters need to grow up, processes need to become unstoppable and be able to be moved around a cluster dynamically. Not even VMware can do this at a process level yet

> Ponder too what might be done around the ginormous price of clustering 
> on OpenVMS, and the not-trivial costs of OpenVMS itself.
> 

I can only say that the folks I've spoken too find this part the most disappointing of all with VMS licensing - just too dam expensive

Once upon a time when VMS clustering was the ants pants so to speak one could charge a premium, but now? Not a chance. 

Claw that cost back some other way but no project will adopt VMS if the set up costs are high (I'm ignoring that little applications that people want run on it for now, that's another whole area of pain)

> Further ponder what can be done to simplify the user interface for 
> OpenVMS-style clustering to better meet current user expectations, as 
> it's positively baroque to configure it -- ill-documented and 
> manually-performed shared file relocations and manual volume mounts and 
> logical name redirections and host-based shadowing configurations and 
> all the rest.  A whole lot of reading, to get this "right"...
> 

Can't disagree with this at all

It's the whole spectrum of management of VMS IMO. From clustering to networking to user management to storage. No single management interface, no interface that hides the real complexities of what you are managing, all make for a disjointed offering that is difficult to manage

It's sort of why I mentioned in another place the concept that the VMS manager must eventually go...I'm sure that caused some to choke on their morning coffee but where I work, they do not want to spend a dime on management of systems. If you start with that concept, that the system manager must go, then you can start developing a framework around making that happen and you can start working on unifying all the tolls and procedures needed to manage VMS. Start with a decent scripting language light also be a good idea ;-)

Companies who supply workplace automation are making inroads with their automation systems where I work. They handle not just common events but also workarounds and through machine learning they are also doing break fixes of unknown situations as well (simple ones). They have their eye already on configuration changes and predictive environment adjustments. Systems managers are a dying breed

VMS cannot even being to participate in these areas of workplace automation, another reason it's been earmarked as a non-compliant architecture in my workplace :-(

> 
> -- 
> Pure Personal Opinion | HoffmanLabs LLC

For sure, VMS is showing it's age and there is more than a ton of work to do on it to bring it up to speed but at least it's being reworked of sorts to get it onto x86-64. It's not dead yet Jim (if you a Trekkie fan)

How much scope VSI will have to modernise it during this transition will be anyone's guess. How much computing will advance in the next 2+ years is also anyone's guess. How many businesses out there will wait or will move away from VMS is also anyone's guess. Lots of unknowns. The only known is that it is being worked upon, for now at least

I was talking with someone just last week and they describe themselves as a dinosaur working on dinosaur systems and they still get so much work they can pick and choose the jobs. They do no advertising, it's all word of mouth only and they get a stack of work coming to them for things like retrieving old data off tapes on systems that have been shutdown for years to helping port old DOS 3 based systems that are still out there running. I was astounded how many places around that have these old systems lurking in the nether-regions of workplaces still churning away. The pyramid of computing technology is just that a pyramid. Lots of places still exist at the bottom of the computing technology pyramid

It's very easy to read new academic papers on the latest and greatest (Yes, I have acm membership and love reading 'new stuff' and it causes me to lament VMS and the state of disrepair that it is in) but the vast majority of mere mortals are still using basic computing technology in their workplaces. 

Even the data warehouse systems where I work do not have the need for something like the higher scale abilities of advanced linux clusters. NoSQL db's have no even begun to make inroads either, Oracle is still good enough at present for what they need and want, although I suspect that might change in the next 2 years though as data mining and process mining ramp up and businesses look to squeeze out money from their customers who are reducing their customer spend.

It's also true, telco is no longer an environment of open ended revenue streams and the whole industry is undergoing change are are having to do more with less - VMS needs to be able to tap this opportunity if it's going to get anywhere

There is probably still a market for VMS, but certainly not in it's current form. It's still one of the most robust systems in my workplace in terms of how many people it servers versus how many are employed to look after it versus how much up-time it manages in keeping it's head above water

Sorry, you started this thread on The Machine and I have sidetracked it somewhat. A lot of my posts are more empirical knowledge for the discussion of ideas rather me trying to assert fact, so please excuse me if my posts come across as anything else