[Info-vax] x86 Server Reliability (was: Re: BASIC compiler in the hobbyist distribution)

[Stephen Hoffman]

On 2015-05-27 17:27:42 +0000, David Froble said:

> seasoned_geek wrote:
>> On Tuesday, May 26, 2015 at 5:53:40 PM UTC-5, David Froble wrote:
>>> So, where do I find the IDEs that work with VAX/DEC Basic?
>> ...
>> Some kind of NetBeans plugin
>> http://compgroups.net/comp.os.vms/netbeans-basic-plugin/523219

>From what I can tell, NetBeans isn't being particularly updated on 
OpenVMS.  Whether VSI or a third-party changes that?

>> There was supposedly an Eclipse plugin at one point but they might only 
>> support VAX COBOL now.

eCube has an Eclipse-based IDE available for OpenVMS.

>> I do not remember the full setup, but you could configure LSE to do 
>> builds directing output to another buffer.

LSE has COMPILE /REVIEW, which works with BASIC and most of the common 
OpenVMS languages, but is rather limited in comparison with other 
platforms and tools.


>> 
>>> Frankly, I don't see where geek's argument is coming from.  HW is HW, 
>>> it does what it's told to do.  Unless I'm entirely clueless (as some 
>>> have claimed) I don't see where x86 (other than being a poorly 
>>> implemented but very well developed CPU) is any different from others.
>> 
>> It's not just the OS which gives up times measured in decades, it's the 
>> hardware. Once migrated to x86 up times will be measured in weeks 
>> (especially with blades instead of rack mounts) or at best months. Yes, 
>> each of you can find ONE x86 system somewhere which was in a perfect 
>> environment and defied all odds. Those are statistical anomalies, not 
>> the norm.
>> 
>> Vendors are finally starting to talk about the failure rate. Page 17 in 
>> this document is interesting.
>> 
>> http://s3.amazonaws.com/isby/lenovopartnernetwork.com/upload/1/docs/1p2-2p-portfolio-positioning-client-pres.pdf 
>> 
>> 
>> I say it is interesting because they scope it to 4 hour failures or 
>> more. No numbers provided about those "random reboots". What is even 
>> more interesting is that no x86 vendor is touting up-times. Marketing 
>> has ceased pushing quality. They now are defining it as being the 
>> lowest percentage of 4+ hour failures instead of 4+ years of contiguous 
>> up time.

That graph shows that around 1% of the Lenovo System x boxes fall as 
low as ~99.95% uptime per year.

Which means that 99% of the System x boxes have less than four hours 
downtime a year.


>> Yes, you can point to special radiation/EMP/etc hardened versions of 
>> x86 made for some satellites. So what? It's the cheap pieces of doo-doo 
>> going into the racks and blades.

So around 1% of the Lenovo System x boxes drop as low as that 99.95% 
uptime, or quite possibly to worse uptime, and the other 99% of the 
System x boxes have better uptime; less than four hours downtime per 
year.  That's presumably across all of the System x boxes sold, and not 
just the ones with configurations for customers that intentionally 
target uptime, too.

What "better uptime" might be here is not presented by the data; the 
cut-over here is that ~4 hour mark.

>From the ITIC 2014 server survey "However, an in-depth analysis of the 
results indicate that the prolonged unplanned downtime of over four 
hours has less to do with the inherent reliability or instability of 
the Dell PowerEdge and HP ProLiant servers and is more indicative of 
end user behavioral patterns. Some 60% of Dell users and 53% of HP 
ProLiant users said they kept their servers for four, five or even six 
years or longer without upgrading/retrofitting or right-sizing the 
servers to accommodate more compute intensive workloads. By contrast, 
only 21% of IBM System x or IBM Power Systems users retained their 
servers for four or more years without upgrading or retrofitting them."

Further along "Human Error is the issue as identified by 44% of survey 
respondents that negatively impacts network reliability, surpassing 
inherent flaws in either server hardware or the server operating 
system. Bugs or flaws in the server OS came in second with 33% of those 
polled stating it undercuts overall reliability, while 30% of survey 
participants said the fact that their IT departments were overworked 
and understaffed undercut reliability."

So if I want really low downtime, head for IBM System z, or for NSK.  
Which was what most folks probably expected, before reading that 
survey.   Both of these high-end platforms are not cheap, and both of 
which involve porting OpenVMS code to another platform, which won't be 
cheap and won't be easy.  The survey also implies that if I upgrade my 
servers more often than twice a decade, and — given this is OpenVMS — 
if I cluster, I might do pretty well with x86-64 hardware.  Or am I 
misreading the data?

> When DEC listened to it's customers, it did well.  When DEC started 
> telling it's customers what to do, they didn't do so well.

Ayup. Getting priced out of the market by higher-volume products — 
which means that those products can sell for less — didn't help with 
that quest, either.

But I'd wonder if more than 99% of the existing OpenVMS boxes had more 
or less than the cited 4 hours downtime per year, too.  I obviously 
don't have that data.



-- 
Pure Personal Opinion | HoffmanLabs LLC