[Info-vax] Volatile, was: Re: yet another sys$qiow question

[Johnny Billquist]

On 2015-08-28 23:05, Bob Gezelter wrote:
> On Friday, August 28, 2015 at 5:02:43 PM UTC-4, Bob Gezelter wrote:
>> On Friday, August 28, 2015 at 4:36:34 PM UTC-4, Bob Koehler wrote:
>>> In article <55e0994f$0$17002$b1db1813$2411a48f at news.astraweb.com>, JF Mezei <jfmezei.spamnot at vaxination.ca> writes:
>>>>
>>>> And when discussing with people who deal in real time, they are all
>>>> adament that it must all be an event loop, no ASTs allowed. So while I
>>>> never had to do this, I think it is possible that some types of apps
>>>> might actually do need to poll the IOSB as part of their real time event
>>>> loop. (if it is more efficient than checking event flags).
>>>
>>>     Did tons of ASTs in real-time.  Hard real-time.  I'd never want an
>>>     event loop.
>>>
>>>     But I know a few who don't understand there systems and do want an
>>>     event loop.
>>
>> Bob,
>>
>> Most of the "event loop" type code originally comes from other systems without ASTs or similar constructs. X-Windows originates in a time when there was no standard Unix threading package.
>>
>> For the record, I have yet to see a case (in over 40 years of systems programming on RSX-11/OpenVMS) where AST delivery overhead was a significant factor. There have been cases where poor planning led to long latencies in AST state (e.g., WAIT-style directives, long computations), but those were poor designs, not a problem with AST efficiency.
>>
>> - Bob Gezelter, http://www.rlgsc.com
>
> Clarification.
>
> I DID get a call from Operations once. It seems that the AST implementation was TOO efficient (on a MV II).
>
> It seems that an AST-based process I had implemented had been observed to not accumulate ANY CPU time over a multi-hour period. The operations team was convinced the code was hung (in actuality, it was processing tens of ASTs/second). Operations calmed down when I pointed out that the DIO count was increasing on a consistent basis.

ASTs are very fine, from a resource usage point. They are also nice from 
a program flow point of view, if you use them right.

However, they do not give you predictable response times. So if that is 
what you are looking for, then ASTs are not the thing for you.
)But then again, if predictable response times interest you, then VMS is 
not the OS for you, and neither is Unix. Not even Unix with "realtime" 
extensions...)

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt at softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol