[Info-vax] Accuweather new contract

[Jan-Erik Soderholm]

Kerry Main skrev den 2015-03-30 16:03:
>> -----Original Message-----
>> From: Info-vax [mailto:info-vax-bounces at info-vax.com] On Behalf Of
>> Jan-Erik Soderholm
>> Sent: 30-Mar-15 7:32 AM
>> To: info-vax at info-vax.com
>> Subject: Re: [New Info-vax] Accuweather new contract
>>
>> Kerry Main skrev den 2015-03-30 05:25:
>>>> -----Original Message-----
>>>> From: Info-vax [mailto:info-vax-bounces at info-vax.com] On Behalf Of
>>>> Craig A. Berry
>>>> Sent: 29-Mar-15 9:13 PM
>>>> To: info-vax at info-vax.com
>>>> Subject: Re: [New Info-vax] Accuweather new contract
>>>>
>>>> On 3/29/15 6:03 PM, Kerry Main wrote:
>>>>>> -----Original Message-----
>>>>>> From: Info-vax [mailto:info-vax-bounces at info-vax.com] On Behalf
>> Of
>>>>>> johnson.eric at gmail.com
>>>>>> Sent: 29-Mar-15 4:49 PM
>>>>>> To: info-vax at info-vax.com
>>>>>> Subject: Re: [New Info-vax] Accuweather new contract
>>>>
>>>>>> My basic mode of comparison was... how long does it take to send
>> a
>>>>>> single
>>>>>> 80 byte UDP packet. I measured that by doing that in a tight loop. It
>>>> was
>>>>>> pretty
>>>>>> easy to write a portable version that ran on both Linux and VMS.
>> For
>>>>>> grins
>>>>>> I also write a version that used the QIO$ interface. That improved
>>>> things
>>>>>> a little
>>>>>> bit, but Linux still won by a wide margin.
>>>>>>
>>>>>> EJ
>>>>>
>>>>> Not saying that Linux might be marginally faster, but usually when
>>>> there Is
>>>>> a "wide margin", there is usually something else under pinning the
>>>> results.
>>>>>
>>>>> When network results like this are seen, I usually think of the mis-
>>>> matched
>>>>> full duplex auto-config issue with older network gear, NIC's and
>> drivers.
>>>>
>>>> Kerry, please read more details of Eric's careful analysis in the
>>>> archives of this group before posting such nonsense.
>>>
>>> As I recall, Eric's previous analysis showed OpenVMS network numbers
>>> to be approx. half those of Linux which would be exactly what would
>> be
>>> he case if the autoconfig issue were present when he did the testing.
>>
>> An missconfigured half/full duplex will usualy give something
>> of 1:100 to 1:1000 diference. More or less a full halt.
>>
>> The network is more or less useless apart from small packages.
>>
>> Jan-Erik.
>
> No - that is not correct.
>
> Symptoms depend on how much the network is used and the speed
> of the NIC.
>
> I have seen this issue a number of times and the symptoms are:
> - no errors at system or in error log

Yes.

> - everything works, but network transfers are very slow
>

Transfers such as small VT-terminal sessions usualy
works just fine. FTP (in one direction usually) more
or less halts. So it is not *all* network transfers.

Now, I do *NOT* think that the measured difference
as reported has anything at all to do with full/half
duplex, particualry if the difference was as small as
an 1:2 differance. Using a missconfigured interface
you wpould have seen a 1:100 or larger difference.

That "other" system was simply twice as fast as the
VMS system it was compared with. Simple as that...





> What would happen sometime is everything is working well, system
> gets rebooted for some reason, when it comes back up, everything
> works, but complaints about network performance start coming in.
>
> A reboot may even fix the issue (depending of course on the boot
> / network hand shaking).
>
> Regards,
>
> Kerry Main
> Back to the Future IT Inc.
>   .. Learning from the past to plan the future
>
> Kerry dot main at backtothefutureit dot com
>