Mailing List Archive

On Sat, 25 May 1996, Jake Khuon wrote:

> The RS2 machine at MAE-West is once again up. The second machine should be
> up as well sometime next week. There is still plenty of cleanup to be done
> and disruptions of service may occur although I will do my best to make sure
> that does not happen.

Does that mean you are replacing one of the Suns with an Alpha? I'm not
suggesting there is anything wrong with Sun equipment, just that it is odd
to see a mission critical function being performed by two identical
machines when you have a wide choice of equipment to use that will do the
job.

Michael Dillon ISP & Internet Consulting
Memra Software Inc. Fax: +1-604-546-3049
http://www.memra.com E-mail: michael@memra.com

- - - - - - - - - - - - - - - - -

### On Sat, 25 May 1996 08:00:53 -0700 (PDT), Michael Dillon
### <michael@memra.com> wrote to Jake Khuon <khuon@Merit.Net> concerning
### "Re: MAE-West is up again":

MD> Does that mean you are replacing one of the Suns with an Alpha?

Personally, I wouldn't mind doing this. However, the Digital Unix port of
RSd was not complete. Ramesh Govindan of ISI and myself spent several hours
in an attempt to get it running but it seems there were some deeply rooted
porting concerns that we didn't have time to find or address last night.


--
/*===================[ Jake Khuon <khuon@Merit.Net> ]======================+
| Systems Research Programmer, IE Group /| /|[.~|)|~|~ N E T W O R K |
| VOX: (313) 763-4907 FAX: (313) 747-3185 / |/ |[_|\| | Incorporated |
+==[. Suite C2122, Bldg. 1 4251 Plymouth Rd. Ann Arbor, MI 48105-2785 ]==*/
- - - - - - - - - - - - - - - - -

Just a side on the the Sun, Alpha Unix port discussion....

As most of you probally know, I have the pleasure of 'maintaining'
the linux benchmarks for a couple of years. We have *not* seen
evidence, to date, that the UNIX port(s) to Alpha is 'so hot', and
I'm not just speaking in terms of linux here, because people
send Suns, Alpha/Foo, HP-UX, SGI, benchmarks as well.

The performance numbers for the Pentium Pro have been coming in
(http://www.silkroad.com/linux-bm.html) and the results are
very exciting. UNIX ports to Intel architecuture are much more
mature than Alphas ports (goes without saying) and the P Pro
numbers are comparable to Alpha.

Don't, however, take my word for it.... Check out the benchmark
collection and feel free to run the test suite (also on the
server) on 'Your Platform of Choice with GCC Compiler'.

As most here know and understand, bare metal CPU Spec numbers do
not necessarily translate to a fast kernel or networking code, etc.

Cheers,

Tim



- - - - - - - - - - - - - - - - -

Tim,

Your message addressed platform performance, not network operations.
My reply is mostly just about network operations.

> Just a side on the the Sun, Alpha Unix port discussion....
> [...]
> The performance numbers for the Pentium Pro have been coming in
> (http://www.silkroad.com/linux-bm.html) and the results are
> very exciting. UNIX ports to Intel architecuture are much more
> mature than Alphas ports (goes without saying) and the P Pro
> numbers are comparable to Alpha.

Route servers don't need to have high performance hardware. A 33MHz 486,
or god help us all, a Sun-3/60 or MicroVAX III would all have enough
horsepower to run RSd.

The same is true of root name servers.

Only three things matter:

1. memory. lots of it (256MB or so, with room to grow). ECC preferred.

2. support. two sun techs were still at Ames last night when I left at 3AM.
neither one of them knew as much about their hardware as most Linux users
know about Intel's, but they got the job done Even Without Any Wizards.

3. port-able. RSd makes some assumptions about sizeof(void*)==sizeof(int)
which aren't valid for the Alpha, else RS1 would have come up first.
this kind of bad code is quite common in our community, but on most
systems it doesn't hurt anything.

> As most here know and understand, bare metal CPU Spec numbers do
> not necessarily translate to a fast kernel or networking code, etc.

I have built and sold more than 50 BSD/OS servers in the last few years, and
I am a _very_strong_ proponent of the approach. My own root name server runs
on a P5-120 BSD/OS machine and I'm quite pleased with it. But if I need to
put a server far away where non-wizard field service technicians have to be
able to fix it, I pick a vendor like DEC or SGI or HP or Sun.

Performance, kernel performance, context switching speed, file system speed,
network performance, and all the rest are just side issues. For a WWW farm,
those things matter. For a Route Server or a root name server, they don't.

(All that said, the hardware behind the current RS1.MAE-WEST.RA.NET is the
latest and greatest Alpha, and it runs about 100% faster than a P6-200. Now
if we could just find the portability ickies in RSd, we'd be finished.)

Paul
- - - - - - - - - - - - - - - - -

Paul writes:

>
> Your message addressed platform performance, not network operations.
> My reply is mostly just about network operations.
>

To be sure, you are correct. My subliminal message was more-than-likely
something like this:

"There is very little justification for being single threaded in
network operations in today's very inexpensive, high performance,
UN*X world.' (IYKWIM)

So, the issue, does not really appear to be "Router Servers: Are
They A Good Thing, Yes or No"? as implied in a much earlier post,
but more an architecutural issue, single threaded configurations,
and so forth.

Even 'experimental, non-operational' (whatever that really means in
the real world.... use at your own risk) services such as RA could
easily and cheaply have some redundancy built in at a very moderate
cost in todays world.


Back to Saturday, Best Regards,

Tim

Side note to Gordon Cook based on his contribution earlier:

Gordon, just a little barb to you, please do not take it personally,
but.. "is your current trip to Russia also funded by the NSF similar to
your previous trip"? ( inquiring minds want to know ;-)

Whoops! There I go again, standing in the corner, just losing my
religion.... Oh well, I said too much, I haven't said enought....

Micheal Stipe, REM.






- - - - - - - - - - - - - - - - -

tim asks who is paying for my trip to Russia. How about me for
incidentals and a total of $13.50 for tax on a delta frequent flyer
ticket? NSF has
paid for NONE of my trips. In Sept october 1994 NATO paid for a $620
airplane ticket and I paid $120 to extend it and for all other expenses.

**********************************************************************
The COOK Report on Internet Individ. hard copy $150
431 Greenway Ave, Ewing, NJ 08618 USA Small Corp & Gov't $200
(609) 882-2572 phone and fax Corporate $350
Internet: cook@cookreport.com Corporate Site Lic. $650
http://pobox.com/cook/ for new report: "Tracking Internet Infrastructure"
***********************************************************************


On Sat, 25 May 1996, Tim Bass wrote:

> Paul writes:
>
> >
> > Your message addressed platform performance, not network operations.
> > My reply is mostly just about network operations.
> >
>
> To be sure, you are correct. My subliminal message was more-than-likely
> something like this:
>
> "There is very little justification for being single threaded in
> network operations in today's very inexpensive, high performance,
> UN*X world.' (IYKWIM)
>
> So, the issue, does not really appear to be "Router Servers: Are
> They A Good Thing, Yes or No"? as implied in a much earlier post,
> but more an architecutural issue, single threaded configurations,
> and so forth.
>
> Even 'experimental, non-operational' (whatever that really means in
> the real world.... use at your own risk) services such as RA could
> easily and cheaply have some redundancy built in at a very moderate
> cost in todays world.
>
>
> Back to Saturday, Best Regards,
>
> Tim
>
> Side note to Gordon Cook based on his contribution earlier:
>
> Gordon, just a little barb to you, please do not take it personally,
> but.. "is your current trip to Russia also funded by the NSF similar to
> your previous trip"? ( inquiring minds want to know ;-)
>
> Whoops! There I go again, standing in the corner, just losing my
> religion.... Oh well, I said too much, I haven't said enought....
>
> Micheal Stipe, REM.
>
>
>
>
>
>

- - - - - - - - - - - - - - - - -

Gordon,

I publically apologize for the misunderstanding. I was sure that
you told me, yourself, the NSF funded your first trip to the Soviet Union,
and that your were very appreciative, considering you had earned
your doctorate in a subject related to that part of the country
and had not had a chance to visit that beautiful part of the world.

As I recall, our entire private conversation was, in some form of
fashion, about my concerns about ethics, commercialization, US taxpayer
funds, and the NSF as it relates to the Internet, (a subject that
I have been known to be critical and introspective). Your gratitude
and kind words to the NSF came up in the converstation, as I recall.

It seems, as you are pointing out, that my recollection of events
is distorted or perhaps foggy. Thank you for kind reply. Again,
I offer my deepest apologies for associating you with my interest
in how the NSF distributes US taxpaper dollars.

On the other hand, it is unusual for me to recall such an enjoyable
conversation so vividly and make such a mistake. Hmmmm.
Sorry for misrepresenting what I perceived was an accurate
recollection of an interesting and lively discusssion.

Keep up the good work and enjoy the beautiful sights of Asia!
I'm considering trekking in Tibet in June myself, if this old
tendon in my foot ever heals from 150 miles/week in-line skating.

Best Regards,

Tim

------

And if the cloud bursts, thunder in your ear
You shout and no one seems to hear
And if the band your're in starts playing different tunes
I'll see you on the dark side of the moon.

--- Roger Waters
- - - - - - - - - - - - - - - - -

Tim,

NSF never paid any of Gordon's expenses for anything, to my knowledge at
least. He made an important contribution to the NATO meeting in Moscow in
1994 by delivering a well-thought-out and constructive charge to the
audience about how a bit more cooperation on everybody's part could improve
Internet service for large numbers of Russians. I, as co-chair of the
event, was quite satisfied that I had spent NATO's money wisely by inviting
him to participate. We even fed him (along with the other participants) on
NATO's tab. Even with all that, it was still a good deal for NATO (he
drank much less vodka than most, BTW...).


At 2:18 AM 5/26/96, Gordon Cook wrote:
>tim asks who is paying for my trip to Russia. How about me for
>incidentals and a total of $13.50 for tax on a delta frequent flyer
>ticket? NSF has
>paid for NONE of my trips. In Sept october 1994 NATO paid for a $620
>airplane ticket and I paid $120 to extend it and for all other expenses.
...
>
>On Sat, 25 May 1996, Tim Bass wrote:
...
>>
>> Side note to Gordon Cook based on his contribution earlier:
>>
>> Gordon, just a little barb to you, please do not take it personally,
>> but.. "is your current trip to Russia also funded by the NSF similar to
>> your previous trip"? ( inquiring minds want to know ;-)
>>
>> Whoops! There I go again, standing in the corner, just losing my
>> religion.... Oh well, I said too much, I haven't said enought....
>>

Go stand in the corner, Tim!

--Steve G.

____________________________________________
Steve Goldstein, National Science Foundation
+1(703)306-1949 Ext. 1119
"Let's not procrastinate until next week!"


- - - - - - - - - - - - - - - - -

> delivering a well-thought-out and constructive charge to the audience
> about how a bit more cooperation on everybody's part could improve
> Internet service for large numbers of Russians.

While this was and unfortunately still is an amazingly worthwhile goal, I
can not help but wonder which is more futile, trying to get the Russian
factions to cooperate or explaining something to Tim Bass. Certainly the
former is vastly more worthwhile.

And, while I applaud NATO for sending Gordon to this particular windmill,
and strongly encourage them and others to keep tilting at it (have had a
bash at it once or twice myself), what does this have to do with North
American network operations? More Tim Bass distraction action?

# from my .procmailrc
:
* ^From: (Tim Bass|Jim Fleming|Eugene Kashpureff)
/dev/null

randy
- - - - - - - - - - - - - - - - -

Steve,

Thanks for taking the time to help set the record straight. I did
not intend to pull the discussion off the Router Server single-thread
crash operational experimental discussion.....

I just wanted to have some fun at Gordon's expense, in return for the
fun times when he endearingly called my CIDRD-WG posts "Quixotic"
(and he was right, BTW ;-)

Plus, I sure envy Gordon spending time in beautiful Russia this time of
year. Which leads to my next question.....

Anyone know when there will be an IETF meeting in that part of the
world? That would be a very, very interesting and enlightening
IETF to participate, don't you think?

Best Regards,

Tim

-----

Really don't mind if you sit this one out, My words but a whisper
in deafness they shout. Can't make you feel, but I can make you
think...... ( continuation censored by your Honorable Exon ... )

Ian Anderson


- - - - - - - - - - - - - - - - -

On Sun, 26 May 1996, Tim Bass wrote:

> Anyone know when there will be an IETF meeting in that part of the
> world? That would be a very, very interesting and enlightening
> IETF to participate, don't you think?

Tim, this topic is best hashed elsewhere, not on NANOG.

Also, could everyone please not include my NOC in responses to this thread?
(noc@noc.ns.itd.umich.edu) They have enough to do without having to read NANOG
mail. Thanks.

-dorian

- - - - - - - - - - - - - - - - -

Randy,

It feels great to be piped into your bit bucket. As least /dev/null
has the wisdom not to burn off into a flame because someone has
an opinion or idea he is not interested in reading.

If you wish to continue to beat me up, I am willing to let you take
your frustrations out on me. Randy, please feel free to do it
privately, however. But, as you stated indirectly, it is
easier to bring "peace to the world" than to convince me to accept and
act on the emotional arguments of irrational, mob-o-cratic (sic) minds.

Now, let's continue the 'very important' topic of 'why is is so
difficult' to have redundant RSs in the NAPS. Let's see, we can
flame Merit.... or those operational in an experimental configuration...
or we can flame Tim Bass, the Clueless, Idiot, Moronic, AntiChrist
(no, can't have that honor, that belongs to S. D. the 666 :-)


Tim


PS: Apologies for responding to Randy's personal attack in public.






- - - - - - - - - - - - - - - - -

On Sat, 25 May 1996, Tim Bass wrote:

> The performance numbers for the Pentium Pro have been coming in
> (http://www.silkroad.com/linux-bm.html) and the results are
> very exciting. UNIX ports to Intel architecuture are much more
> mature than Alphas ports (goes without saying) and the P Pro
> numbers are comparable to Alpha.

This not just true for UNIX, a good friend of mine works out at Microsoft
on the NT development team. He claims they've been seeing similar
numbers on Alpha and Pentium Pro machines under Windows NT.

Christopher E Stefan http://www.ironhorse.com/~flatline
System Administrator Home: (206) 706-0945
Ironhorse Software, Inc. Work: (206) 783-6636
flatline@ironhorse.com finger for PGP key

- - - - - - - - - - - - - - - - -

Warning: this is irrelevant to the stated charter of NANOG. Hit "D" now.

I mean it. Don't read this. It's not about network operations.

> This not just true for UNIX, a good friend of mine works out at Microsoft
> on the NT development team. He claims they've been seeing similar
> numbers on Alpha and Pentium Pro machines under Windows NT.

My experience differs significantly, but then my P/Pro is 200MHz and my
Alpha is 333MHz. Microsoft probably has older Alphas and newer P/Pro's.

Any of you installing Sun or Intel machines for netnews or shell or file
service or anything else requiring good source code compatibility (barring
32-bit pointer assumptions such as those in the route servers) and large
numbers of computrons are being silly. DEC would loan you an Alpha for a
month to prove you wrong, since they know you won't be able to live without
it and the chance of them having to take it back is small.
- - - - - - - - - - - - - - - - -

Following Paul's example, this is a warning that this is inappropriate for
the list topic, hit "D" now ;-)


I think that the MP HyperSPARC machines and in particular the new Ultras
are doing Just Fine performance wise and legacy code wise. I've been
working with them on a daily basis for a year and a half at Axil, which
makes more Sparcs than anyone except Sun, and we keep pushing them further
and further. Alphas have a lot going for them, but don't write SPARC off.
It's made up its performance deficit from the early 90s.

-george
- - - - - - - - - - - - - - - - -

On Sat, 1 Jun 1996, Paul A Vixie wrote:

> Warning: this is irrelevant to the stated charter of NANOG. Hit "D" now.
> 
> I mean it. Don't read this. It's not about network operations.
> 
> My experience differs significantly, but then my P/Pro is 200MHz and my
> Alpha is 333MHz. Microsoft probably has older Alphas and newer P/Pro's.

However, since Pentium processors are CISC and Alphas are RISC, the
difference in clock speed may _still_ yield similar performance since
the Alphas have to execute _more_ instructions to accomplish similar
amounts of "work" (on the average since some tasks are simple and some
more complex). Don't get me wrong -- we love our Alphas (so far) and
feel it's probably a more stable platform at this point since the Pro's
are still fairly new.

Northwest Nexus Inc. (206) 455-3505 (voice)
Professional Internet Services
edm@nwnexus.WA.COM

- - - - - - - - - - - - - - - - -

> On Sat, 1 Jun 1996, Paul A Vixie wrote:
>
> > Warning: this is irrelevant to the stated charter of NANOG. Hit "D" now.
> > 
> > I mean it. Don't read this. It's not about network operations.
> > 
> > My experience differs significantly, but then my P/Pro is 200MHz and my
> > Alpha is 333MHz. Microsoft probably has older Alphas and newer P/Pro's.
>
> However, since Pentium processors are CISC and Alphas are RISC, the
> difference in clock speed may _still_ yield similar performance since
> the Alphas have to execute _more_ instructions to accomplish similar
> amounts of "work" (on the average since some tasks are simple and some
> more complex). Don't get me wrong -- we love our Alphas (so far) and
> feel it's probably a more stable platform at this point since the Pro's
> are still fairly new.
>
> Northwest Nexus Inc. (206) 455-3505 (voice)
> Professional Internet Services
> edm@nwnexus.WA.COM
>

Before we continue on a path of naive CPU arguments (or worse, delve
into 32-bit vs. 64-bit, MMU path width, cache, pipelining and stages,
etc. etc.), can we take this offline please?

Daniel
~~~~~~
- - - - - - - - - - - - - - - - -

On Sun, 2 Jun 1996, Daniel W. McRobb wrote:

> Before we continue on a path of naive CPU arguments (or worse, delve
> into 32-bit vs. 64-bit, MMU path width, cache, pipelining and stages,
> etc. etc.), can we take this offline please?

Why? I think most of us can use this info.

Nathan Stratton CEO, NetRail, Inc. Tracking the future today!
---------------------------------------------------------------------------
Phone (703)524-4800 NetRail, Inc.
Fax (703)534-5033 2007 N. 15 St. Suite 5
Email sales@netrail.net Arlington, Va. 22201
WWW http://www.netrail.net/ Access: (703) 524-4802 guest
---------------------------------------------------------------------------
"Therefore do not worry about tomorrow, for tomorrow will worry about
itself. Each day has enough trouble of its own." Matthew 6:34


- - - - - - - - - - - - - - - - -

> On Sun, 2 Jun 1996, Daniel W. McRobb wrote:
>
> > Before we continue on a path of naive CPU arguments (or worse, delve
> > into 32-bit vs. 64-bit, MMU path width, cache, pipelining and stages,
> > etc. etc.), can we take this offline please?
>
> Why? I think most of us can use this info.
>
> Nathan Stratton CEO, NetRail, Inc. Tracking the future today!

Because there are better forums than NANOG to discuss architecture issues.

If people wanted to discuss distributed algorithms & data structures for
route re-computation or compare effectiveness of feeding routes from
RPs to SPs, or even just rant one way or another about using the route
servers, it might be relevant.

Avi

- - - - - - - - - - - - - - - - -