[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ga] Fw: ADNS NEWSLETTER: ASLAN SERVER VISIBILITY JUMPS BY 54%




Hi Roeland,
	You didn't understand what I was getting at apparently. I agree
completely with you that the CPU-power you need currently is very low. What
I was saying is that ORSC as it stands today is not handling more than a
completely marginal proportion of internet traffic (probably less than
0.01% of legacy IANA traffic), so for all practical purposes it just doesn;t
exist. The problem of having enough CPU or not just doesn't come into it.

So this lovely newletter stating over half a million hits in one month is
really an attempt of glorifying something that doesn't need to be glorified.
Your servers may be able to handle all the load you want. I'm not arguing
about that. What I *am* arguing is that you don't have any load at all
because (in all practical senses) nobody is using you. From the perspective
of the size of the internet, a mere 500K uses in one month works out to one
use every 5 seconds, which means that nobody uses you or has heard about
you. For all I care, you may have a Beowolf linux cluster of 5000 alpha-500
boxes and have ten units like that spread over the world making you the most
impressive computing facility alive, and would never have a CPU problem in
this matter. Unfortunately it would all be a wasted effort (for the moment).
Please do inform us when you get something like a million hits per DAY. (How
many users on the 'net today? aprox 200 million some say? Even if you get
just ONE hit per month per person, that would be 200 million hits... -yes
oversimplification because of cache etc, but you get the idea-)
The LARGEST rogue^H^H^H^H^Halternative root system out there NEEDS less than
64Kb of bandwidth to run and that's an overkill.
Your 500K hits in one month work out to one hit every 5 secs aprox. Lets say
that 200 bytes are exchanged in each hit (I'm being extremely exagerated
there), that means you are consuming an average of 40bytes per second, in
other words (because of CRC etc) lets say around 400 bits/second. So, in
fact, you don't even need a 64Kb link. You just need a 0.4Kb link. You could
run your current server over a 1200baud modem (over a 300baud you would
probably saturate, but as I've estimated on the higher side, it *might* just
work). So, run it now on a 300 baud modem, check the bandwidth saturation
and if it proves to be too bad, you can migrate to a 1200 baud modem to have
mode than enough bandwidth spare. Hmmm then again that would make your
latency pretty bad, so maybe (if you have the cash to spare) a 9600baud
modem would really eliminate all your latency/bandwidth problems, and at
your current rate of growth shoul allow you another couple of years. Yes,
very relevant I don't think. You're a technical minded guy. Care to indicate
where I've made a mistake?
As far as an excel sheet to backup the above analysis, sorry I don't use
excel to do 2+2 type equations.
So, as you see, there is no FUD, apart from that put out by ORSC trying to
disguise a 300baud-modem system as something that is currently handling
traffic in a major way. I am *NOT* disputing the fact that you MIGHT be able
to handle it. What I'm saying is that the effort as-is today is completely
irrelevant (from the figures you give.

Yours, John Broomfield.

P.S.
/whine on
I *wish* I had 2 E1's. I only have one (E1=2Mb bandwidth). It's a standard
leased circuit, not SDSL. And it costs aprox 240Kff per month (around US$35K).
And I can't colocate out of here, because I need to get the bandwidth from
the 'net to my customers (who ARE local). If you're wondering why it's so
expensive, it's because we're located in Guadeloupe -Caribbean-. Bandwidth
to under-developed areas and third world countries is EXTREMELY expensive
because of monopolistic structuring of the local telcos (you don't even have
to go that far, just ask how much a 2Mb link costs between London and Paris,
and you will see the advantages of breaking up the national telco as they
did in US).

> > Behalf Of John Charles Broomfield
> > Sent: Friday, July 10, 2893 3:44 PM
> >
> > Hi all,
> > 	Apart from the fact that it's just a shameless plug,
> > I'd like to put it into perspective.
> >
> > (...)
> > > 1. Visibility of ORSC Top-Level domains jumps 54% in March
> > (...)
> > > VISIBILITY OF ORSC TOP-LEVEL DOMAINS JUMPS 54% IN MARCH
> > > The results are in and they show that the number of hits on ADNS's
> > > root server / resolver ASLAN.OPEN-RSC.ORG jumped in March
> > to 530,634 hits
> > > compared to 353,000 in February. This is an increase of over 54%.
> >
> > /flame on heat-level=nova
> 
> Really? This sounds like FUD to me ...
> 
> > This means that on average, it was getting around 1 hit every
> > 5 seconds.
> > Note that if I look at the hits that *my* main resolver gets (the main
> > resolver for an ISP with around 3500 dialup customers, and a
> > bunch of hosted
> > sites, but with a total outbound connectivity of 2Mb which is
> > not full -ie a VERY small ISP by todays standards-) it's aproximately 5
> > times that (one hit per second aprox).
> 
> Okay, you have a pair of E1's ... I believe that matches the smallest system
> feeding alternate roots, but they're a pair of SDSL lines (1.1 Mbps each).
> 
> > My system services 3500+ customers. ORSC roots are trying to
> > indicate that
> > they are a RELEVANT alternative to the legacy IANA roots
> > (which serve for
> > all *practical* purposes ALL of the internet).
> > If someone can post the hits per second of the legacy IANA
> > roots, it would
> > put it even further in perspective, but I suspect that the
> > ORSC is something
> > around 0.01% of hits at the VERY most. I don't call that
> > relevant. In fact I
> > call it totally IRRELEVANT.
> 
> As pointed out before, the DNS is a very light load. Even a K6-200 can
> handle multiple hits per second and a dual PIII-800 could probably handle
> the entire Internet, with all zones local. But, if one really feels like
> spending money, a VALinux Cluster City would handle all possible loads, with
> one gig-ether NIC
> disconnected.<http://www.valinux.com/systems/clustercity.html>
> 
> > /flame off
> >
> > A burning commentary, but true nevertheless.
> 
> I'd like to see a capacity analysis supporting your statement. An Excel
> spreadsheet would be fine. I you have a problem with MS products, xspread is
> also useable. Even my esteemed opponent, Kent Crispin, agrees that you don't
> need a big box for DNS. What you need is decent pipes and, with DSL and
> whatnot, those are getting cheaper every day. (Covad SDSL [1.1 Mbps] is
> $348US per month, retail. [Sorry John, I know what those E1's are costing
> you]).
> 
> I suspect that in the process of building such an analysis (which you
> clearly haven't done) you will come to the realization that making such
> claims, in an audience containing folks who do that sort of thing, as part
> of their livelyhood, is more than slightly risk-prone. The numbers had
> better be there, they had better be right, and you MUST show your work.
> Anything less is FUD.
> 
> BTW, didn't we go throught this effort on the IFWP list, or was it
> DOMAIN-POLICY.
> 
> --
> This message was passed to you via the ga@dnso.org list.
> Send mail to majordomo@dnso.org to unsubscribe
> ("unsubscribe ga" in the body of the message).
> Archives at http://www.dnso.org/archives.html
> 

--
This message was passed to you via the ga@dnso.org list.
Send mail to majordomo@dnso.org to unsubscribe
("unsubscribe ga" in the body of the message).
Archives at http://www.dnso.org/archives.html