Re: [vulture@aoi.dyndns.org: Bug#100744: Binary should be in /usr/bin, since it's useful to non-admins.]
On 06/19/2001 09:33:41 PM Manfred Wassmann wrote:
>> On Mon, 18 Jun 2001, Vince Mulhollon wrote:
>> > It's not "a rather expensive network analyzing tool".
>> >
>> > Please quantify the costs of a traceroute compared to net
connectivity,
>> > electrical power, and labor.
>>
>> It is rather expensive compared to a ping probe when you only want to
>> test if a host is reachable. By default traceroute uses three probes
per
>> hop with a data length of 38B for the request (not 40B as is stated in
the
>> manpage) and 66B for the response. With an average distance between 10
>> and 20 hops a single traceoute probe generates an average traffic of
about
>> 5k. A three packet ping would only need 444B.
>>
>> If you have 10 users makeing 10 probes a day on 20 working days a month
>> the traceroute probes sum up to about 10 Megs of data.
Well, yes there is a measurable difference, but the point I'm making is
that the "expense" is irrelevant in the greater scheme of things.
Using your numbers, lets assume a 9600 baud modem and a 10 cent per minute
internet access, which is silly-expensive compared to "most peoples
connections now a days". In Midwestern USA, that would have been typical
for long distance dialup access in perhaps late 80s, at least ten years
ago. Much faster and cheaper now, of course.
Your ping example will cost about 444*8/9600*0.10/60 = 0.06 cents for the
network connection.
Your traceroute example will cost about 5000*8/9600*0.10/60 = 0.6 cents for
the network connection.
That is a fairly cheap network analyzing tool, compared to a $40000
protocol analyzer.
Now, lets assume the labor required to really think about the results costs
$20 per hour, and it takes at least a minute to determine an intelligent
destination address, type in the command, read over the results and
comprehend it, compare results to past results, lookup the addresses on a
paper network map, etc. That would be a cost of about 33 cents per
traceroute.
So, using the cheapest labor available to analyze the slowest and most
expensive network I can think of, the difference in total overall cost
between a traceroute and a ping is about (33.6-33.06)/33.06*100 = 1.6%
difference in total overall cost. With better network connectivity and
higher paid employees, the percentage is vastly smaller. My previous
example of a fairly cheap DS1 at a fixed monthly cost (no per bit charge)
and more of a senior sysadmin level of pay, resulted in more like one ten
thousandth of a percent difference.
In your example of 10 users, 10 probes a day, and 20 days a month, the
killer expense is not going to be the network connectivity, which will cost
"dollars" (if not far less), the real expense will be labor, paying 10
employees for a month is going to cost a good fraction of a million
dollars. The net connectivity is literally pocket change in comparison to
the salaries.
Really, you'd save alot more money by patching the kernel to force the
numlock LED on all the time, to help illuminate the users keyboards,
resulting in lower lighting costs. (more seriously I don't know if an LED
and it's power conditioning circuitry is more or less efficient than a
typical light bulb)
The point I'm making is that in a business situation, the difference in
cost is really non existant. The effort required to "work around" the cost
saving measure far exceeds the actual cost savings. There is an English
saying "Penny wise and pound foolish".
Reply to: