|
Grex > Coop11 > #32: How should we determine how many dialin lines we should have? | |
|
| Author |
Message |
| 25 new of 154 responses total. |
rcurl
|
|
response 25 of 154:
| 
Oct 6 13:54 UTC 1998 |
Isn't there any way for Grex to accomplish anything that most decide
is a "good thing" but for which no volunteers come forward? How about - paying
someone? It is not a new idea. There seems to be a general feeling that
Grex does not do enough advertising, contacting its users to ask for support,
etc, but I can understand the "techies" not being interested in doing these
things and most board members are "techies". I don't like doing those things
either, but can appreciate their necessities. So, either *find* volunteers
liking to do those things (advertise for them!) or hire the work done.
|
aruba
|
|
response 26 of 154:
|
Oct 6 17:42 UTC 1998 |
Mary, there is no such thing as an "unused line" - all of our dialin lines
are used for some amount of time. The question is, how much use merits
keeping a line? How much use merits adding another line? In what units
do we measure use?
We need a quantitative answer here, so we need quantitative criteria as
well.
Certainly the answers might depend on our financial situation; if times
are good, we can afford to have less busy signals than when times are
lean.
I think there is little question that at the October board meeting, we
will vote to drop some lines. I'd like to have some way to figure out how
many, and I'd like to be able to make the same decision every three months
from now on. (Of course, some quarters we may decide to add lines rather
than drop them.) I want to be scientific and consistent about the
decisions, and that means not depending on a lot of anecdotal evidence.
We have some hard data, and can obtain more, but we need to decide how to
use it.
|
mary
|
|
response 27 of 154:
|
Oct 6 20:39 UTC 1998 |
Mark, we have unused lines. They are the amount of times lines
are not in use. I don't see this as so complicated. You
figure out what percentage of the lines are unused at the
busiest part of the day and that is the excess.
I must be missing something here.
|
scg
|
|
response 28 of 154:
|
Oct 6 22:02 UTC 1998 |
I guess my suggested criteria would that we should cut whatever lines are not
in use for at least 15 consecutive minutes on at least four days a month, and
maybe add lines if all lines are in use for at least 15 consecutive minutes
8 days in a month, or one hour on four days a month. That would be a rough
starting point for me, but those numbers are pretty random.
|
aruba
|
|
response 29 of 154:
|
Oct 6 22:29 UTC 1998 |
Re #27: If you look at the data which shows how many lines are used each day,
most days have a max of between 10 and 12 lines. (In other words, the most
lines in use at one time on those days is between 10 and 12.) But some days
the max is 13 or 14, meaning that all the lines were full, and anyone dialing
in during that period would get a busy signal. So the last couple of lines
are not "unused", merely "seldom used". The question is, how seldom a use
is worth paying for?
|
aruba
|
|
response 30 of 154:
|
Oct 6 22:37 UTC 1998 |
Steve's #28 is the kind of thing I think we should come up with here.
|
scott
|
|
response 31 of 154:
|
Oct 6 23:21 UTC 1998 |
I'd agree with Steve's #28. But it still seems a bit too in favor of keeping
lines that don't get much use.
How about if we cut lines if there are any lines in use less than 5 minutes
on an average day?
|
mary
|
|
response 32 of 154:
|
Oct 7 01:48 UTC 1998 |
If the dialin lines see the most use during the evening hours, say 4:00 to
12:00, that's 8 hours. Say we have 14 dialin lines. That's a total of
6,720 minutes of peak-use dial in time available. Do we know the number
of non-used minutes we used during those hours, total, over a recent 30
consecutive day period? If so, divide that number by 30. What percentage
of those 6,720 minutes were unused? Multiply 14 phones lines by that
percentage, rounding up and down to make it an even number. I have a
feeling that's how many lines we could cut without too much hassle.
The hours I suggest and the number of dialin phones lines might
not be accurate. But if the accurate times and numbers were
available, and it wasn't a lot of work, I'd really like to
see what that formula suggests we do.
|
scg
|
|
response 33 of 154:
|
Oct 7 02:00 UTC 1998 |
The problem with that method (if I understand it correctly without doing all
the math involved) is that it goes to average usage, while what matters is
peak usage. If we have 100 users, all of whom spend 1/10 of every day on
line, then if they all spread their usage out evenly, 10 or 11 lines might
be enough to not have busy signals. If all 100 want to connect at the same
time, we will need lines to accomodate them. I suspect we're at neither
extreme.
If we were a business, with the main goal of keeping our customers happy, we
would want to guarantee that there wouldn't be any busy signals. That would
mean that if there were some time of day when 50% of our customers were on
line, we would want enough modems to accomodate 50% of the customers at once,
without a problem. We're a non-profit chairity, providing a free public
service. Our goal can't be to provide the best service around, because it
just isn't possible given our resources (ok, maybe it is, but that would
reflect badly on the commercial services rather than particularly well on us).
What we need to do is figure out how we can best serve our mission of
providing access to everything that Grex is while living within our shoestring
financial limitations. Given that, the occasional busy signal is ok. The
question we have to answer is how long is it acceptable for it to take for
somebody to get through. 15 minutes very occasionally is probably ok,
especially since it hopefully won't be the same person hitting the busy
signals every month. I'm not sure if five minutes a day would be acceptable
or not. If it's always the same people gettin gthe busy signals, I suspect
it isn't. If it's different people different days, which it probably would
be, we can probably live with it.
|
mary
|
|
response 34 of 154:
|
Oct 7 13:20 UTC 1998 |
Er, my percentage formula called for only looking
at peak use time.
I agree with your entire second paragraph.
|
rtgreen
|
|
response 35 of 154:
|
Oct 8 05:47 UTC 1998 |
Mary and Steve: You're both right, sort of. Mary's formula would yield
the average use during the peak hours. ;-)
I, too, like the second paragraph. We're on the right track here. In my
thinking, the number of incidents of a busy signal is much less important
than the duration of an incident. If each incident is only a minute, I
think we could tolerate them every hour, but if each incident were fifteen
minutes, I would get worried if it happened more than once a day.
I haven't looked recently. What does qgraph tell us for the past two
months? Is it realistic to look at July & August as typical, with
vacations giving people more time to grex, or leave town and not grex?
|
janc
|
|
response 36 of 154:
|
Oct 28 20:45 UTC 1998 |
I kludged up my dialup usage program to make it possible to skip over
mangled parts of the wtmp file. I was able to get August data, and
data for the current month, but September was kind of chewed up. I'm
not 100% confident of this data. But here it is:
Usage between Sat Aug 1 00:00:00 1998 and Mon Aug 31 23:59:59 1998
of IP addresses: 204.212.46.131
LINES HOURS PERCENTAGE
0: 131.54 17.68%
1: 82.23 11.05%
2: 104.12 13.99%
3: 101.13 13.59%
4: 93.41 12.55%
5: 77.11 10.36%
6: 55.66 7.48%
7: 38.16 5.12%
8: 28.64 3.84%
9: 15.23 2.04%
10: 8.38 1.12%
11: 4.91 .65%
12: 2.46 .33%
13: 1.02 .13%
TOTAL 744.00 100.00%
Average number of lines in use: 3.36
Usage between Thu Oct 1 00:00:00 1998 and Wed Oct 28 15:07:48 EST 1998
of IP addresses: 204.212.46.131
LINES HOURS PERCENTAGE
0: 29.46 4.43%
1: 94.28 14.19%
2: 113.62 17.10%
3: 121.61 18.31%
4: 108.76 16.37%
5: 83.13 12.51%
6: 52.55 7.91%
7: 30.95 4.66%
8: 15.92 2.39%
9: 8.09 1.21%
10: 3.86 .58%
11: 1.38 .20%
12: 0.39 .05%
13: 0.10 .01%
TOTAL: 664.10 100.00%
Average number of lines in use: 3.5068
Mark is also working on generating some data, and Scott should be able to
get some from the terminal servers. Personally, I think we can drop two
lines, and should certainly drop one.
|
janc
|
|
response 37 of 154:
|
Oct 28 21:06 UTC 1998 |
Here are some snapshots of who was on the dialin when 13 people were dialed
on this month:
--USER-- --LINE-- --------SINCE---------
sixx ttyqa Tue Oct 6 19:44:08 1998
pthomas ttyuc Tue Oct 6 19:48:15 1998
wrestler ttytd Tue Oct 6 19:54:07 1998
jor ttyp5 Tue Oct 6 19:16:12 1998
aruba ttys3 Tue Oct 6 18:34:39 1998
zeroacid ttyrf Tue Oct 6 19:08:04 1998
happyboy ttyq7 Tue Oct 6 19:43:57 1998
bookie ttys8 Tue Oct 6 19:52:51 1998
kme ttyr6 Tue Oct 6 18:29:06 1998
tiyose ttyse Tue Oct 6 19:25:34 1998
acorn ttyt2 Tue Oct 6 18:34:02 1998
headdoc ttyq6 Tue Oct 6 19:44:28 1998
benni ttyud Tue Oct 6 19:54:23 1998
--USER-- --LINE-- --------SINCE---------
bookie ttypb Tue Oct 6 22:49:06 1998
acorn ttytb Tue Oct 6 22:41:48 1998
dpfitzen ttyue Tue Oct 6 22:12:33 1998
sekari ttype Tue Oct 6 22:28:51 1998
lbmiller ttyqe Tue Oct 6 22:42:16 1998
morpheus ttyqb Tue Oct 6 22:49:51 1998
gabriel ttyr1 Tue Oct 6 22:37:24 1998
suzie ttyua Tue Oct 6 22:26:48 1998
russ ttyrf Tue Oct 6 22:44:36 1998
kme ttyr6 Tue Oct 6 18:29:06 1998
eeyore ttyp7 Tue Oct 6 22:49:56 1998
jbo ttys5 Tue Oct 6 22:50:46 1998
ricke ttyp9 Tue Oct 6 22:50:49 1998
--USER-- --LINE-- --------SINCE---------
dpfitzen ttyp4 Sun Oct 18 20:27:32 1998
traveler ttyp2 Sun Oct 18 20:21:09 1998
lbmiller ttyp5 Sun Oct 18 20:28:56 1998
other ttyq7 Sun Oct 18 19:56:56 1998
krj ttyq1 Sun Oct 18 20:14:22 1998
snookie ttys7 Sun Oct 18 20:28:45 1998
aruba ttyq2 Sun Oct 18 20:21:22 1998
kharder ttyua Sun Oct 18 20:02:10 1998
mermaid ttyuf Sun Oct 18 20:07:28 1998
hmv ttyub Sun Oct 18 20:20:03 1998
laotzu ttys2 Sun Oct 18 20:27:57 1998
illogic ttysf Sun Oct 18 20:09:00 1998
dsmith ttyu8 Sun Oct 18 20:08:13 1998
--USER-- --LINE-- ----------SINCE----------
mary ttys0 Thu Oct 22 21:26:53 1998
acorn ttyt9 Thu Oct 22 21:28:20 1998
eeyore ttypb Thu Oct 22 21:12:48 1998
sixx ttyq6 Thu Oct 22 21:28:35 1998
senna ttyp5 Thu Oct 22 21:28:41 1998
thea ttyr4 Thu Oct 22 21:07:13 1998
n8rxs ttyr8 Thu Oct 22 20:40:16 1998
imax ttyp1 Thu Oct 22 21:31:09 1998
indi ttyrc Thu Oct 22 21:18:12 1998
ckallery ttyqc Thu Oct 22 20:34:06 1998
kharder ttys5 Thu Oct 22 21:34:38 1998
mrmat ttys3 Thu Oct 22 21:17:07 1998
orinoco ttyrd Thu Oct 22 21:32:05 1998
I don't think aruba uses the staff dialin, and I'm sure none of the others
do, so I think this means we were full with regular users, and anyone else
dialing in during the 6 minutes in October that these cover would have
gotten a busy tone. If we cut one line, we'd probably have had a bit more
than 35 minutes of busy tones this month.
Unless other people get data that looks much difference, my inclination would
be to cut just one dialin.
|
aruba
|
|
response 38 of 154:
|
Nov 3 19:40 UTC 1998 |
I called Ameritech today and had them drop the last two lines in our trunk
hunt, per the board's vote last Wednesday. See the minutes for details.
|
janc
|
|
response 39 of 154:
|
Nov 6 05:19 UTC 1999 |
At the last board meeting, Mark reminded me that I once wrote a program
that measures dialin line use. I happened to find it today, and ran it:
August:
NUMBER TOTAL
OF LINES HOURS PERCENT
0: 111.53 14.99
1: 141.76 19.05
2: 146.90 19.74
3: 124.61 16.74
4: 94.99 12.77
5: 60.64 8.15
6: 32.91 4.42
7: 17.04 2.29
8: 8.73 1.17
9: 3.81 0.51
10: 0.90 0.12
11: 0.19 0.03
September:
NUMBER TOTAL
OF LINES HOURS PERCENT
0: 81.64 11.34
1: 167.46 23.26
2: 176.80 24.56
3: 126.28 17.54
4: 82.78 11.50
5: 45.93 6.38
6: 22.25 3.09
7: 10.65 1.48
8: 4.18 0.58
9: 1.57 0.21
10: 0.45 0.06
October:
NUMBER TOTAL
OF LINES HOURS PERCENT
0: 101.93 13.70
1: 162.66 21.86
2: 178.22 23.95
3: 124.79 16.77
4: 86.01 11.56
5: 46.84 6.29
6: 25.89 3.48
7: 12.07 1.62
8: 4.82 0.65
9: 1.46 0.20
10: 0.28 0.04
11: 0.02 0.00
|
scg
|
|
response 40 of 154:
|
Nov 6 05:22 UTC 1999 |
So it looks like our last line's barely in use. In theory we could cut it.
The same could probably be said for the last four or so lines.
However, one thing I've noticed at work is that dial-in line usage goes way
up as the weather outside gets colder. I'm not sure if that applies to Grex
or not, but it makes sense that it would. It will be interesting to see if
usage of Grex's dial-ins picks up over the winter.
|
gelinas
|
|
response 41 of 154:
|
Nov 6 06:00 UTC 1999 |
Yes, it will. When first reading Jan's stats, my thought was to reduce
to 8 or 9 lines. After reading the text and responses, I'd say dropping
to 9 would be reasonable.
|
other
|
|
response 42 of 154:
|
Nov 6 06:22 UTC 1999 |
since dropping doesn't cost, but adding does, i would prefer to see drops done
at a rate of no more than one per month, just to allow time to reassess.
|
don
|
|
response 43 of 154:
|
Nov 6 23:13 UTC 1999 |
How much is each line costing us per month now? I don't think it would be that
much compared to the profits/losses per month.
|
scott
|
|
response 44 of 154:
|
Nov 7 00:33 UTC 1999 |
Each line is about $20/mo? That tends to add up pretty fast.
|
gelinas
|
|
response 45 of 154:
|
Nov 7 01:37 UTC 1999 |
I just realised that my first sentence in #41 is ambiguous. The pronoun
refers to "interesting", not to dial-in usage. I've no opinion of what
will happen to dial-in usage.
|
aruba
|
|
response 46 of 154:
|
Nov 7 16:07 UTC 1999 |
Cutting a phone line would save us approximately $18.81 per month.
|
remmers
|
|
response 47 of 154:
|
Nov 8 17:51 UTC 1999 |
And $18 is equivalent to three monthly memberships. That's a
significant amount.
I agree with Eric that we shouldn't drop more than one line at once,
and should allow time to assess the consequences before making
further decisions. Line 11 is definitely a candidate for removal
if low usage continues during the cold weather months.
|
kaplan
|
|
response 48 of 154:
|
Nov 10 01:54 UTC 1999 |
How much would it cost to reconnect one?
I'd say users would barely notice if get rid of two or three lines
based on Jan's numbers. We're not running a business here where people
will take their money elsewhere at the first busy signal. I think
most of our users will tolerate dialing back in again later if they get
busy signals.
|
aruba
|
|
response 49 of 154:
|
Nov 10 15:08 UTC 1999 |
I believe it costs $42 (plus tax) to install a new line. At least that was
the price last time we did so.
|