|
|
| Author |
Message |
| 25 new of 270 responses total. |
ajax
|
|
response 200 of 270:
| 
Oct 3 21:33 UTC 1995 |
Interesting, STeve! It occurs to me that if our users rarely
rise above 52 (or thereabouts?) that the "forties" might have
a mean of around 44 users, whereas the "fifties" might have a
mean of around 51, indicating a difference of seven users. (Do
you have the files handy to compute those numbers?)
If those guesses are right, the figures kind of substantiate
the theory that the last few users put a disproportionate load
on the system (the last 15% of users increased the average load
by 33%), although not by quite as much as I'd have figured.
|
steve
|
|
response 201 of 270:
|
Oct 3 23:55 UTC 1995 |
The data comes from /var/stats/UptimetimeLog; I can do that
tomorrow, or you can.
|
mlady
|
|
response 202 of 270:
|
Oct 4 01:18 UTC 1995 |
Weird. Try it, I guess. If I can't get on sometime, I'll know who
to decapitate.
|
janc
|
|
response 203 of 270:
|
Oct 4 02:33 UTC 1995 |
If it's any reassurance, if you see the "out of ports" error message then
the only thing you are missing is a *really* slow Grex, and if you are already
on when Grex gets *really* slow, then at least you will be saved a couple more
users who would make it even slower. It's not such a bad deal.
|
selena
|
|
response 204 of 270:
|
Oct 4 05:32 UTC 1995 |
<Selena *hates* the "out of ports" message!>
Damndestly most annoying thing I ever run into- my hardware's
running well, michnet lets me on, msu-gopher's not tied up, get all the way
here and, WHAMMO! No ports. Hate it.
I am not reassured by the thought of missing grex at all, slow or
not.
I gotta admit- I like debra's style.. deapitation, hmmm..
|
scg
|
|
response 205 of 270:
|
Oct 4 07:44 UTC 1995 |
I've said this before, so I should probably stop repeating myself, but I would
rather wait a while for a faster Grex. Then again, I seem to be managing that
ok now, without the limit, simply by not Grexing in prime time. I've also
generally got real work to do with my computer, meaning that if I can't Grex
I can generallly find something else to do without getting up. I guess that
may not be the case for some people.
|
steve
|
|
response 206 of 270:
|
Oct 4 14:30 UTC 1995 |
Before you sharpen the axe Debra, just remember that we already
have the 'no ports available' message from time to time on Grex right
now. By taking 8 pty's away I think we'll increase that to about
1.5 hours, spread mostly between two time periods: late afternoon
and late night. Since people come and go constantly, getting one
of those messages basically means that you should try again in about
30 seconds, becuase lots of people pop in for a minute to look for
friends, or check mail and then get out quickly.
|
ajax
|
|
response 207 of 270:
|
Oct 4 15:34 UTC 1995 |
I ran the numbers myself, and here are the results:
Num. Load Num
Users Avg. Times
1: 1.74 175 26|
10: 2.81 890 l 24| -
20: 4.00 2741 o 22| _
30: 5.64 3285 a 20|
31: 5.87 3365 d 18|
32: 6.46 3360 16| _
33: 6.80 3200 a 14|
34: 7.41 3049 v 12| _
35: 7.95 3003 e 10|
36: 8.70 2709 r 8| _
37: 9.33 2693 a 6| _
38: 9.69 2537 g 4| _ _ _
39: 11.00 2264 e 2|_ _ -
40: 11.63 2114 0|___________________________________
41: 12.28 1786 01 10 20 30 40 50
42: 13.00 1725
43: 14.41 1545 users
44: 15.26 1381
45: 16.01 1243
46: 16.63 1122
47: 17.67 851
48: 18.57 716
49: 19.24 574
50: 19.70 435
51: 22.08 306
52: 21.29 192
53: 21.97 123
54: 24.18 70
55: 25.03 47
56: 23.43 18
57: 24.45 1
I think the curve's tapering off toward the end is because as the
curve hits its top, people will log off a lot more, so the times
when 55+ users are on are "relatively fast" for that many users.
Of the 96,176 minutes (27 days) logged, if we capped the maximum
users at 45, it would have turned people away for 4,455 minutes,
around 5% of the time. (This doesn't differentiate ptys vs. ttys).
|
janc
|
|
response 208 of 270:
|
Oct 4 16:52 UTC 1995 |
I think the tapering off of the curve toward the end is probably due
to the fact that at a certain point people find the system so slow
that they begin to modify their behavior to avoid especially slow
tasks. I occasionally join party when other parts of the system are
just too slow for words.
If that theory is correct (and the taper-off is not just an anomoly arising
from having less data a those levels) then maybe the point where the
taper-off begins is the point where Grex is, in most people's perceptions,
"too slow," and we should consider setting the user limit near that point.
Say around 52 users.
|
rcurl
|
|
response 209 of 270:
|
Oct 4 18:38 UTC 1995 |
People are already doing that for you. What you want, I would think,
is to (say) halve the saturation load to say 15, which means you limit
the ptys to about 40.
A thought has occurred to me. If we increase the limit, the system will
slow down so much, that everyone will give up and stop doing things,
and *the load average will drop precipitously*. That is, the system
will have cracked the capacity barrier, like the sonic barrier, and
from there on the load average will decrease as the number of users
increase. Eureka!
|
davel
|
|
response 210 of 270:
|
Oct 4 20:51 UTC 1995 |
Um, Rane, what planet are you telnetting in from?
|
chelsea
|
|
response 211 of 270:
|
Oct 4 20:53 UTC 1995 |
Hey, no fair, I'm the only one who can tell him that!
|
rcurl
|
|
response 212 of 270:
|
Oct 4 21:47 UTC 1995 |
Yes! I forgot for a while there who I was.
|
adbarr
|
|
response 213 of 270:
|
Oct 4 23:40 UTC 1995 |
Who did you think you were? Have you ever had this feeling before?
Did you have a happy childhood?
|
popcorn
|
|
response 214 of 270:
|
Oct 5 00:56 UTC 1995 |
Hey, I liked Rane's idea. :)
|
ajax
|
|
response 215 of 270:
|
Oct 5 04:26 UTC 1995 |
<rofl@rane> I think srw's observation from #176 is more likely: after
cracking the capacity barrier, grex would run backwards!
(: !ti yrt ton s'teL
|
mdw
|
|
response 216 of 270:
|
Oct 5 05:37 UTC 1995 |
I'd be really profoundly disappointed to see us skewing the number of
pty's to favour the "no ports" message. I don't think that's what grex
should be about, and I think it's likely to badly backfire as people
start "camping out" on pseudolines to keep them.
|
srw
|
|
response 217 of 270:
|
Oct 5 07:56 UTC 1995 |
As a thought experiment, suppose we increased the bandwidth and CPU speed
by a factor of 3 or 4 each. Would you support proportional increases in
ptys to keep the pseudolines from being camped on? You'd have to, because
the faster system would attract more people until it was just as slow
as it is today. I wouldn't.
I think the users need to be limited somehow, and my preference is to
speed the system up, at least a little, rather than rely on loadavg and
netlag to keep more users out, which is what we do today.
|
popcorn
|
|
response 218 of 270:
|
Oct 5 10:39 UTC 1995 |
I agree. I'd rather have a usable system than a big, hopelessly slow,
system.
|
ajax
|
|
response 219 of 270:
|
Oct 5 15:44 UTC 1995 |
Same here. Another thing I wonder though: if people "attack
telnet" Grex (issuing continuous telnet requests until they
get in), how would that impact Grex? Do people do that?
On the other hand, as pure conjecture, if people get turned away
more in prime time, they might shift/reduce their Grex time, so
the amount of "connections full" time might diminish somewhat on
its own, providing a form of self-regulation. Plausible?
|
steve
|
|
response 220 of 270:
|
Oct 5 18:36 UTC 1995 |
Marcus, I'm confused at what you said in #216. Are you for or against
the idea of removing some ptys to make the system slightly faster when
nearing a full-house condition?
|
ajax
|
|
response 221 of 270:
|
Oct 5 20:48 UTC 1995 |
(Seemed pretty clearly against removing ptys to me.)
|
lilmo
|
|
response 222 of 270:
|
Oct 6 17:12 UTC 1995 |
Re #219: If i consistently got "Ports full" msgs at a given time of day, *I*,
at least, would make an effort to avoid calling near that time of day.
|
mlady
|
|
response 223 of 270:
|
Oct 6 19:05 UTC 1995 |
Well, I've never gotten that message, si if I do, I'll have to figure
it's mostly from this. I don't think you guys really need this- you
aren't going to be fast, even if you do cut it down to 40, so
don't try and fool yourselves. I don't come here for the speed, that's
for sure, and I don't think anyone else does either.
|
steve
|
|
response 224 of 270:
|
Oct 6 19:32 UTC 1995 |
Heh. Thats for certain. So the question remains, would people rather
have a slightly faster Grex that is full a little more often, or would
folks prefer a system that is slower, but more available?
|