#Suicide as an option <g>
36 messages in this thread
Thanks David S. for all the insight. I am a little confused by one
paragraph and that is:
>> My own preference is not to use non-dedicated servers. I have to
approach this with the thought of reliability though. I have 47
workstations out there and if one of them is also a server the
likelyhood of the application being run on it crashing and requiring
a hard reset is unacceptable to me. That hard reset takes out the
server as well. <<
It sounds like you'd rather use dedicated servers but then you state
"if one of them is also a server the likelyhood of the application
being run on it crashing and requiring a hard reset is unacceptable
to me." Which sounds to me like if you have a dedicated server and it
crashes, everything else goes down as well. Could you elaborate on
this for me? Thanks for your time, and I'll check to see what the
status is regarding your recommendation to ADESK. -JE
Sorry for the confusion <G>
On a non dedicated server your machine is doing double duty. It is
the server, and at the same time it is someones workstation. If the
application they are running crashes so badly that you have to reset
the computer to recover (as opposed to just using <CTRL><ALT><DEL>)
then you are rebooting your server too, much to the annoyance of your
other users. With a dedicated server your machine is only doing one
thing, being a file server. The likelyhood of having to reboot it is
less than if it were non dedicated.
In a small operation a non dedicated server may not be much of a risk.
You need to look at how your users function and how critical their
work is and then decide if the cost savings of using a non dedicated
server (you save the price of one machine) is worth the possibility
of having your non dedicated server lock up and need to be reset
while there are active files open.
This is not that big a disaster. Network operating systems are
pretty flexible, expecially the peer to peers. If an application on
a non dedicated peer to peer server hangs you can usually just walk
around the office and ask folks to save their work and log off, then
go reset the server. The server software and the local application
are running in different virtual machines. Having one lock usually
doesn't affect the other. Because of the size of our network and the
type of applications I feel much safer with a dedicated server.
Thankyou very much David, your insight has been most helpful. I think
I'm going to start peer to peer and if things start getting
overloaded, I'll dedicate my 368/40 as a server. I like the server
idea as it would seem to eleviate some headaches in troubleshooting.
This would seem to be another advantage to this approach. And thanks
for taking the time. -JE
My pleasure…As I was saying to David Taffet, I have gotten so much
from this forum by hanging around the edges and listening that I was
delighted to be able to be a small help for a short while and "repay"
my debt a tiny bit! Have a great weekend!
Well David S. you've helped alot, and I hope you stick around, I think
you'll find that your expertise will be valuable to others as well
and maybe you'll catch the 3DS bug and I can reciprocate 🙂 Hope you
have a great weekend too. -JE
In a network with a dedicated server, the traffic is totally managed
by the server, in that, if the server crashes, the network is no
more.
David,
I have a 3 machine lantastic 4.1 network, with a non-dedicated server.
If I'm doing network rendering and the server crashes the slaves will
hang-up too, and also require rebooting. So much for a non-dedicated
solution. I was hoping that if my non-dedicated server crashed and I
got it rebooted fast enough, the slaves wouldn't notice, or would
give me some error message (server connection broken. retry?) Oh
well..:-(
Martin – Networks are very weird beasts. While no two PCs are the
same, even when all the hardware is "the same" no two networks are
even on the same planet. There are some characteristics in common,
tho'. Like most people have 4 limbs, etc. If a person's heart stops
they die, regardless of upbringing. Similarly, if a server dies, the
network goes bye-bye…
David,
networks _are_ very weird beasts! A lot of us would probably love a
dedicated server solution, but can't afford it. The next best thing
might be to pick the _most_ stable system you have and make that the
non-dedicated server. I, being very stupid at the time, picked the
most unstable of my 3 systems to be the server. (pass the pointed
"dunce" hat, derrrr!). This became such a problem that I started
using the network just to send out projects to each rendering
station, so that each rendering station had the required project and
all maps locally. I then just have the 3ds network directory on the
local system – this is so that no system is depending on the server
if it crashes, everything it needs is on it's local drive. This was
more of a desperate measure, and once I have a stable server I'll
probably go the intended route.
Martin – As a network *professional*, I feel that, if your network
configuration is working, you have a functional network! Hey, any port on a
storm and all that… Your department of redundancy department of the
tautological guard… <stupid g>
DT
David,
wow, thanks for the free advice. Probably would have cost me $100 per hour to
get that kind of advice from you _if_ this wasn't CI$. I still get the
occasional corrupt file written across the network.
Nah. I'm too much of a woose to charge what I'm worth. <g> For *you*
it'd be $30/hr. Like I charged the lady the other evening who
couldn't figure out why her MIDI card would not play through the
speakers on her MIDI keyboard. (She was plugged into a line out. The
keyboard had *no* line in… That'll be $75, please.)
don't you just those type of clients. You look like a hero, without
even trying. Hey, $30/hr in Wisconsin is probably equivalent to
$60/hr in LA – your expensive!!!
And, like the ride to *the future*, "well worth a dollar"…
Martin, could you do me a huge favor?
If you've got some time (an hour or so at the very most), could you set up a
video post rescaling of some bitmaps (no mesh rendering, just FAST stuff) and
put it on the network queue for all machines? I'd be very grateful, and would
be exceedingly curious to hear if it works for you. At this point I'm really
convinced that I've bumped into a limitation of Lantastic peer to peer.
Thanks,
John Tissavary (La Luna cie)
John,
sure I could do that. Just to make sure I understand what you are saying: I
would have a sequence of bitmaps on my server (rather than locally on each
machine), then I render in video post with rescaling, send this job to each of
the machines in the queue. OK, I'll give it a bash.
You got it! Thanks for humoring me, I really appreciate it. Don't forget to
write the rendered files to the same disk as you are reading from.
John Tissavary (La Luna cie)
John,
I just finished my test and here are my results. I rescaled 270
512×486 targa files on the server down to 320×200 and saved them back
to the server as .gifs. Everything rendered without errors (no
crashes and no missing files) BUT 2 out of 270 of the files that were
saved back to the server were corrupted and could not be loaded back
into 3ds. To test the integrity of the .gifs I set up an .ifl file
and just rendered them as a background. I got the message the first
file was corrupted and then after removing this from the .ifl I
started the rendering again and on the second bad file I got the
message "can't find resz0113.gif in map-path(s) or image path. This
message was erroneous because the file was there, but it was
corrupted. Maybe that's what was happening to you – a corrupted file
giving a misleading error message. Another thing that confused me is
that I can load the files that 3ds calls corrupted into animator pro.
Weird!
Anyway, it seems I have network problems too (I was already aware of
the corrupted file problem), only they are different from yours.
I think I stumbled onto the real cause of my network problems tonight,
thanks to your message and that of a couple of other people. I think
one of my AE-2 cards is tweeky. When I queue VP stuff like what you
did on the other two machines, things seem to move fine. I'll trade
the card for a new one and see what happens.
I wonder why those images were corrupted? I haven't had any corrupted
images out of 3DS yet. But who knows, maybe I'll get lucky <g>.
John,
turns out the images weren't corrupted but 3ds reported them as such in that
session. I quit 3ds, then restarted 3ds again, then 3ds had no problem with
those 2 images. I don't know why it would have a problem with 2 out of 270
images in one session but not in the next. Too weird :-%
I have strange things like that happen in 3DS sessions, too. I had a preview
show up as corrupted, but when I exited and started 3DS again it was fine. I
think there are still bugs in R3… perhaps we should post these to the Yost
group as bug reports.
John Tissavary (La Luna cie)
John,
trouble is you need to have a reproducible case to make a bug report, and most
of these things are hard to reproduce. Yet again, it's probably some memory
management weirdness that's going on.
They aren't bugs, John. They're spurious system problems that you're having
due to a hardware problem. There is no way that we can develop a piece of
software that will work perfectly in every possible DOS configuration and
hardware setup. Of course, if you can come up with a reproducible problem, let
me know and I'll look at it.
– G
Martin:
>> I was hoping that if my non-dedicated server crashed and I got it
>> rebooted fast enough, the slaves wouldn't notice
Only if you reboot while they are in the middle of a render! And pray
a lot, besides!
This is why a "stateless" network works so well. If the server goes
down and comes back up, the clients won't notice. On some systems,
however, if it goes down, everything else must be reset.
We've had occasion to reboot our server and all renderings continued
when it came back up, except for those dreaded "Error writing Targa
file" boxes, which I agree would be great if they could give you an
option to retry instead of you just giving them permission to crap
out by hitting the 'continue' button. <g>
I've found that if my slaves are in the waiting mode, they always
hang up if the server crashes. If they are rendering they will
continue to render until and even write the file, if it's set to
write locally, but get a error mesage or hang up if the server goes
down. More crash testing is necessary, I feel – I sound like an auto
manufacturer <g>. What's a stateless network? Is that what you have,
oh netmeister? <g>. I assume I don't have one.
Martin:
>> What's a stateless network? Is that what you have, oh netmeister?
Yes, PC-NFS is a stateless network. The easiest way to see if yours
is also would be to 'mount' a drive on the server, copy a file from
the client to the server, reboot the server, and see if you can copy
it back without doing anything else on the client.
It basically means that the server doesn't keep track of what is
mounted on it, so everything can continue just where it left off.
Greg, oh exalted net-highness of the Sun, the moon, and the PC-NFS, I
think I like stateless! But mine isn't, aargh! I will always get the
message: "Server connection to network node xxxx broken reading
drive G Abort, Retry, Fail?" Retry will always get the connection
back, if you have brought the server back on line, but 3ds doesn't
seem to perform retries <sob>. Anybody, how can I get a stateless
network???
Martin:
>> Anybody, how can I get a stateless network???
Well, the lower priced Suns are below $4,500 now, which includes 16"
flatscreen color monitor, NFS networking software, internal hard disk,
modem software, windowing operating system, ethernet card, 16 megs of
memory, etc., etc., etc.
The only other things you would want would be a big SCSI hard disk and
SCSI tape backup (Exabyte preferred), and maybe a CD-ROM player. At
least with everything being SCSI, you can usually get identical
hardware to what you would get on your PC, most can just be moved
right over.
If you just use it as a file server, you could easily add 25 to 35
PC's to it, probably much more, but I haven't verified that myself
yet!
Once it is up and running, keeping it going is rather painless.
Anyone who understands getting 3DS hardware installed could easily
learn enough keep it going – it is as easy as Novell, and, from what
I see here, definately easier than Lantastic! If you have to learn a
network, why not make it a good one?
that sounds like a slick solution for the non-financially challenged.
As you know I'm so poor right now that I'm lucky to have 2 pennies to
rub together. When I came over to have those prints done the other
day, it caused major wallet anxiety. Fortunately, the prints had the
desired effect on the prospective client 🙂
I am able to do just what Greg stated with Lantastic 5.0. Only if I
push part of the network software into different segments of upper
memory do the other nodes complain.
Regards,
John Tissavary (La Luna cie)
so, you are saying that if a server goes down while slaves are
rendering and you reboot the server quick enough you won't get the
dreaded disconnection message? If that's true I may just need an
upgrade from Lantastic 4.1 to 5.0 which would be a bit more fiscally
feasible than buying a Sun server.
I think you'll be pleased. I've rebooted on several occasions and
there was no interruption in the slave renderers, as long as they
were busy thinking, and not trying to write a file to disk.
John,
sounds like time for an upgrade! 🙂
So David T. it appears there are trade-offs to each approach :-), but
I think the consensus is that for speed and overall reliability the
dedicated server approach is the best, and fastest. I think for the
time being however I will be happy to get three systems linked "peer
to peer" and try that first. If I have trouble with that there are
trouble-shooting advantages to the server approach as well. I think
that ultimately I will end up with a dedicated server. (I see the
little beasts grinning at me <G>) Thanks for the thought. -JE
No question, in my mind, that the dedicated server approach is most
stable. If there are enough things going on, even breaking functions
off to different servers is a *very* good idea. We have 4 servers. #1
acts as a bridge to a little LocalTalk network, a staging area for
files and supports SCSI DAT backup and optical archiving. #2 does
double duty as a gateway to an AS/400 and as our Postscript print
server (7 printers). #3 is a another file server, for our image
library and art dept. archive staging area. #4 is dedicated to
database and word processing functions only. The whole thing is
working quite well, too.
David – Sounds like where you work there's alot going on, and that
you've got a good network administrator. Thanks again 🙂 -JE