HPC
CPU and RAM hogs overstaying their welcome? Here's a fix
Multicore processors drive everything these days from the biggest HPC cluster to the lowliest tablet – even smartphones. While parallel programming has come quite a way, there are still many apps that aren’t well-behaved at all. They’re the worst kind of guests – acting like they own the whole damned house while paying …
This topic is closed for new posts.
Heh....
I'd like to see how it handles a minecraft server with a handful of users on it on different sides of a fairly large world- Minecraft's server code is notoriously BAD with memory and processor management. (either that, or the java VM it's running in has that problem...
Re: Heh....
Tribes / Tribes 2 had the sweetest network code for gaming in 2001 .. that became torque game engine ... there's still this open source version that hasn't been updated since 2005
http://sourceforge.net/projects/opentnl/
forced all the other games and sims to go with client server software for online play, and often give the server with the game .. 128 player server software is pretty standard now depending on how much action is going on to be pushed to all the clients .. single socket server and a big enough data pipe with minimum packet loss is all that's required
I have a better idea
Why don't we get the main user interface and rearange the letters into the least common used layout making the user alot slower in doing common tasks. Beyond that the operating systems of the world I live in allow you to control priority of users, quota's etc. Been that way since java was a wee twinkle in the minds eye.
That all said there is no such thing as a bad user, only bad programs and bad pro-grammers(sic). This is not the solution you are seaking. Worst case redesign your charging models to accomodate this untapped source of income, what next - limited to one chicken nugget per order. Roll with it and profit, and learn to admin you man, learn to admin.
Re: I have a better idea
Does sound a lot like Solaris' resource management, doesn't it?
Re: I have a better idea
Damn, you beat me: this all sounds a lot like Solaris zones.
Solaris Zones
Similarities to Zones, but it's dynamic, so if a containers owner i snot using some of the resource another user can "borrow" that resource, so maximizing system utilization.
So let me get this right ...
... they've written a script to automate "nice" and "kill"?
Re: So let me get this right ...
Gotta be better than the OOM Killier, aka "self-inflicted DoS attack".
Re: So let me get this right ...
> Gotta be better than the OOM Killier, aka "self-inflicted DoS attack".
If you are out of virtual memory as well as real memory, you're probably thrashing or something has gone badly wrong. You are probably I/O bound and nothing much is happening anyway.
People pay big money to VMWare to prevent apps from bringing down the entire host. This is the sort of resource management which should be in the base OS.
However, I'm not sure that the complexity of managing resourcing is cheaper than putting tasks on their own physical hosts. Time to bring on the ARM server chips & blades!
Re: So let me get this right ...
... are you saying they've recreated a shell version of sysadmin doom ?
Memory bloat...
..is often a web browser in our world. Why, oh, why does a web browser need GB of memory to keep a dozen or two tabs open?
Memo to Mozilla et al - please fix you damn leaky code and stop buggering around with version numbers and GUI changes.
Re: Memory bloat... <--THIS
2 GB RAM, 4-core CPU, one instance of FF12 (visibly) running, one tab open.
PID USER PRI NI VIRT RES SHR S CPU% MEM% TIME+ Command
27341 user 20 0 847M 146M 40680 S 0.0 7.4 0:43.10 /usr/lib/firefox/firefox -no-remote -P SARA
27416 user 20 0 272M 23676 16328 S 0.0 1.2 0:00.05 /usr/lib/firefox/plugin-container /usr/lib/flashplugin-installer/libflashplayer.so -greomni /usr/lib/fire
27417 user 20 0 272M 23676 16328 S 0.0 1.2 0:00.00 /usr/lib/firefox/plugin-container /usr/lib/flashplugin-installer/libflashplayer.so -greomni /usr/lib/fire
27415 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27410 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27372 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.03 /usr/lib/firefox/firefox -no-remote -P SARA
27371 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27367 user 20 0 847M 146M 40680 S 0.0 7.4 0:02.85 /usr/lib/firefox/firefox -no-remote -P SARA
27364 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.04 /usr/lib/firefox/firefox -no-remote -P SARA
27363 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27361 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27360 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.01 /usr/lib/firefox/firefox -no-remote -P SARA
27359 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.03 /usr/lib/firefox/firefox -no-remote -P SARA
27358 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.01 /usr/lib/firefox/firefox -no-remote -P SARA
27355 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.04 /usr/lib/firefox/firefox -no-remote -P SARA
27353 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.81 /usr/lib/firefox/firefox -no-remote -P SARA
27349 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.02 /usr/lib/firefox/firefox -no-remote -P SARA
27348 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.16 /usr/lib/firefox/firefox -no-remote -P SARA
27347 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.00 /usr/lib/firefox/firefox -no-remote -P SARA
27346 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.18 /usr/lib/firefox/firefox -no-remote -P SARA
27345 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.25 /usr/lib/firefox/firefox -no-remote -P SARA
27344 user 20 0 847M 146M 40680 S 0.0 7.4 0:00.01 /usr/lib/firefox/firefox -no-remote -P SARA
Re: Memory bloat... <--THIS
Don't know how you managed that, FF never seems to use that much memory for me on either Linix or Windows. So, for example, I've got this and four other tabs open and I'm using 124MB.
Re: Memory bloat... <--THIS
> I've got this and four other tabs open and I'm using 124MB
I can beat that. My Windows Firefox process right now is holding 807 MB of virtual memory - but it has 101 tabs open (in three windows) for an average of just under 8 MB per tab. Your FF is averaging nearly 25 MB per tab.
Sure, at the moment Firefox has the biggest memory footprint of any single process running on the system (though it wouldn't if I had any VMs up, or if I were running a large build or test suite). But considering what I'm having it do at the moment it's not unreasonable.
As a point of comparison, Venomous Studio is taking a cool 457 MB, for one 48-project solution with a dozen open source tabs.
Re: 101 tabs?
You win, I normally have about 20-30 open tabs, and use about 400-500mb of RAM doing so. Everyone thinks I'm crazy and ADD for the 20 I do have, thanks for making me feel normal.
Here's another cheaper fix...
At a handy shell prompt, type:
killall -9 java
Re: Here's another cheaper fix...
Might want to make that:
sudo killall -9 java
If there's really a problem...
fix the apps causing the problem or fix the job scheduler in the kernel, don't introduce a layer of bloat which simply tries to deal with the symptoms.
It's worse than that, Jim!
Don't introduce a layer of bloat which gives code-monkeys another excuse to do sloppy coding.
Pricing?
Timeout is pretty good and free. I cannot see the slightest hint on pricing on the website so I'm assuming it costs the same as buying another server, kinda self-defeating if that's the case.
Especially as most of what this does can be achieved with a small bash script (or the aforementionned timeout).
t they don’t have the ability to completely control and prioritize system resources
Excuse me - isn't that the entire point of any half-decent OS?
Is this driven by GPL fears?
Is a part of the reason for this a reluctance to entangle a product with the GPL license used for the kernel?
lxc on steroids?
lxc seems to be able to do a lot of what's going on here. I must admit the idea of being able to control badly-behaved programs and keep them from bringing the machine to its knees is quite appealing.
I can think of a few large applications suites that could do with some serious controlling.
This topic is closed for new posts.
