| 1 |
Am Sun, 19 Feb 2017 14:41:20 +0100 |
| 2 |
schrieb Meino.Cramer@×××.de: |
| 3 |
|
| 4 |
> Helmut Jarausch <jarausch@××××××.be> [17-02-19 14:04]: |
| 5 |
> > Hi, |
| 6 |
> > |
| 7 |
> > sometime I have some memory hungry ebuilds in the background, when |
| 8 |
> > I start (e.g.) Chromium which needs very much memory if you have a |
| 9 |
> > lot of open tabs. |
| 10 |
> > |
| 11 |
> > In that case my system nearly freezes. I cannot even kill chrome. |
| 12 |
> > What can I do in that case. (Remote login doesn't work either) |
| 13 |
> > |
| 14 |
> > Can I have any additional program (like Chromium) die if there is |
| 15 |
> > not enough memory. |
| 16 |
> > |
| 17 |
> > Many thanks for a hint, |
| 18 |
> > Helmut |
| 19 |
> > |
| 20 |
> |
| 21 |
> Hi Helmut, |
| 22 |
> |
| 23 |
> I know that situation very well...additionally I have Blender |
| 24 |
> open... |
| 25 |
> |
| 26 |
> But I think that the "freeze" of the system is not due to the memory |
| 27 |
> amount but due to the heavy I/O while swapping. |
| 28 |
|
| 29 |
Yes, and it's the small tiny write IO every application does every now |
| 30 |
and then which then blocks the application or even the whole OS for |
| 31 |
minutes. Because IO work-queues tend to be really big. |
| 32 |
|
| 33 |
> May be a tool like ionice could help you to keep the possibility |
| 34 |
> of killing certain processes. Ionice the emerge itself and additinally |
| 35 |
> nice it also. |
| 36 |
|
| 37 |
No, ionice cannot help here. If a process blocks due to exhausting |
| 38 |
dirty_background_bytes, it is blocked. Nothing will help to write back |
| 39 |
that data any earlier. And if even dirty_bytes is exhausted, the whole |
| 40 |
system blocks. |
| 41 |
|
| 42 |
I suggested another approach in my other reply: |
| 43 |
|
| 44 |
Lower dirty_background_bytes, maybe switch to deadline scheduler for |
| 45 |
better in-time servicing of write requests. |
| 46 |
|
| 47 |
> The emerge may take longer, but a frozen system is even slower... |
| 48 |
> ;) |
| 49 |
|
| 50 |
Not really, most of it can run from cache (if you allow for |
| 51 |
"swappiness"). It will have almost no impact on system performance if |
| 52 |
running in tmpfs. I'm having 16GB of RAM and allow portage to use 32GB |
| 53 |
of tmpfs - that means: Parts of ongoing big emerges WILL be swapped |
| 54 |
out. But since using deadline, this has almost no impact. Emerging is |
| 55 |
still lots faster than without tmpfs. |
| 56 |
|
| 57 |
-- |
| 58 |
Regards, |
| 59 |
Kai |
| 60 |
|
| 61 |
Replies to list-only preferred. |
Archives