Re: [lopsa-tech] linux and swapping

2012-09-01 Thread Robert Brockway
On Fri, 31 Aug 2012, Michael C Tiernan wrote: That's a very good explanation (I think I actually understand it better than I did before.) Thank you for that. However, now I'm going to switch back to our primary discussion and ask, how do I, the system administrator, who is charged with the op

Re: [lopsa-tech] linux and swapping

2012-09-01 Thread Michael Tiernan
On Fri, Aug 31, 2012 at 2:49 PM, Nathan Hruby wrote: [...] > Easy to say, not so easy to do. A lot of what Adam talks about is a > cultural issue [...] There's also a degree to which this culture? .. is almost part of the business model in a way. At my previous job, the decisions were made

Re: [lopsa-tech] linux and swapping

2012-08-31 Thread Josh Smift
ATW == Adam Tauno Williams ATW> When management says "we've decided to run XYZ". You can push to get ATW> things changes, but in the interim it is the admins job to make it go. ...but ideally, management should get the sysadmins involved before they commit to that decision. (But sometimes they

Re: [lopsa-tech] linux and swapping

2012-08-31 Thread Adam Tauno Williams
On Fri, 2012-08-31 at 11:49 -0700, Nathan Hruby wrote: > On Fri, Aug 31, 2012 at 11:27 AM, Dan Ritter wrote: > > On Fri, Aug 31, 2012 at 02:03:36PM -0400, Adam Tauno Williams wrote: > >> Ditto, this is the life of the System Administrator: here is the box, > >> here is the application, go... Aft

Re: [lopsa-tech] linux and swapping

2012-08-31 Thread Nathan Hruby
On Fri, Aug 31, 2012 at 11:27 AM, Dan Ritter wrote: > On Fri, Aug 31, 2012 at 02:03:36PM -0400, Adam Tauno Williams wrote: >> Ditto, this is the life of the System Administrator: here is the box, >> here is the application, go... After doing this for 20+ years the >> sermons about proper developm

Re: [lopsa-tech] linux and swapping

2012-08-31 Thread Dan Ritter
On Fri, Aug 31, 2012 at 02:03:36PM -0400, Adam Tauno Williams wrote: > Ditto, this is the life of the System Administrator: here is the box, > here is the application, go... After doing this for 20+ years the > sermons about proper development and design techniques get tired; it > may all be true

Re: [lopsa-tech] linux and swapping

2012-08-31 Thread Adam Tauno Williams
On Fri, 2012-08-31 at 07:31 -0400, Michael C Tiernan wrote: > - Original Message - > > From: "Edward Ned Harvey (lopser)" > > If you have an application that requires more ram than you physically > > have in the system... > However, now I'm going to switch back to our primary discussion a

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Edward Ned Harvey (lopser)
> From: Michael C Tiernan [mailto:michael.tier...@gmail.com] > > - Original Message - > > From: "Edward Ned Harvey (lopser)" > > > Why couldn't you memory map? > > Ok, I'll assume the mantle of the least smartest and least experienced guy in > the room. > How would you resolve the issue

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Narayan Desai
We've done OK with the HP quad socket DL580G7's. They have 64 dimm sockets, so we can stick with 16s and get 1TB. The big issue is the MTBF on dimms; infant mortality was a real pain on these. Aside from that, they've been pretty good. -nld On Aug 29, 2012, at 10:33 AM, Doug Hughes wrote: > On

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Doug Hughes
On 8/29/2012 8:57 AM, Edward Ned Harvey (lopser) wrote: I happen to support a lot of high mem high performance jobs, including EDA tools, with sims that will consume 100G+ memory... I haven't yet seen the situation where we deemed the cost of the hardware upgrade to be the most significant fact

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Doug Hughes
On 8/29/2012 7:54 AM, Edward Ned Harvey (lopser) wrote: From: Travis [mailto:hcoy...@ghostar.org] Never is such a harsh word. The point is not all use-cases can be simply solved by physically adding more RAM to a system or by mmaping. Why couldn't you memory map? vendor binary, for one. _

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Edward Ned Harvey (lopser)
> From: tech-boun...@lists.lopsa.org [mailto:tech-boun...@lists.lopsa.org] > On Behalf Of Edward Ned Harvey (lopser) > > OMG, holy crap. I don't care how big your processes are, or how much > memory you have in your system, you should never be swapping active > memory. You can always solve this

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Michael C Tiernan
- Original Message - > From: "Edward Ned Harvey (lopser)" > Why couldn't you memory map? Ok, I'll assume the mantle of the least smartest and least experienced guy in the room. How would you resolve the issue by memory mapping? -- << MCT >> Michael C Tiernan. Is God a perform

Re: [lopsa-tech] linux and swapping

2012-08-29 Thread Edward Ned Harvey (lopser)
> From: Travis [mailto:hcoy...@ghostar.org] > > On Mon, Aug 27, 2012 at 9:41 PM, Edward Ned Harvey (lopser) > wrote: > > > > OMG, holy crap. I don't care how big your processes are, or how much > memory you have in your system, you should never be swapping active > memory. You can always solve

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Andrew Hume
good point. it is weird i didn't pick this up, as i certainly remember when my early systems (Cray and CDC 6600) did real swapping. unix swapped, if i recall correctly, until the VM systems (like Vax Unix) appeared. On Aug 28, 2012, at 9:18 PM, Leon Towns-von Stauber wrote: > > On Aug 28, 2012,

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Leon Towns-von Stauber
On Aug 28, 2012, at 6:09 AM, Jack Coats wrote: > Once virtual memory was being used we 'paged'. Read/wrote one page at > a time. Shortly after we started using swapping to take a 'set of > pages', normally defined by a 'reasonable buffer' for the underlying > device. This was especially true a

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Skylar Thompson
On 08/28/12 06:59, Brandon Allbery wrote: > On Tue, Aug 28, 2012 at 12:31 AM, Skylar Thompson > mailto:skylar.thomp...@gmail.com>> wrote: > > It can be a real problem for latency-sensitive applications that are > cohabiting on a system that's also doing heavy I/O. For instance, > IBM's

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Marc David Rovner
On Tue, Aug 28, 2012 at 04:43:16PM -0500, Jack Coats wrote: > It would be nice if someone could update this allegory to modern > machines, and include some SAN and/or NAS type access times, > especially since we don't 'do tape' much anymore. Funny you should say that, as I just looking at this tod

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Dan Ritter
On Tue, Aug 28, 2012 at 04:43:16PM -0500, Jack Coats wrote: > It would be nice if someone could update this allegory to modern > machines, and include some SAN and/or NAS type access times, > especially since we don't 'do tape' much anymore. https://gist.github.com/2843375 Look at the 1 billion t

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Yves Dorfsman
On 2012-08-28 15:43, Jack Coats wrote: to my college's Chemistry Dept, and we fixed the logic by replacing geranium based transistors that were 2 flip-flops per circuit board (about 2.5x2.5" each). 48bit words, and most instructions were 'half Truly, the first organic computer! Too sad we then

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Jack Coats
Only slightly off topic, Back when mainframes roamed the earth, I was a systems programmer on IBM VM systems for an oil company. One of the performance and tuning trivia that went around went something like this (my memory is probably wrong, but you can get the idea. Assume 1 cpu cycle takes one

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Michael C Tiernan
- Original Message - > From: "Travis" > Never is such a harsh word. > The point is not all use-cases can be simply solved by physically > adding more RAM to a system or by mmaping. I agree with that. Every time these conversations come up I am always amazed at how many times "Never" an

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread bergman
In the message dated: Tue, 28 Aug 2012 02:41:51 -, The pithy ruminations from "Edward Ned Harvey (lopser)" on were: => > From: tech-boun...@lists.lopsa.org [mailto:tech-boun...@lists.lopsa.org] => > On Behalf Of Andrew Hume => > => > err, that is the whole point. => > SwapTotal - SwapFree re

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Doug Hughes
On Aug 28, 2012 11:35 AM, "Travis" wrote: > > On Mon, Aug 27, 2012 at 9:41 PM, Edward Ned Harvey (lopser) > wrote: > > > > OMG, holy crap. I don't care how big your processes are, or how much memory you have in your system, you should never be swapping active memory. You can always solve this p

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Travis
On Mon, Aug 27, 2012 at 9:41 PM, Edward Ned Harvey (lopser) wrote: > > OMG, holy crap. I don't care how big your processes are, or how much memory > you have in your system, you should never be swapping active memory. You can > always solve this problem by either adding more memory, or using a

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Andrew Hume
jack presents a reasonable summary of things. i still believe there is a "bug/feature" with how linux reports swap usage. i was looking at one on the weekend where it claimed 7GB of swap use, and the total VM of all running processes was significantly less than this. nevertheless, i have my answe

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Brandon Allbery
On Tue, Aug 28, 2012 at 12:31 AM, Skylar Thompson wrote: > It can be a real problem for latency-sensitive applications that are > cohabiting on a system that's also doing heavy I/O. For instance, > IBM's TSM database process (often consuming 75% of the physical memory > of the machine for indices

Re: [lopsa-tech] linux and swapping

2012-08-28 Thread Jack Coats
I think the TSM example is what is showing the strength of the architecture. Every computer load (mix of programs used for your application) have a different need for memory. My first home UNIX system was a non-virtual memory system, so EVERYTHING stayed in memory, and on a 386 with under 1G ram

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Yves Dorfsman
On 2012-08-27 22:31, Skylar Thompson wrote: Why exactly do you want to waste RAM on dead code/data? It served its purpose, it should get out of the way and let something productive use the RAM. It can be a real problem for latency-sensitive applications that are cohabiting on a system that's

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Robert Hajime Lanning
On 08/27/12 21:56, Yves Dorfsman wrote: Desktop: With swappiness = 0: You can have a whole bunch of apps started, a bunch of xtrem with scripts I am working on in one pane, some spreadsheet I update once or twice a day in another pane, my email client in a third one, a couple of browser windows i

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Yves Dorfsman
On 2012-08-27 22:25, Brandon Allbery wrote: On Tue, Aug 28, 2012 at 12:20 AM, Yves Dorfsman mailto:y...@zioup.com>> wrote: I for one, totally disagree with your statement, I do not want any of my process to be pushed to swap to just buy some buffer space. If I have an I/O issue, I'll

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Skylar Thompson
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 08/27/2012 09:25 PM, Brandon Allbery wrote: > On Tue, Aug 28, 2012 at 12:20 AM, Yves Dorfsman > wrote: > >> I for one, totally disagree with your statement, I do not want >> any of my process to be pushed to swap to just buy some buffer >> space.

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Brandon Allbery
On Tue, Aug 28, 2012 at 12:20 AM, Yves Dorfsman wrote: > I for one, totally disagree with your statement, I do not want any of my > process to be pushed to swap to just buy some buffer space. If I have an > I/O issue, I'll look into it. This is why I set swappiness to zero on > Linux, and I sure

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Yves Dorfsman
On 2012-08-27 20:41, Edward Ned Harvey (lopser) wrote: Here is what swap is meant for: At every opportunity, the kernel will grow the system buffer & cache to consume all physical memory in the system. It is normal to see near-zero "free" memory in the system, provided that you have a large c

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Brandon Allbery
On Mon, Aug 27, 2012 at 10:41 PM, Edward Ned Harvey (lopser) < lop...@nedharvey.com> wrote: > memory? It is normal for some processes to sit almost completely idle for > the life of the computer. Or some process dies in a zombie state, or > whatever. Swap is useful, so the kernel Zombie proce

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Edward Ned Harvey (lopser)
> From: tech-boun...@lists.lopsa.org [mailto:tech-boun...@lists.lopsa.org] > On Behalf Of Andrew Hume > > err, that is the whole point. > SwapTotal - SwapFree reflects the high water mark problem. > for example, if we are swapping, and process P1 has 20GB of swap, > then SwapTotal - SwapFree = 20G

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread david
The thing is that what is happening is a little different from what you are thinking you run out of physical ram and P1 has cause you to use an additional 20GB of swap space. That swap space is very unlikely to contain any address space allocated to process P1, it's almost always going to be a

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Matt Simmons
AFAIK, the "high water mark" thing in Linux swap is because the dirty pages in memory still haven't been written or are in use. When a page of memory is swapped to disk, it's because that memory is in use, and has changed from the on-disk data (or has been generated, or whatever), it can't just be

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Andrew Hume
err, that is the whole point. SwapTotal - SwapFree reflects the high water mark problem. for example, if we are swapping, and process P1 has 20GB of swap, then SwapTotal - SwapFree = 20GB. if i start a new process P2 which uses 3GB swap, then SwapTotal - SwapFree = 23GB. Great! process P1 now exits

Re: [lopsa-tech] linux and swapping

2012-08-27 Thread Dan Ritter
On Mon, Aug 27, 2012 at 06:49:28AM -0700, Andrew Hume wrote: > i have some colleagues who are being frustrated by the stupid way > Linux measures swap space consumption (the high water mark > of currently running processes). > > does anyone know of a way to measure how much swap space is actually