Excessive Pageouts! Or not, you tell me...

bergz

My immediate question is this: If I'm getting a whole bunch of pageouts, but still have a whole hell of a lot of System Memory inactive, are the excessive pageouts affecting my performance at all?





I don't know what's causing it at this point. On the one hand I can be surfing with Camino with nothing else running but quicksilver and menucalendarclock, and I can get 20-100 pageouts. Then I open Photoshop, bits on wheels, toast, ical, unreal, itunes, outlook, classic and mail.app AT ONCE and I get none. Nada.



I've got a 1.67 PB with 1.5 GB RAM. (512 Apple + 1 GB Kingston which memtest says is "ok" all around)



I've got 8 GB VM at the moment.



Half of my system memory is inactive.



And yet I've got 7000 pageouts after restarting an hour ago. All I've got running now is Bits on Wheels, Activity Monitor, Camino and iTunes.



I've found that MPlayer can give me 1000 pageouts in 90 seconds. Other than that I haven't been able to pin it down to anything in particular.



Thanks for any clarity you might be able to provide.



--B

ipodandimac

just restart your computer from time to time.

bergz

Thanks for that, as always. I was just wondering if someone who knows more about swapping and such could tell me what kind of a performance hit it would represent.



I mean, if I've got all that inactive memory, I suppose that up to a certain point I shouldn't notice any hit, no? But then again, I don't know much about it.



--B

kickaha

Well 7000 in an hour is only about 100 a minute. Considering that each pageout is measured in microseconds...



Yes, it is a speed hit. A very very small one.



Sounds to me like the apps in question are repeatedly hammering on the same memory chunklet (which could be a file mapping) in a repeated pattern.



Have you been able to narrow it down to any *single* apps other than MPlayer? (And in that case are you sure it's not page*ins* you're seeing as it maps the file into VM a chunk at a time?)

bergz

Quote:

Originally posted by Kickaha

Sounds to me like the apps in question are repeatedly hammering on the same memory chunklet (which could be a file mapping) in a repeated pattern.



Have you been able to narrow it down to any *single* apps other than MPlayer? (And in that case are you sure it's not page*ins* you're seeing as it maps the file into VM a chunk at a time?)

Thanks for the response, Kix.



I'll do more experimentation this weekend, playing with particular apps, but the only thing that I can think of are the tracker errors that Bits on Wheels was getting (a bunch of trackers still don't support it). Only program running last night and I got > 71,000 (and I didn't sleep long, either).







--B

hiro

This is definitely odd behavior. My best guess would go something like this: Your current version of MPlayer (or whichever is the responsible app) has some sort of memory leak in an area that has a process it shuts down on occasion, so the memory doesn't stay perpetually leaked.



If that was the case, on occasion with the leak you would see your free memory go down lower than the 160MB or so you show in your last post. Someplace in that neighborhood OS X trips on a threshold which tells the memory management logic to clean out the free list. OS X uses lazy methods here keeping a list of pages it can page out but doesn't actually move them until the free space goes somewhere less than 10% installed. So the leak causes the page-outs, free list gets bigger, the page-outs stop and then by some completely unrelated reason, the leaking process is terminated and the memory it grabbed is returned to the OS which puts it on the free list as well. Now you just cycled a lot of nothing all because there is some errant process that happens to be playing when your used memory is near that magic clean out the free list threshold.



A couple full top readouts might be a little more help in determining what's going on (one with MPlayer, one without), but it definitely isn't what should be happening. Only modified pages should be paged out and playing a media file shouldn't modify enough pages to notice anything. But a memory leak can create a whole shitload of unnecessary modified pages in a downright hurry.

bergz

Thanks for the extensive reply. My brain's doing pageouts just trying to understand it.



Here are some tops. I'm having trouble reproducing the phenomenon. The most reliable apps that give me immediate pageout surges are: Limewire, Azureus, eMule, and Bits on Wheels.



However, when these apps are not running, installers, opening disk images, mplayer, or other programs *especially upon closing them* nudge up the pageouts a bit. I just opened the VPC installer and it caused about 100 pageouts by opening it. I just opened it again twice to try to grab a top screenshot for you and there were no pageouts.



Here's a top with a bunch of apps open. I'm racking up pageouts.









Here I've just got Limewire. Still racking them up.







Here's one without.









What do you think?



--B

kickaha

I think you are *WAY* overanalyzing this.



Right now, my pagein/pagouts are nearly 1:1 at around 480k, after a few days uptime, with pageins slightly higher. Granted, I usually run a solid <10MB free RAM.

bergz

Quote:

Originally posted by Kickaha

I think you are *WAY* overanalyzing this.



Right now, my pagein/pagouts are nearly 1:1 at around 480k, after a few days uptime, with pageins slightly higher. Granted, I usually run a solid <10MB free RAM.

I definitely agree with you. Only thing is while checking out what apps cause them etc. I crashed my machine and lost my camino bookmarks. See this desperate thread...



I'm on another machine afraid to touch anything. I really don't want to lose them. Any recovery techniques would be greatly appreciated.



--B

dobby

Your pageins are fine.

Here is a couple of examples of various macs we have. (using vm_stat from terminal)

Dual 1.25 2GB mem running as Distiller server uptime 28 days

Mach Virtual Memory Statistics: (page size of 4096 bytes)

Pages free: 267890.

Pages active: 28039.

Pages inactive: 65716.

Pages wired down: 31571.

"Translation faults": 563519048.

Pages copy-on-write: 34705151.

Pages zero filled: 55987828.

Pages reactivated: 0.

Pageins: 30463.

Pageouts: 48367.

Object cache: 11132243 hits of 11919867 lookups (93% hit rate)



XServe Dual 1.33 2GB mem uptime 198 days but sod all running.

Mach Virtual Memory Statistics: (page size of 4096 bytes)

Pages free: 4128123.

Pages active: 513994.

Pages inactive: 202162.

Pages wired down: 316009.

"Translation faults": 2664025863.

Pages copy-on-write: 255541912.

Pages zero filled: 840392709.

Pages reactivated: 0.

Pageins: 313208.

Pageouts: 17724.

Object cache: 124094671 hits of 128223471 lookups (96% hit rate)

G5 Dual 1.8 2GB mem running 8x java apps, OPI Server and 80 odd user connections.

Mach Virtual Memory Statistics: (page size of 4096 bytes)

Pages free: 267797.

Pages active: 49345.

Pages inactive: 158837.

Pages wired down: 48309.

"Translation faults": 1138949951.

Pages copy-on-write: 35097152.

Pages zero filled: 99918974.

Pages reactivated: 10860250.

Pageins: 84703371.

Pageouts: 398466.

Object cache: 16705401 hits of 33398547 lookups (50% hit rate)



As you can see a big varity of usage.

No you dont need to reboot your machine. Unix knows how to manage memory unlike older OS9/Windows OSs.

Notice our distiller server has more pageouts than ins. Performance is still fine.



Dobby.

bergz

Thanks for the analyses, guys. Not going to worry about it.



And thanks for no one telling me to repair permissions



--B