On Wed, Jan 18, 2017 at 3:36 PM, Christopher Myers wrote:

Weird...what does free say? Just curious because we've got a fileserver with 64GB of memory, and it'll use maybe 1-2GB for actual applications, and then turn the rest of it into cache. So, top will say it's got like 700M free, but it's basically all just used as buffers and cache.

==== This machine was booted within the hour. Running openSUSE 13.1 from a boot DVD. I've now added a 200GB swapfile (plus 128GB or RAM). I'm not sure I trust "top"s numbers. It says they are in KiB, but the math doesn't work. It says I have a total of 1461804 KiB RAM and 33554428 KiB SWAP. Reality should be 128GB of RAM and 200GB of swap. top shows "87368 free" for Mem and "227284 used" for swap. In theory that is 87 MB free and 227 MB of swap in use. I think is is closer to 8.8 GB of free RAM with about 23 GB of swap in use. Greg -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On Wed, Jan 18, 2017 at 4:01 PM, Christopher Myers wrote:

...

...

...

Greg Freemyer 01/18/17 2:56 PM >>> top shows "87368 free" for Mem and "227284 used" for swap.

In theory that is 87 MB free and 227 MB of swap in use.

I think is is closer to 8.8 GB of free RAM with about 23 GB of swap in use.

Sorry, by "free", I meant the free command, eg. -

cmyers@chrismyers:/> free -h total used free shared buffers cached Mem: 7.6G 6.9G 693M 125M 79M 1.0G -/+ buffers/cache: 5.8G 1.7G Swap: 2.0G 130M 1.9G

My mistake: ============ total used free shared buffers cached Mem: 1.4G 1.3G 93M 2.3M 1.8M 245M -/+ buffers/cache: 1.1G 340M Swap: 31G 217M 31G ============ That really makes no sense. Maybe my hardware is too new? But the swap file should definately be 200GB even if the RAM is wrong. I used fallocate to create the 200GB swapfile, mkswap to initialize it and swapon to put it in use. Unfortunately, I have to boot this machine back to Windows for the next 18 or so hours. Greg -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

...

...

"Carlos E. R." 01/18/17 3:33 PM >>> On 2017-01-18 22:26, Greg Freemyer wrote:

You are not using 32bit 13.1, per chance?

Ah, good thought -- That's what I'm running, but with a different kernel, and it's seeing my extra memory, so maybe a difference from the kernels? -- cmyers@chrismyers:/> cat /etc/SuSE-release openSUSE 13.1 (i586) VERSION = 13.1 CODENAME = Bottle # /etc/SuSE-release is deprecated and will be removed in the future, use /etc/os-release instead cmyers@chrismyers:/> uname -a Linux chrismyers.millikin.edu 3.12.62-52-desktop #1 SMP PREEMPT Wed Aug 10 18:37:26 UTC 2016 (069dc39) i686 i686 i386 GNU/Linux cmyers@chrismyers:/> -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

Carlos E. R. wrote:

On 2017-01-18 22:26, Greg Freemyer wrote:

...

My mistake:

============ total used free shared buffers cached Mem: 1.4G 1.3G 93M 2.3M 1.8M 245M -/+ buffers/cache: 1.1G 340M Swap: 31G 217M 31G ============

That really makes no sense. Maybe my hardware is too new?

It clearly is not finding all the memory. You need to look at the log of the booting.

You are not using 32bit 13.1, per chance?

With that size memory, it would be running a PAE kernel. -- Per Jessen, Zürich (-5.4°C) http://www.cloudsuisse.com/ - your owncloud, hosted in Switzerland. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/19/2017 02:49 AM, Per Jessen wrote:

Carlos E. R. wrote:

...

On 2017-01-18 22:26, Greg Freemyer wrote:

...

My mistake:

============ total used free shared buffers cached Mem: 1.4G 1.3G 93M 2.3M 1.8M 245M -/+ buffers/cache: 1.1G 340M Swap: 31G 217M 31G ============

That really makes no sense. Maybe my hardware is too new?

It clearly is not finding all the memory. You need to look at the log of the booting.

You are not using 32bit 13.1, per chance?

With that size memory, it would be running a PAE kernel.

OMG! The PAE mechanism involves a mapping table for the 32-bit system to address more than 4G. I'll leave aside the issue of why anyone might need for than 4G when many of us run excellent systems in that or less. I believe Per Jensen mentioned a server that had been up for over 4 years with just 2G. I'm sure that more than 4G I have in my 64-bit desktop would be nice for my photo and video editing, but realistically I know that the real limitation is the framebuffer and rendering speed/capability of my graphics system. I need to throw money/technology at that first even though memory is cheaper, but to throw more memory I'd need a new mobo as well, so in reality a new GPU would be cheaper overall. Santa was kind in the camera department this year, not the computer department. If I were corporate and the LAN/SAN were the limit I'd see about gigabit networking or optical networking. But then again, all my machines, except for a few from the Closet of Anxieties with 800Mhz 32-bit/1GRAM are 64-bit. Yes there was always the argument that for a given RAM speed and datapath (remember the 8088, the 8086 with an 8-bit data path for 'backward compatibility' and the need for fewer TTL support components, which was why it was chosen for the original IBM-PC? well by the time it actually got into production the whole microprocessor/TTL landscape and pricing had changed to invalidate the original justification) that the smaller opcodes and reduced fetch time and the fact that most loops were small, requiring only an 8-bit counter and such blitherations meant that upgrade wasn't worth it? Heck,, I remember in the 1960 hearing a guy argue that stereo hi-fi wasn't worth it, the extra cost of components for the second channel, the extra loudspeaker, the more complicated pick-up on the record player, the more complex recording/mixing equipment, it was all a sale conspiracy to get us to spend more money. The along come LPs, cassettes and DVDs, each invalidating the previous technology ... But "-pae"? Even so, a processes address space remains at 32-bits, meaning it can only access a maximum of 4GB of memory. The OS however can access a 64GB address space, allocating 4GB chunks to processes. This is done via a mapping table. The size of the mapping table is going to depend on the actual amount of physical memory. so the more physical memory the bigger that table is. And that bale has to be in the space under the 4G boundary, in kernel space. I'm not sure I like the way this is going. If this were critical, if this were corporate, I'd seriously look at getting a new mobo and a 64-but processor. As a manager, I look at this way: the cost of time (that is, salary or consultant's fee) flutzing around with this, ongoing, compared to the one-time cost of a new mobo and CPU and the one-time cost of installing the same will almost certainly work out in favour of the latter. At home, on a hobby budget, things might be different. The hobby budget is competing with other household and family expenses. -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

That is definitely odd... Maybe try free -g ? On my laptop with 8G memory and 1G swap: cmyers@chrismyers:/> free -g total used free shared buffers cached Mem: 7 7 0 0 0 0 -/+ buffers/cache: 5 1 Swap: 1 0 1 On our ODA with 192G and 18G swap: [cmyers@muoda1 ~]$ free -g total used free shared buffers cached Mem: 192 144 47 0 2 26 -/+ buffers/cache: 116 76 Swap: 18 0 18 -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/18/2017 03:36 PM, Christopher Myers wrote:

Weird...what does free say? Just curious because we've got a fileserver with 64GB of memory, and it'll use maybe 1-2GB for actual applications, and then turn the rest of it into cache.

Well, lets face it, the original UNIX was designed for short-lived small programs, in sharp contrast to the mainframes of the day that ran huge, long lived monolithic programs that did everything all in one such as CICS The 'each thing does one thing and only one thing" ... and then disappears .. was the attitude that differentiated UNIX back then. Well, OK the shell (of that day) was logn lived, but it was basically a dispatcher and was much, much smaller than our current BASH. A file server, not least of all a "black box" NAS, isn't running an application suite as is a desktop. It isn't exactly running the huge monolithic CICS, but there isn't the 'churn' in the code space. Perhaps the turnover ion the file management in the kernel and the buffer space counts for something, but how much of NFS is in user space and how much in the kernel?

With that said -- my server at home routinely "runs out of memory" after a few months of uptime, even though it has plenty of memory free. I've always wondered if it maybe has issues with memory fragmentation or something like that(?)

Basically, Linux discovers all memory and tries to use it. I'm not sure that running a cache application in user space helps with memory management. I *do* know that the VM system is highly tunable and that out-of-the-box linux is tuned for the desktop. Somehow you need to tell it to use more space in the kernel. I'm not sure what you mean by "turning the rest of it into cache" without some more specifics. is that a cache application? *HOW* are you "turning it" into cache? https://lonesysadmin.net/2013/12/22/better-linux-disk-caching-performance-vm... -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/18/2017 08:40 PM, Carlos E. R. wrote:

On 2017-01-19 01:11, Anton Aylward wrote:

...

I'm not sure what you mean by "turning the rest of it into cache" without some more specifics. is that a cache application? *HOW* are you "turning it" into cache?

That is what the Linux kernel does. You don't need to do anything to activate. It just happens.

That's what I thought. Never the less, for a file server I can imagine that you'd want to tune the vm characteristics, the number of NFS threads, and even, perhaps, the network MTU differently from a desktop. And since this is a NAS on the LAN, you can probably alter the rsize and wsize (provided you don't exceed the network MTU). Since this is a NAS, you probably want to alter the network buffer queue size as well, the kernel won't normally allocate enough memory there. Of course anything that you modify in /proc file system is temporary, because its the value that's stored in the RAM, which does not persist across reboots. You can make these entries permanent across reboots by making an entry in sysctl.conf. -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/19/2017 02:47 AM, Per Jessen wrote:

Anton Aylward wrote:

[big snip]

NFSv4 automagically does that.

Any other advantages/disadvantages to V4? What about crossing network boundaries, routers, bridges on a more extended corporate LAN? How does it compare to CIFS? -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On Thu, Jan 19, 2017 at 5:50 PM, Per Jessen wrote:

...

What about crossing network boundaries, routers, bridges on a more extended corporate LAN?

I don't know if there are any problems, I wouldn't expect any. It's just plain TCP traffic.

NFS v3 and below are using dynamic ports and several auxiliary protocols running over separate ports as well. So it was rather challenging to use across firewall/NAT. NFS v4 consolidates protocols and is using fixed port number. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 19/01/17 14:58, Carlos E. R. wrote:

On 2017-01-19 15:54, Andrei Borzenkov wrote:

...

On Thu, Jan 19, 2017 at 5:50 PM, Per Jessen wrote:

...

...

What about crossing network boundaries, routers, bridges on a more extended corporate LAN?

I don't know if there are any problems, I wouldn't expect any. It's just plain TCP traffic.

NFS v3 and below are using dynamic ports and several auxiliary protocols running over separate ports as well. So it was rather challenging to use across firewall/NAT. NFS v4 consolidates protocols and is using fixed port number.

Oh. This is very interesting.

Could you point to some info link on how to configure the suse firewall for NFS v4?

I have some problems with NFS not working after upgrading to 42.2 and this would help to review the configs.

I would have thought YaST could do that for you in the Allowed Services tab. Basically, you need to open 111 and 2049 for both tcp and udp. HTH Bob -- Bob Williams System: Linux 4.4.36-8-default Distro: openSUSE 42.2 (x86_64) Desktop: KDE Frameworks: 5.26.0, Qt: 5.6.1 and Plasma: 5.8.2 -- Bob Williams System: Linux 4.4.36-8-default Distro: openSUSE 42.2 (x86_64) Desktop: KDE Frameworks: 5.26.0, Qt: 5.6.1 and Plasma: 5.8.2

Andrei Borzenkov wrote:

On Thu, Jan 19, 2017 at 5:50 PM, Per Jessen wrote:

...

...

What about crossing network boundaries, routers, bridges on a more extended corporate LAN?

I don't know if there are any problems, I wouldn't expect any. It's just plain TCP traffic.

NFS v3 and below are using dynamic ports and several auxiliary protocols running over separate ports as well. So it was rather challenging to use across firewall/NAT.

I can imagine that - I've never had reason to do it, we've sometimes used NFS over a VPN, that's about as "advanced" it's been :-) -- Per Jessen, Zürich (-3.8°C) http://www.hostsuisse.com/ - dedicated server rental in Switzerland. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/18/2017 05:50 PM, Patrick Shanahan wrote:

why have swap at all if you are going to have it in ram. Just limits your system's usage of free ram. I have 36g ram and *no* swap and no noticible need for it.

Yes, but what is your application suite? -- A: Yes. > Q: Are you sure? >> A: Because it reverses the logical flow of conversation. >>> Q: Why is top posting frowned upon? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/18/2017 02:50 PM, Patrick Shanahan wrote:

* jdd [01-18-17 16:55]:

...

Le 18/01/2017 à 21:23, Greg Freemyer a écrit :

...

Setting up swap space now!

make it in ram :-) why have swap at all if you are going to have it in ram.

Whoosh Patrick. Whoosh. Smiley face and all. -- After all is said and done, more is said than done. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On 01/18/2017 04:55 PM, jdd wrote:

Le 18/01/2017 à 21:23, Greg Freemyer a écrit :

...

Setting up swap space now!

make it in ram :-)

Or /dev/fd0. ;-) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

You can stop it from swapping by setting swappiness to zero. sysctl vm.swappiness=0 Then it will only swap if it runs out of memory. I just thought I would throw that into the mix in case no one has mentioned it already. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org

On Mon, Feb 20, 2017 at 5:59 AM, Peter Suetterlin wrote:

Hi,

I'm late on this (been on vacation :D), so sorry if I repeat stuf that has been mentioned already. But as I had a similar issue half a year ago I thought I might throw the word 'slab' in. In my case it had been a berzerk process acquiring more and more memory from the kernel slab without freeing it ever again. Check 'slabtop' and or /proc/slabinfo for huge usage. In my case it had been 'kmalloc-4096'

Cheers,

Pit

PIT, My issue was inadvertantly using a 32-bit kernel on a new generation CPU. Resolved now. Greg -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org