Odd swap space behavior
One of our Redhat servers got a LOT of activity yesterday and the swap space looks funny to me. swapon -s FilenameTypeSizeUsedPriority /dev/dasda2 partition 1023976 3692-1 /dev/dasdb1 partition 194964 420 2 /dev/dasdc1 partition 64976 152 1 /dev/dasdd1 partition 196596 25244 3 Why would the system use swap space on dasdc1, dasdb1 and dasda2 if dasdd1 hasn't run out? Bobby Bauer Center for Information Technology National Institutes of Health Bethesda, MD 20892-5628 301-594-7474 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
On Wed, Nov 2, 2011 at 12:37 PM, Bauer, Bobby (NIH/CIT) [E] baue...@mail.nih.gov wrote: One of our Redhat servers got a LOT of activity yesterday and the swap space looks funny to me. swapon -s Filename Type Size Used Priority /dev/dasda2 partition 1023976 3692 -1 /dev/dasdb1 partition 194964 420 2 /dev/dasdc1 partition 64976 152 1 /dev/dasdd1 partition 196596 25244 3 Why would the system use swap space on dasdc1, dasdb1 and dasda2 if dasdd1 hasn't run out? It filled the first ones and eventually used the last one. Some processes were killed in the fight and their pages on swap space got released. As long as it did not go too quick, a performance monitor could show you total swap usage over time and reveal you (briefly) had that much swapped out. There's nothing in Linux that will migrate things back to the first swap disks in the list, other than when you swapoff the last ones in the chain. Remember that when you swapoff a VDISK, z/VM will still hold the old data (and use memory for that). Rob -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
On Wed, Nov 02, 2011 at 07:37:17AM -0400, Bauer, Bobby (NIH/CIT) [E] wrote: Date: Wed, 2 Nov 2011 07:37:17 -0400 From: Bauer, Bobby (NIH/CIT) [E] baue...@mail.nih.gov To: LINUX-390@VM.MARIST.EDU Subject: Odd swap space behavior One of our Redhat servers got a LOT of activity yesterday and the swap space looks funny to me. swapon -s FilenameTypeSizeUsed Priority /dev/dasda2 partition 1023976 3692-1 /dev/dasdb1 partition 194964 420 2 /dev/dasdc1 partition 64976 152 1 /dev/dasdd1 partition 196596 25244 3 haven't done Linux on Z for a while, but I have always used the same Priority for the swapdisks so that linux could spread out the IO to several disks (preferably on separate spindles). This works well on x86 (real VMware) and P-Series quote from http://www.vm.ibm.com/perf/tips/linuxper.html Swap extents of equal priority are used in round-robin fashion. Equal prioritization can be used to spread swap I/O across chpids and controllers, but if you are doing this, be careful not to put all the swap extents on minidisks on the same physical DASD volume, for if you do, you will not be accomplishing any spreading. /quote I'd be interested to see what today's thinking is. //rhi - now back to lurking -- ... Point and click ... ... probably means that you forgot to load the gun ... Have a nice day ;-) Richard Higson mailto:richard.hig...@gt.owl.de -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
That's what you want when you're using spindles, but on z, you're usually talking about v-disks, which are really virtual disks in memory. When they're not in use, they take up no space at all, but when you start using them, they start to occupy real memory and become a burden. So you set priorities on the swap spaces so that they each get used one at a time in turn. Ideally, you don't want to use them at all; they're a safeguard to keep the image from coming down. When they are used, they're an indication that you need more memory allocated to the image, and they give you a buffer to get to the moment when you can safely cycle the image to add that memory. Having four swap spaces allocated seems like a bit of overkill to me. It should be sufficient to have one to be the buffer, and a second larger one to be the trigger to increase the size of the image. -- Robert P. Nix Mayo Foundation.~. RO-OC-1-18 200 First Street SW/V\ 507-284-0844 Rochester, MN 55905 /( )\ -^^-^^ In theory, theory and practice are the same, but in practice, theory and practice are different. On 11/2/11 8:25 AM, Richard Higson richard.hig...@gt.owl.de wrote: On Wed, Nov 02, 2011 at 07:37:17AM -0400, Bauer, Bobby (NIH/CIT) [E] wrote: haven't done Linux on Z for a while, but I have always used the same Priority for the swapdisks so that linux could spread out the IO to several disks (preferably on separate spindles). This works well on x86 (real VMware) and P-Series -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
Yes, having 4 is a little odd. We are struggling with this server. It sits almost idle most of the month then for 1 or 2 days it gets 60 to 80 thousand hits/hour. Not sure what to make of this current display of the swap space. Bobby Bauer Center for Information Technology National Institutes of Health Bethesda, MD 20892-5628 301-594-7474 -Original Message- From: RPN01 [mailto:nix.rob...@mayo.edu] Sent: Wednesday, November 02, 2011 10:14 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Odd swap space behavior That's what you want when you're using spindles, but on z, you're usually talking about v-disks, which are really virtual disks in memory. When they're not in use, they take up no space at all, but when you start using them, they start to occupy real memory and become a burden. So you set priorities on the swap spaces so that they each get used one at a time in turn. Ideally, you don't want to use them at all; they're a safeguard to keep the image from coming down. When they are used, they're an indication that you need more memory allocated to the image, and they give you a buffer to get to the moment when you can safely cycle the image to add that memory. Having four swap spaces allocated seems like a bit of overkill to me. It should be sufficient to have one to be the buffer, and a second larger one to be the trigger to increase the size of the image. -- Robert P. Nix Mayo Foundation.~. RO-OC-1-18 200 First Street SW/V\ 507-284-0844 Rochester, MN 55905 /( )\ -^^-^^ In theory, theory and practice are the same, but in practice, theory and practice are different. On 11/2/11 8:25 AM, Richard Higson richard.hig...@gt.owl.de wrote: On Wed, Nov 02, 2011 at 07:37:17AM -0400, Bauer, Bobby (NIH/CIT) [E] wrote: haven't done Linux on Z for a while, but I have always used the same Priority for the swapdisks so that linux could spread out the IO to several disks (preferably on separate spindles). This works well on x86 (real VMware) and P-Series -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
On Wed, Nov 2, 2011 at 2:25 PM, Richard Higson richard.hig...@gt.owl.de wrote: haven't done Linux on Z for a while, but I have always used the same Priority for the swapdisks so that linux could spread out the IO to several disks (preferably on separate spindles). This works well on x86 (real VMware) and P-Series The OP is correct in using different priority for the swap devices. The issue with Linux on z/VM using VDISK as swap is not to spread the I/O (there is no I/O for VDISK). Multiple swap devices with different priority is to force Linux to reuse blocks rather than take fresh ones. This reduces the amount of pages that must be backed by z/VM. It's also important when you have a mix of disk types (like VDISK and real disk). Rob -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
You might consider a manual 'swapoff' (then 'swapon') of one large swap volume after that crunch time. In any case, this is one where you should reconsider how much VDISK to use. Obviously, there's a lot happening when it gets that end-of-month workload, so remember to include CPU and other I/O when you profile this server. As Rob said, there's no page migration in Linux. (Other than to force the issue with a 'swapoff' and 'swapon' cycle.) So what you're seeing is random pages which got pushed out at various times during the stress period. If not needed, they will sit there forever. I like to differentiate between swap occupancy and swap movement. The occupancy doesn't really hurt you in terms of response time. -- R; Rick Troth Velocity Software http://www.velocitysoftware.com/ On Wed, Nov 2, 2011 at 10:21, Bauer, Bobby (NIH/CIT) [E] baue...@mail.nih.gov wrote: Yes, having 4 is a little odd. We are struggling with this server. It sits almost idle most of the month then for 1 or 2 days it gets 60 to 80 thousand hits/hour. Not sure what to make of this current display of the swap space. Bobby Bauer Center for Information Technology National Institutes of Health Bethesda, MD 20892-5628 301-594-7474 -Original Message- From: RPN01 [mailto:nix.rob...@mayo.edu] Sent: Wednesday, November 02, 2011 10:14 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Odd swap space behavior That's what you want when you're using spindles, but on z, you're usually talking about v-disks, which are really virtual disks in memory. When they're not in use, they take up no space at all, but when you start using them, they start to occupy real memory and become a burden. So you set priorities on the swap spaces so that they each get used one at a time in turn. Ideally, you don't want to use them at all; they're a safeguard to keep the image from coming down. When they are used, they're an indication that you need more memory allocated to the image, and they give you a buffer to get to the moment when you can safely cycle the image to add that memory. Having four swap spaces allocated seems like a bit of overkill to me. It should be sufficient to have one to be the buffer, and a second larger one to be the trigger to increase the size of the image. -- Robert P. Nix Mayo Foundation .~. RO-OC-1-18 200 First Street SW /V\ 507-284-0844 Rochester, MN 55905 /( )\ - ^^-^^ In theory, theory and practice are the same, but in practice, theory and practice are different. On 11/2/11 8:25 AM, Richard Higson richard.hig...@gt.owl.de wrote: On Wed, Nov 02, 2011 at 07:37:17AM -0400, Bauer, Bobby (NIH/CIT) [E] wrote: haven't done Linux on Z for a while, but I have always used the same Priority for the swapdisks so that linux could spread out the IO to several disks (preferably on separate spindles). This works well on x86 (real VMware) and P-Series -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
You might also consider using a real disk to back the v-disk for the peak period swap, so that it doesn't add additional memory pressure to the underlying z/VM system. On 11/2/11 10:29 AM, Richard Troth vmcow...@gmail.com wrote: You might consider a manual 'swapoff' (then 'swapon') of one large swap volume after that crunch time. In any case, this is one where you should reconsider how much VDISK to use. Obviously, there's a lot happening when it gets that end-of-month workload, so remember to include CPU and other I/O when you profile this server. -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
As Rob said, there's no page migration in Linux. Yet. 8-) -- db -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Netbackup and SLES 11
We are running Netbackup successfully on SLES11. NetBackup-IBMzSeriesSuSE2.6 6.5.4 Joe -Original Message- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Bern VK2KAD Sent: Thursday, October 06, 2011 6:20 PM To: LINUX-390@VM.MARIST.EDU Subject: Netbackup and SLES 11 Hi All Is anyone deploying Netbackup on SLES11. I am about to embark on this trail and am looking for footprints to follow. We already have Master and Media servers deployed in the mid-range SANs - we are looking to exploit this infrastructure to give us file level backups of our zVM guests. Our preferred direction is do the backups via fibre channel rather than the network - our z10 only had OSA Express2 cards so bandwidth is limited. Early research reveals not all features/functions are available on s390x architecture. All comments appreciated Bern -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Netbackup and SLES 11
Joe: Which oracle version is running? Our Unix group test SLES 11 with Oracle 11 and doesn't work. Regards, Victor Echavarry System Programmer Technology Systems Operations Division EVERTEC -Original Message- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Joe Comitini Sent: Wednesday, November 02, 2011 2:39 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Netbackup and SLES 11 We are running Netbackup successfully on SLES11. NetBackup-IBMzSeriesSuSE2.6 6.5.4 Joe -Original Message- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Bern VK2KAD Sent: Thursday, October 06, 2011 6:20 PM To: LINUX-390@VM.MARIST.EDU Subject: Netbackup and SLES 11 Hi All Is anyone deploying Netbackup on SLES11. I am about to embark on this trail and am looking for footprints to follow. We already have Master and Media servers deployed in the mid-range SANs - we are looking to exploit this infrastructure to give us file level backups of our zVM guests. Our preferred direction is do the backups via fibre channel rather than the network - our z10 only had OSA Express2 cards so bandwidth is limited. Early research reveals not all features/functions are available on s390x architecture. All comments appreciated Bern -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/ - CONFIDENTIALITY NOTE: This email communication and its attachments contain information that are proprietary and confidential to EVERTEC, INC., its affiliates or its clients. They may not be disclosed, distributed, used, copied or modified in any way without EVERTEC, Inc.’s authorization. If you are not the intended recipient of this email, you are not an authorized person. Please delete it and notify the sender immediately. EVERTEC, Inc. and its affiliates do not assume any liability for damages resulting from emails that have been sent or altered without their consent. Moreover, EVERTEC, Inc. has taken precautions to safeguard its email communications, but cannot assure that such is the case and disclaim any responsibility attributable thereto. -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/
Re: Odd swap space behavior
On Wed, 2 Nov 2011 11:29:40 -0400 Richard Troth wrote: So what you're seeing is random pages which got pushed out at various times during the stress period. If not needed, they will sit there forever. Well, maybe not forever ... ;-) This lazy (de-)allocation behaviour of Linux is worth remembering. It's just too expensive to continually run the q's to clean this up. Later kernels expose the per-pid (actual) swap usage - I haven't figured out if there is yet a reliable means of discerning disk vs. cache swap usage. I like to differentiate between swap occupancy and swap movement. The occupancy doesn't really hurt you in terms of response time. Most of the time. If memory pressure ramps up *really* quickly, kswapd gets kicked into action to run the q's to free up pages. And it can cycle back through chasing enough memory to free. Not likely in this scenario, but if kswapd needs to work, you wait. Shane ... -- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For more information on Linux on System z, visit http://wiki.linuxvm.org/