IRC chat logs for #ltsp on irc.libera.chat (webchat)
Channel log from 29 January 2015 (all times are UTC)
| 00:41 | TatankaT has left IRC (TatankaT!~tim@193.190.253.114, Ping timeout: 255 seconds) | |
| 00:57 | gdi2k_ has left IRC (gdi2k_!~gdi2k@222.127.254.113, Ping timeout: 252 seconds) | |
| 00:59 | gdi2k_ has joined IRC (gdi2k_!~gdi2k@222.127.254.113) | |
| 01:58 | andygraybeal has joined IRC (andygraybeal!~andy@h73.206.130.174.dynamic.ip.windstream.net) | |
| 02:06 | ogra_ has left IRC (ogra_!~ogra_@p5098ed03.dip0.t-ipconnect.de, Ping timeout: 264 seconds) | |
| 04:33 | telex has left IRC (telex!teletype@freeshell.de, Read error: Connection reset by peer) | |
| 04:34 | telex has joined IRC (telex!teletype@freeshell.de) | |
| 04:39 | vagrantc has left IRC (vagrantc!~vagrant@freegeek/vagrantc, Ping timeout: 252 seconds) | |
| 05:11 | cyberorg has joined IRC (cyberorg!~cyberorg@opensuse/member/Cyberorg) | |
| 05:24 | freedomrun has joined IRC (freedomrun!~quassel@unaffiliated/freedomrun) | |
| 06:06 | <work_alkisg> map7, vagrantc: please forward upstream bugs to the upstream bug tracker, don't leave them in distro-specific bug trackers...
| |
| 06:06 | work_alkisg is now known as alkisg | |
| 06:07 | <alkisg> !ltsp-bug
| |
| 06:07 | <ltsp`> ltsp-bug: To file a bug report for upstream LTSP, go to https://bugs.launchpad.net/ltsp
| |
| 06:07 | <alkisg> !ubuntu-bug
| |
| 06:07 | <ltsp`> ubuntu-bug: To file a bug report for Ubuntu LTSP, go to https://bugs.launchpad.net/ubuntu/+source/ltsp
| |
| 06:07 | <alkisg> !debian-bug
| |
| 06:07 | <ltsp`> Error: "debian-bug" is not a valid command.
| |
| 06:09 | <alkisg> !learn debian-bug as To file a bug report for Debian LTSP, go to https://bugs.debian.org/ltsp or one of the ltsp binary packages mentioned there
| |
| 06:09 | <ltsp`> The operation succeeded.
| |
| 06:09 | <alkisg> Distro-specific bug trackers should only contain distro-specific bugs, all the rest should be forwarded upstream
| |
| 06:20 | vagrantc has joined IRC (vagrantc!~vagrant@freegeek/vagrantc) | |
| 06:20 | <alkisg> (01:06:24 πμ) championofcyrodi: it's official. ramdisk hosted nbd image is much faster for reading from multiple clients ==>
| |
| 06:20 | ==> do you have some benchmark for that?
| |
| 06:20 | I believe that the first client will be slower, since it will read from the nbd disk, but all the rest will be as fast as accessing it from the ram image, because the nbd image will be cached by the kernel in ram
| |
| 06:32 | Here's my results: server side: echo 3 > proc/sys/vm/drop_caches, client side: echo 3 > proc/sys/vm/drop_caches; dd if=/dev/nbd0 of=/dev/null bs=1024 count=1000000; and again on the client the same drop caches and dd.
| |
| 06:33 | The first dd was at 126 MB/s, the second one at 222 MB/s
| |
| 06:34 | And server-side, `free` reported 245212 cached before, 1241748 cached after
| |
| 06:34 | So all the nbd image was indeed cached in RAM automatically by the kernel
| |
| 06:35 | So if you're indeed able to see some benefit from putting the nbd in a server-side tmpfs, that can only be due to bad kernel implementation of caching, and I don't think that's very likely...
| |
| 06:37 | The limit of 222 MB/sec in my case was the VM's CPU, shared between nbd-client and dd
| |
| 06:37 | alkisg is now known as work_alkisg | |
| 07:02 | ricotz has joined IRC (ricotz!~rico@ubuntu/member/ricotz) | |
| 07:28 | ame has joined IRC (ame!75c1338a@gateway/web/freenode/ip.117.193.51.138) | |
| 07:29 | <ame> can i configure ltsp-pnp on CAE linux??
| |
| 07:58 | Grembler has joined IRC (Grembler!~Ben@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net) | |
| 08:00 | work_alkisg is now known as alkisg | |
| 08:16 | <alkisg> ame: it says it's based on ubuntu, so probably yes, try it
| |
| 08:42 | ame has left IRC (ame!75c1338a@gateway/web/freenode/ip.117.193.51.138, Ping timeout: 246 seconds) | |
| 08:54 | ogra_ has joined IRC (ogra_!~ogra_@p5098ed03.dip0.t-ipconnect.de) | |
| 09:30 | vmlintu has joined IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi) | |
| 09:39 | vmlintu has left IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi, Ping timeout: 265 seconds) | |
| 09:43 | TatankaT has joined IRC (TatankaT!~tim@193.190.253.114) | |
| 09:52 | vmlintu has joined IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi) | |
| 10:06 | ame has joined IRC (ame!75c135c3@gateway/web/freenode/ip.117.193.53.195) | |
| 10:07 | <ame> which made my ubuntu system as LTSP server....Is there any configuration file is there??Can i get that pls??
| |
| 10:13 | Grembler has left IRC (Grembler!~Ben@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net, Ping timeout: 245 seconds) | |
| 10:13 | Grembler has joined IRC (Grembler!~Ben@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net) | |
| 10:23 | freedomrun has left IRC (freedomrun!~quassel@unaffiliated/freedomrun, Remote host closed the connection) | |
| 10:23 | cyberorg has left IRC (cyberorg!~cyberorg@opensuse/member/Cyberorg, Ping timeout: 240 seconds) | |
| 10:28 | <ame> hi i have installed CAE linux in an system and starting configuring ltsp-pnp with help of ubuntu documentation...But i cant able to generate image...How to generate image??
| |
| 10:46 | cyberorg has joined IRC (cyberorg!~cyberorg@opensuse/member/Cyberorg) | |
| 11:14 | ame has left IRC (ame!75c135c3@gateway/web/freenode/ip.117.193.53.195, Quit: Page closed) | |
| 11:27 | alkisg is now known as work_alkisg | |
| 11:29 | AlexPortable has joined IRC (AlexPortable!uid7568@gateway/web/irccloud.com/x-uiufkdhvelibrebh) | |
| 11:45 | freedomrun has joined IRC (freedomrun!~quassel@unaffiliated/freedomrun) | |
| 11:56 | ame has joined IRC (ame!75c135c3@gateway/web/freenode/ip.117.193.53.195) | |
| 11:57 | <ame> hi i have installed CAE linux in an system and starting configuring ltsp-pnp with help of ubuntu documentation...But i cant able to generate image...How to generate image??
| |
| 12:01 | ame has left IRC (ame!75c135c3@gateway/web/freenode/ip.117.193.53.195, Client Quit) | |
| 12:04 | vmlintu has left IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi, Read error: Connection reset by peer) | |
| 12:12 | cyberorg has left IRC (cyberorg!~cyberorg@opensuse/member/Cyberorg, Ping timeout: 252 seconds) | |
| 12:42 | gdi2k_ has left IRC (gdi2k_!~gdi2k@222.127.254.113, Ping timeout: 265 seconds) | |
| 12:44 | Faith has joined IRC (Faith!~paty@unaffiliated/faith) | |
| 12:46 | gdi2k_ has joined IRC (gdi2k_!~gdi2k@222.127.254.113) | |
| 13:13 | cliebow has joined IRC (cliebow!~cliebow@gw-rsu24-co.rsu24.org) | |
| 14:09 | telex has left IRC (telex!teletype@freeshell.de, Remote host closed the connection) | |
| 14:10 | telex has joined IRC (telex!teletype@freeshell.de) | |
| 14:32 | mgariepy has left IRC (mgariepy!~mgariepy@ubuntu/member/mgariepy, Quit: Leaving) | |
| 14:35 | mgariepy has joined IRC (mgariepy!~mgariepy@ubuntu/member/mgariepy) | |
| 14:42 | gdi2k_ has left IRC (gdi2k_!~gdi2k@222.127.254.113, Ping timeout: 276 seconds) | |
| 14:43 | gdi2k_ has joined IRC (gdi2k_!~gdi2k@222.127.254.113) | |
| 14:53 | vmlintu has joined IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi) | |
| 15:14 | gdi2k_ has left IRC (gdi2k_!~gdi2k@222.127.254.113, Ping timeout: 240 seconds) | |
| 15:15 | NeonLicht has left IRC (NeonLicht!~NeonLicht@darwin.ugr.es, Quit: leaving) | |
| 15:16 | gdi2k_ has joined IRC (gdi2k_!~gdi2k@222.127.254.113) | |
| 15:17 | NeonLicht has joined IRC (NeonLicht!~NeonLicht@darwin.ugr.es) | |
| 15:17 | NeonLicht has left IRC (NeonLicht!~NeonLicht@darwin.ugr.es, Client Quit) | |
| 15:19 | NeonLicht has joined IRC (NeonLicht!~NeonLicht@darwin.ugr.es) | |
| 15:45 | gdi2k_ has left IRC (gdi2k_!~gdi2k@222.127.254.113, Ping timeout: 264 seconds) | |
| 15:48 | <championofcyrodi> alkisg: droping cache from the client before running dd shows 100 MB/s for 1.0 GB. dd's w/o clearing cache after show 1.5 GB/s, indicating that the client is caching the image. are you saying that the nbd-server is caching the image as well, so tmpfs would be redundant?
| |
| 15:49 | when i drop cache from the server, 'free' doesnt change
| |
| 15:50 | going to switch back and to normal disk mount img file and do some more tests.
| |
| 16:03 | gdi2k_ has joined IRC (gdi2k_!~gdi2k@203.177.248.238) | |
| 16:20 | adrianorg has left IRC (adrianorg!~adrianorg@177.156.59.219, Ping timeout: 255 seconds) | |
| 16:22 | adrianorg has joined IRC (adrianorg!~adrianorg@177.156.59.219) | |
| 16:29 | work_alkisg is now known as alkisg | |
| 16:31 | <alkisg> championofcyrodi: yes, the nbd image is cached in the server RAM. Linux by default fills all the available RAM by whatever it read from the disk, and only clears parts of that disk cache when applications ask to allocate RAM, so if you have enough RAM then all the nbd image will be cached for a long time
| |
| 16:32 | if you drop caches on the server, then run free on the server, then use dd on the client, and then again run free on the server, you should be seeing that cache
| |
| 16:38 | <championofcyrodi> so this is interesting...
| |
| 16:38 | i shutdown my client, and server.
| |
| 16:38 | started the server and let it sit idle for a moment, then ran free -m
| |
| 16:39 | 3145 MB Used.
| |
| 16:39 | Buffers = 11
| |
| 16:39 | cached = 3021
| |
| 16:40 | booted the client ran dd ~ 100mb/sec, reran dd ~ 1.4GB /sec.
| |
| 16:40 | re-ran free -m on the server...
| |
| 16:40 | 3187 MB used
| |
| 16:40 | Buffers = 14
| |
| 16:40 | cached = 3058
| |
| 16:41 | so it would seem that copying the image to a ram disk makes the 1st client boot/read almost as fast as the second, and so on...
| |
| 16:41 | i would have thought that nbd might double the memory usage. but it did not.
| |
| 16:42 | e.g. tmpfs image + nbd cache
| |
| 16:42 | but it seems as though nbd knows that the blocks are already cached via tmpfs
| |
| 16:43 | my image is 1.5 GB
| |
| 16:43 | or is nbd-server caching the image up front at startup?
| |
| 16:44 | which is why I am seeing ~3GB at boot
| |
| 16:46 | <alkisg> championofcyrodi: linux is smart enough to only cache once the disk data
| |
| 16:46 | It cached them once with your script reading the nbd image to the tmpfs
| |
| 16:46 | It won't cache them again until you drop the server caches
| |
| 16:46 | I do think that you'll be needing twice the RAM though, once for the cache and twice for the tmpfs
| |
| 16:47 | I.e. by using the tmpfs you're just wasting RAM...
| |
| 16:58 | <championofcyrodi> so my mistake was assuming that nbd-server only cached blocks read from the client. but rather it caches the entire image up front.
| |
| 16:59 | should have took metrics on memory usage before switching to tmpfs :)
| |
| 16:59 | <alkisg> championofcyrodi: no, it doesn't cache the entire image normally
| |
| 16:59 | it only caches it now because you're using your tmpfs script, and you're reading the whole image
| |
| 17:00 | Normally, it only caches the parts that the clients read, and that's the ideal implementation (unless you have other specific needs)
| |
| 17:00 | <championofcyrodi> i just switch my config and i am not using tmpfs. on boot i'm using 1.5 GB before a client even boots.
| |
| 17:00 | <alkisg> Where are you copying the nbd image to RAM?
| |
| 17:00 | Ah
| |
| 17:01 | What's the output of free, (1) right upon boot, and (2) right after you drop caches?
| |
| 17:01 | <championofcyrodi> then doing an inital dd on the client for 1024 bs x 3000000, i get 100Mb/sec, and free -m on the server doesnt change
| |
| 17:01 | let me stop nbd server, drop cache, then start it back up...
| |
| 17:02 | <alkisg> No need to stop it
| |
| 17:02 | Ah you put the startup script as a wrapper to nbd-server?!
| |
| 17:02 | Normally you'd just need to drop the caches, to see that 1.5 gb cache go away...
| |
| 17:02 | <ogra_> if you measure with dd you want to use conv=fsync
| |
| 17:02 | that disables all caching from the start
| |
| 17:05 | <alkisg> ogra_: we're writing to /dev/null, no write cache there at all
| |
| 17:05 | fsync is about write cache, not read cache, right?
| |
| 17:05 | <ogra_> the kernel always caches
| |
| 17:05 | (unless you tell it not to :) )
| |
| 17:06 | yeah, thats write cache
| |
| 17:06 | <alkisg> I believe fsync means "sync on writes", we're testing read cache now though
| |
| 17:06 | <championofcyrodi> when i disabled creating the tmpfs, i also disabled copying the image in the nbd-server script.
| |
| 17:06 | <alkisg> nbd-server doesn't write anywhere... it's read only
| |
| 17:06 | (for ltsp)
| |
| 17:06 | <ogra_> ah, there is another option for read caches ... forgot how it was called though
| |
| 17:07 | <championofcyrodi> weird... on boot i'm using 1.7GB, then booting a client and running dd for the first time hardly impacts my server memory usage.
| |
| 17:07 | with the normal configuration of ltsp-server reading from /opt/ltsp/...
| |
| 17:07 | <alkisg> championofcyrodi: how much do you dd?
| |
| 17:07 | <championofcyrodi> 1.6GB
| |
| 17:08 | but if i stop nbd, clear caches, and start it back up... memory usage drops to around 300MB
| |
| 17:08 | and stays there.
| |
| 17:08 | as though nbd is caching the entire image on boot... but not when starting the service. which doesnt make sense to me
| |
| 17:08 | <alkisg> When a client boots, it reads about 50 MB of the nbd image
| |
| 17:08 | I can't imagine why you're seeing 1.5 GB cache, I don't see it here
| |
| 17:09 | <championofcyrodi> let me look over it again... i must be missing something.
| |
| 17:09 | <alkisg> The server reads and caches some hundred MB, but it shouldn't be that much
| |
| 17:09 | (while booting)
| |
| 17:11 | <championofcyrodi> just did a reboot... 1541 MB Cached on boot.
| |
| 17:11 | with no tmpfs
| |
| 17:11 | and of course... booting a client and reading 1.6GB will only increase that cache by a nominal ammount, ~20MB
| |
| 17:12 | but if i stop nbd-server, drop cache and start it back up... it just hangs around 300 MB.. aka, doesnt cache the whole image.
| |
| 17:12 | yup... now my server only has 23MB cached
| |
| 17:13 | <alkisg> You don't need to stop the nbd-server, you can just drop caches
| |
| 17:13 | It's weird that your image is read on boot, it shouldn't be
| |
| 17:13 | Maybe you've got something left over from your tmpfs script
| |
| 17:13 | E.g. maybe you're doing the copy even though the tmpfs target isn't there?
| |
| 17:15 | <championofcyrodi> i'll double check... but all i did was add a tmpfs entry to /etc/fstab, and add a 'cp' command to the nbd-server 'start' function, and point the /etc/nbd-server/conf.d/ltsp-amd64 config at the tmpfs version of the img.
| |
| 17:15 | and going back over it... it's all been commented out and the nbd-server config is pointing back to the /opt/ltsp path on the disk.
| |
| 17:15 | possibly an issue doing 'reboot' instead of a clean shutdown?
| |
| 17:15 | e.g. hard reboot
| |
| 17:17 | fyi... this isn't really a blocker. i'm just trying to understand better how the image is served up. initially i just didn't want clients thrashing the servers disk for nbd reads. and i've seen now that it wont happen.
| |
| 17:17 | using tmpfs or not.
| |
| 17:18 | <alkisg> Try moving the i386.img image elsewhere, then reboot, then check cache + syslog
| |
| 17:18 | Maybe you'll see what tried to access the image on boot, if something did try...
| |
| 17:19 | <championofcyrodi> k
| |
| 17:20 | but it's def. caching a majority of the image... just did a hard reboot, had 1.5 GB cached... ran the echo 3 drop_caches w/o restarting nbd and its just staying around 22MB
| |
| 17:21 | so i'll just rename the img to img.bak and reboot, then check syslog + cache
| |
| 17:21 | maybe it just caches on boot based on how much memory vs. image size?
| |
| 17:22 | e.g. i have 8GB RAM, and the OS uses 100MB on boot outside of nbd's image
| |
| 17:22 | <alkisg> The first question would be, "what reads the image on server boot?!"
| |
| 17:22 | <championofcyrodi> okay, i moved the image and rebooted, now i'm only seeing about 60MB cached... but def not 1500
| |
| 17:23 | <alkisg> nbd-server only access it when the clients ask it, on a "sector by sector" basis...
| |
| 17:23 | Put the image back, and try to move aside the nbd-server executable
| |
| 17:23 | mv /bin/nbd-server /bin/nbd-server.original
| |
| 17:23 | +reboot...
| |
| 17:24 | About available RAM, I believe that linux caches everything and only frees caches when applications try to malloc()
| |
| 17:28 | <championofcyrodi> okay, going to try to move the nbd-server executable. I didn't see any complaints in the logs about the image file being missing... there is just nothing in syslog at all, where as before it showed 'starting to serve', 'size of exported file/device is 1552523264'
| |
| 17:30 | okay... moving nbd-server executable shows 1.5GB cached... which means that something else is caching the image on boot, rather than nbd-server
| |
| 17:31 | <alkisg> I believe it's something you did as part of your tmpfs effort... maybe something before the tmpfs solution?
| |
| 17:32 | You can also change your ltsp/nbd config so that the image is renamed, e.g. not-i386.img
| |
| 17:32 | Everywhere, even on dhcp for the clients to boot from it
| |
| 17:32 | and pxelinux.cfg/default too...
| |
| 17:33 | So that you verify that everything will work fine and it's something unrelated to standard ltsp setups that caches your i386.img on boot...
| |
| 17:39 | freedomrun has left IRC (freedomrun!~quassel@unaffiliated/freedomrun, Read error: Connection reset by peer) | |
| 17:39 | freedomrun has joined IRC (freedomrun!~quassel@unaffiliated/freedomrun) | |
| 17:42 | <championofcyrodi> not sure what it could be...
| |
| 17:43 | again, all i did for tmpfs was create the mount point, copy the img on boot, and change where nbd-server is looking to serve it up.
| |
| 17:43 | i don't think i could have done anything else, because i wouldnt know what else to do.
| |
| 17:44 | it could be another service i installed on the server, sssd
| |
| 17:44 | but that is supposed to cache credentials and integrate w/ pam.d to support ldap sshfs
| |
| 17:45 | and when i boot my first client and start reading from nbd, i don't see any significant increase in cache
| |
| 17:45 | oh well, i'm over it. moving on now.
| |
| 17:56 | freedomrun has left IRC (freedomrun!~quassel@unaffiliated/freedomrun, Read error: Connection reset by peer) | |
| 17:57 | freedomrun has joined IRC (freedomrun!~quassel@unaffiliated/freedomrun) | |
| 18:03 | freedomrun has left IRC (freedomrun!~quassel@unaffiliated/freedomrun, Read error: Connection reset by peer) | |
| 18:17 | alkisg is now known as work_alkisg | |
| 18:32 | vagrantc has joined IRC (vagrantc!~vagrant@freegeek/vagrantc) | |
| 18:38 | work_alkisg is now known as alkisg | |
| 19:09 | freedomrun has joined IRC (freedomrun!~quassel@unaffiliated/freedomrun) | |
| 19:14 | alkisg is now known as work_alkisg | |
| 19:33 | cliebow has left IRC (cliebow!~cliebow@gw-rsu24-co.rsu24.org, Quit: Ex-Chat) | |
| 19:36 | Grembler has left IRC (Grembler!~Ben@cpc29-aztw22-2-0-cust128.18-1.cable.virginm.net, Quit: I Leave) | |
| 19:54 | telex has left IRC (telex!teletype@freeshell.de, Remote host closed the connection) | |
| 19:56 | telex has joined IRC (telex!teletype@freeshell.de) | |
| 19:56 | mgariepy has left IRC (mgariepy!~mgariepy@ubuntu/member/mgariepy, Ping timeout: 264 seconds) | |
| 19:56 | mgariepy has joined IRC (mgariepy!~mgariepy@ubuntu/member/mgariepy) | |
| 20:01 | khildin has joined IRC (khildin!~khildin@ip-213-49-87-210.dsl.scarlet.be) | |
| 20:15 | Faith has left IRC (Faith!~paty@unaffiliated/faith, Quit: Saindo) | |
| 20:31 | championofcyrodi has left IRC (championofcyrodi!~cott@50-205-35-98-static.hfc.comcastbusiness.net) | |
| 21:38 | vmlintu has left IRC (vmlintu!~vmlintu@a91-152-200-70.elisa-laajakaista.fi, Ping timeout: 245 seconds) | |
| 21:44 | work_alkisg is now known as alkisg | |
| 22:13 | alkisg is now known as work_alkisg | |
| 22:19 | gdi2k_ has left IRC (gdi2k_!~gdi2k@203.177.248.238, Ping timeout: 264 seconds) | |
| 22:28 | khildin has left IRC (khildin!~khildin@ip-213-49-87-210.dsl.scarlet.be, Quit: I'm gone, bye bye) | |
| 23:05 | <muppis> map7, good you got it solved.
| |
| 23:47 | ricotz has left IRC (ricotz!~rico@ubuntu/member/ricotz, Quit: Ex-Chat) | |