1473557430 J * derjohn_mobi ~aj@x590e0bde.dyn.telefonica.de 1473557871 Q * aj__ Ping timeout: 480 seconds 1473560070 J * fstd_ ~fstd@x4db6de36.dyn.telefonica.de 1473560071 Q * fstd Read error: Connection reset by peer 1473560086 N * fstd_ fstd 1473561692 M * Bertl off for a nap ... bbl 1473561693 N * Bertl Bertl_zZ 1473574287 J * Aiken ~Aiken@d63f.h.jbmb.net 1473574346 Q * Aiken 1473574412 J * Aiken ~Aiken@d63f.h.jbmb.net 1473574415 Q * Aiken 1473574516 J * Aiken ~Aiken@d63f.h.jbmb.net 1473579541 J * Ghislain ~aqueos@adsl1.aqueos.com 1473582029 N * Bertl_zZ Bertl 1473582031 M * Bertl morning folks! 1473582990 M * Ghislain hello bertl, running 4.1.32 in test, seems fine for now 1473583043 M * Ghislain about xfs i have one server using it 1473583092 M * Ghislain it runs 3.4.108-vs2.3.3.9aq, the support is in the 3.4 familly it seems 1473583415 M * Ghislain i guess the switch to 4.x was the problem 1473583614 M * Bertl well, yes and no, the 'problem' was, that they changed the kernel code too often for my taste and adapting it over and over again for no apparent user was tedious 1473585053 M * Ghislain yes i see :) 1473585098 M * Ghislain if i use a on supported fs (yes i think i allready asked this) i loose quota and dlimits but that's it the rest will work as usual ? 1473585203 M * Ghislain i mean an 'unsupported' 1473585729 M * Bertl you don't have the barrier, but it is not sure if it is still required to create secure guests 1473585782 M * Bertl an that only applies if the parent dir (for the guest) is on the unsupported fs 1473585905 M * Bertl IMHO the pivot root done by recent util-vserver and the cleanup of unused mounts should make a guest secure without the use of the barrier flag 1473586060 M * Ghislain ok, thx bertl 1473586079 M * Bertl np 1473586118 M * Ghislain i would have loved to switch to xfs as my disk start to have capacity i do not trust ext4 anymore but we do not have the ressource to make 4.1 work so lets stay ext4 for now :) 1473586416 M * Bertl what is your current capacity if I may ask? 1473587080 M * Ghislain not so much it is 4x4tB drives 1473587104 M * Ghislain but ext4 fsck and quota is worring me 1473587159 M * Ghislain the more i try to figure out FS on linux the more i get XFS is clean and solid and ext4 is a mess, my experience is the more the capacity the more issue i have when i encounter a pb 1473587178 M * Ghislain still not loos important data but i am listening to get better things 1473587203 M * Ghislain btrfs would be great but 50% say it is ok, 50% say they lost data so i cannot make my choice here 1473587221 M * Bertl well, ext* has the advantage that it is really common (thus well tested) 1473587250 M * Bertl but from the fs design, I probably would go for zfs 1473587265 M * Ghislain yes, xfs is also well tested but btrfs and zfs are not at all. But snapshotting is such an important things for us would would LOVE to have it 1473587290 M * Ghislain for zfs i mean on linux not on solaris/bsd 1473587347 M * Ghislain the issue in zfs is that it seems you have to have a zfs specialist to make it work the rigth way, it eats quite some ram, even it is less and less an issue 1473587565 J * bonbons ~bonbons@2001:a18:205:7601:bc18:6235:2008:d83 1473587735 M * Ghislain there is also bcachefs in the works that sound promising but this is a one man project it seems 1473587793 M * Ghislain hard to compete with the others. With all the facebook ingeneers on the btrfs side i wonder how it is not seen as stable yet, they do not communicate about their work on it 1473587830 M * Ghislain and still DB perf are horrible making it an issue as i do not want to have one partition for each app so it can work well :p 1473588284 M * Ghislain fb use btrfs for data and xfs for DB i heard 1473588397 M * Bertl I guess there simply is no one-for-all filesystem :) 1473588447 M * Ghislain too bad no ? i was dreaming of a fs where it can be told: espect lots of small file here or bunch of big ones here, and it tune for it :) MAGIC ! 1473588484 M * Ghislain well i also have a dream where cron is easy and you do not have to search in 37 different places if someone put a cron here so.. 1473588652 M * Bertl it's always nice to dream big :) 1473589274 M * Ghislain yes, i start to think i could give a try to the cron thing as a pet project 1473589286 M * Ghislain that sounds a cool thing to do ^^ 1473589891 M * Ghislain Load average: 3008.20 3007.00 3005.29, hum linux 4.x has really make improvement oin stability. Not so long ago at this 10x less load i could not connect to the server in ssh 1473589899 M * Ghislain here i can connect and issue commands 1473590049 M * Bertl it really depends on what the processes are doing 1473590079 M * Bertl on heavy I/O without proper scheduler setup, you won't get much interactive feedback 1473590109 M * Ghislain this is sysbench in loop mainly 1473590139 M * Ghislain sysbench --test cpu --cpu-max-prime 20000 run and sysbench --test fileio --file-total-size 5G on 2 guest 1473590184 M * Ghislain so cpu load is maxed on all cores and disk is used also 1473590243 M * Ghislain plus several other so i mess the more possible 1473590252 M * Ghislain well not the more but reasonably well 1473590281 M * Ghislain like sysbench --test threads --thread-yields 5000 --thread-locks 16 run added too, i am up for suggestion to mess it up more 1473590328 M * Ghislain vps auxwf 1473590331 M * Ghislain oups 1473590462 M * Ghislain the 3.4.112 do not work on the 3.4 patch, i get : 2 hunk failed on fs/file_table.c 1473590674 M * Ghislain fyi, was just toying with it ^^ 1473591334 M * Bertl yeah, good to know ... 1473595339 Q * Carpoon Ping timeout: 480 seconds 1473595531 J * Carpoon ~Carpoon@carpoon.hu 1473598561 Q * Romster Remote host closed the connection 1473601304 M * [Guy] Ghislain: fwiw, I've been using zfsonlinux in production since about 2012 and am quite happy with it 1473601310 N * [Guy] Guy- 1473602031 Q * geos_one Ping timeout: 480 seconds 1473604144 M * Bertl off for now ... bbl 1473604146 N * Bertl Bertl_oO 1473605887 M * Ghislain guy-: on what scale, you mean the out kernel module ? 1473606451 M * Guy- Ghislain: yes, the one from github.com/zfsonlinux; on about 20 physical boxes, the largest pool consisting of about 30 disks with a total capacity of about 70TB 1473606462 M * Guy- Ghislain: what other scale aspect are you interested in? 1473606499 M * Ghislain you beat me by a large margin, was to know if it was on 2 machine with 2x1tb or a real production use ^^ 1473606534 M * Ghislain the ram usage is not to high ? 1473606573 M * Guy- Ghislain: no; also, it's tunable 1473606606 M * Guy- Ghislain: but with today's amounts of RAM, it's not much of an issue anyway 1473606636 M * Ghislain it depends, we have some limits , most machine are 32gb for us 1473606640 M * Guy- Ghislain: I wouldn't recommend having a large pool with lots of SSD cache and a small amount of RAM, though 1473606648 M * Guy- Ghislain: 32GB is plenty 1473606657 M * Guy- I have zfs on boxes with as little as 16GB 1473606665 M * Ghislain i must have plenty of this for mysql apache and other spamd 1473606667 M * Guy- (as well as on my netbook, which has 4GB :) 1473606683 M * Guy- zfs uses RAM for cache 1473606691 M * Ghislain how do you see the ram usage of the FS for you ? 1473606706 M * Ghislain reserving 4gb could be enough ? 1473606716 M * Guy- there is a script called arcstat.py which is similar to vmstat 1473606729 M * Guy- depends on your i/o workload 1473606778 M * Guy- if you have lots of writes and lots of snapshots, most data will not be contiguous on disk 1473606788 M * Guy- (because it's never overwritten in place) 1473606813 M * Guy- if you use spinning disks, this will cause lots of seeking if you don't also have a large cache 1473606843 M * Guy- so I'd say having just 4GB for cache would be too little in typical circumstances 1473606865 M * Ghislain what do you use yourself on that case ? 1473606876 M * Ghislain on your typical workload 1473606877 M * Guy- the box with the 70TB pool currently uses 17GB RAM for zfs cache 1473606942 M * Guy- another one with a ~10TB pool and heavy i/o (samba, ldap, build servers, interactive use, subversion) uses 15GB 1473606983 M * Guy- the 2nd server also uses 206GB of SSD cache as of this moment 1473607011 M * Ghislain seems quite a beast, i deal with very small customer so i have small machines 1473607057 M * Ghislain thanks for your infos that certainly help lower the fear on zfs on linux 1473607061 M * Guy- in contrast, I have a box with lots of RAM, many CPUs but only ~1TB of mirrored storage 1473607067 M * Guy- that uses 2GB of RAM for zfs 1473607095 M * Ghislain yes it depend really on the workload 1473607097 M * Guy- (it also uses 37GB of SSD cache) 1473607248 M * Guy- Ghislain: I'd recommend giving zfsonlinux a spin on your personal desktop/notebook to get a feel for it 1473607285 M * Guy- note though that there is a significant issue with zfs vs. linux-vserver: you can't get a guest to see a newly created zfs instance without restarting the guest 1473608516 J * LongyanG ~long@15255.s.t4vps.eu 1473608725 Q * Long_yanG Ping timeout: 480 seconds 1473613768 M * Ghislain i think i will try it on personal workstation first then test more 1473616787 J * thierryp_ ~thierry@2a01:e35:2e2b:e2c0:d408:e2c3:89c4:fcf5 1473617224 Q * thierryp Ping timeout: 480 seconds 1473621062 J * hijacker ~hijacker@52D9A971.cm-11-1c.dynamic.ziggo.nl 1473621113 Q * hijacker 1473628172 Q * Ghislain Quit: Leaving. 1473629215 Q * bonbons Quit: Leaving 1473630121 Q * thierryp_ Quit: ciao folks 1473630193 J * thierryp ~thierry@2a01:e35:2e2b:e2c0:2500:44b3:5c1c:e055 1473631035 Q * thierryp Quit: ciao folks 1473636130 J * Ghislain ~aqueos@adsl1.aqueos.com 1473636226 Q * Ghislain