[one-users] OpenNebula Drain CPU
Hello List, I have installed my servers in datacenter last Friday: I have a Cloud Controller made on old Xeon quadcore (a dell PowerEdge 860) with 6 GB RAM and 2x2TB SATA disks and I use it only for storage via MooseFS and to run OpenNebula daemon; I have also 2 KVM nodes that use VM raw disk on MooseFS daemon. I plan to move the datas and some VM from other 2 old PE860 on this OpenNebula infrastructure and, after that, use the other 2 servers to add MooseFS nodes to the storage cluster to enable HA. At the moment I have 10 VM running on this infrastructure, all ready but with the only load of incoming emails (no DNS, no web sites). Without OpenNebula runnung, I have a load under 1 process; CPU is 70% idle and 25% waiting for disk. This morning, (yesterday was OK) with OpenNebula running I have a load about 9-10 process; CPU is 0% idle, 0% waiting for disk, 60% system and 35% nice. RAM is OK: 1 GB used, 1 GB free an 4 GB cached; no swap at all. I checked the system for possible rootkits ecc., I checked logs (debug level 3) and I tried to move out one.db and restart OpenNebula but the load is always high. Nokogiri gem is installed. Sunstone, radacctd, econe, ecc. not running, only One daemon. When I start OpenNebula, if I run netstat, I see many connection in Time Wait state to MooseFS. In this situation I don't unterstand if it is a MooseFS problem or a OpenNebula problem, but without OpenNebula running all KVM VM works well, no other problems. What can I do to debug (and solve) this? Thanks, Alberto -- Alberto Zuin via Mare, 36/A 36030 Lugo di Vicenza (VI) Italy P.I. 04310790284 Tel. +39.0499271575 Fax. +39.0492106654 Cell. +39.3286268626 www.azns.it - albe...@azns.it ___ Users mailing list Users@lists.opennebula.org http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
Re: [one-users] contribution of LVM2 transfer manager driver for opennebula 3.4.x
On 2012-06-30 19:14, Shankhadeep Shome wrote: Will this work with 3.6? Is there any reason clustered lvm is required over I didn't try yet 3.6, but as seen in release notes for beta version, there are no changes in transfer manager drivers part, so it should work (but please test before using in production). For normal LVM with shared disks (SAN), there I have no experience. I suspect, that in this case probably mv script could be improved, by not copying from one node to another in some cases (and/or not creating the copy in file for STOP). Regards, Rolandas Naujikas normal lvm with shared disks? I've been using glusterfs for shared storage but it's been unstable and I've lost data with high i/o. Shank On Tue, Jun 26, 2012 at 6:58 AM, Rolandas Naujikas rolandas.nauji...@mif.vu.lt wrote: Hi, Because Debian 6.0 Xen doesn't support tap:aio: and because LVM disks are faster, I wrote modified transfer manager driver for opennebula 3.4.x, that use LVM volumes on local disks in virtualization hosts. There are 3 kinds: *) lvm2 - works with shared or not shared filesystem datastore (for system datastore there is parameter in lvm.conf to tell shared or not it is). *) lvm2ssh - the same, but removed code to detect that datastore is shared. *) lvm2shared - the same, but with only assumption, than datastore is shared. URL of all them is at http://mif.vu.lt/~rolnas/one/**one34/tm/http://mif.vu.lt/~rolnas/one/one34/tm/ Regards, Rolandas Naujikas P.S. This driver was in almost working condition in opennebula 3.2, but was lost in opennebula 3.4. lvm driver in opennebula 3.4.1 is not for system datastore and is different by design. __**_ Users mailing list Users@lists.opennebula.org http://lists.opennebula.org/**listinfo.cgi/users-opennebula.**orghttp://lists.opennebula.org/listinfo.cgi/users-opennebula.org ___ Users mailing list Users@lists.opennebula.org http://lists.opennebula.org/listinfo.cgi/users-opennebula.org