Re: [ceph-users] Public Network Meaning
Public network is clients-to-OSD traffic - and if you have NOT explicitely defined cluster network, than also OSD-to-OSD replication takes place over same network. Otherwise, you can define public and cluster(private) network - so OSD replication will happen over dedicated NICs (cluster network) and thus speed up. If i.e. replica count on pool is 3, that means, each 1GB of data writen to some particualr OSD, will generate 3 x 1GB of more writes, to the replicas... - which ideally will take place over separate NICs to speed up things... On 14 March 2015 at 17:43, Georgios Dimitrakakis gior...@acmac.uoc.gr wrote: Hi all!! What is the meaning of public_network in ceph.conf? Is it the network that OSDs are talking and transferring data? I have two nodes with two IP addresses each. One for internal network 192.168.1.0/24 and one external 15.12.6.* I see the following in my logs: osd.0 is down since epoch 2204, last address 15.12.6.21:6826/33094 osd.1 is down since epoch 2206, last address 15.12.6.21:6817/32463 osd.2 is down since epoch 2198, last address 15.12.6.21:6843/34921 osd.3 is down since epoch 2200, last address 15.12.6.21:6838/34208 osd.4 is down since epoch 2202, last address 15.12.6.21:6831/33610 osd.5 is down since epoch 2194, last address 15.12.6.21:6858/35948 osd.7 is down since epoch 2192, last address 15.12.6.21:6871/36720 osd.8 is down since epoch 2196, last address 15.12.6.21:6855/35354 I 've managed to add a second node and during rebalancing I see that data is transfered through the internal 192.* but the external link is also saturated! What is being transferred from that? Any help much appreciated! Regards, George ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Andrija Panić ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
[ceph-users] Public Network Meaning
Hi all!! What is the meaning of public_network in ceph.conf? Is it the network that OSDs are talking and transferring data? I have two nodes with two IP addresses each. One for internal network 192.168.1.0/24 and one external 15.12.6.* I see the following in my logs: osd.0 is down since epoch 2204, last address 15.12.6.21:6826/33094 osd.1 is down since epoch 2206, last address 15.12.6.21:6817/32463 osd.2 is down since epoch 2198, last address 15.12.6.21:6843/34921 osd.3 is down since epoch 2200, last address 15.12.6.21:6838/34208 osd.4 is down since epoch 2202, last address 15.12.6.21:6831/33610 osd.5 is down since epoch 2194, last address 15.12.6.21:6858/35948 osd.7 is down since epoch 2192, last address 15.12.6.21:6871/36720 osd.8 is down since epoch 2196, last address 15.12.6.21:6855/35354 I 've managed to add a second node and during rebalancing I see that data is transfered through the internal 192.* but the external link is also saturated! What is being transferred from that? Any help much appreciated! Regards, George ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Public Network Meaning
I thought that it was easy but apparently it's not! I have the following in my conf file mon_host = 192.168.1.100,192.168.1.101,192.168.1.102 public_network = 192.168.1.0/24 mon_initial_members = fu,rai,jin but still the 15.12.6.21 link is being saturated Any ideas why??? Should I put cluster network as well?? Should I put each OSD in the CONF file??? Regards, George Andrija, thanks a lot for the useful info! I would also like to thank Kingrat at the IRC channel for his useful advice! I was under the wrong impression that public is the one used for RADOS. So I thought that public=external=internet and therefore I used that one in my conf. I understand now that I should have specified in CEPH Public's Network what I call internal and which is the one that all machines are talking directly to each other. Thanks you all for the feedback! Regards, George Public network is clients-to-OSD traffic - and if you have NOT explicitely defined cluster network, than also OSD-to-OSD replication takes place over same network. Otherwise, you can define public and cluster(private) network - so OSD replication will happen over dedicated NICs (cluster network) and thus speed up. If i.e. replica count on pool is 3, that means, each 1GB of data writen to some particualr OSD, will generate 3 x 1GB of more writes, to the replicas... - which ideally will take place over separate NICs to speed up things... On 14 March 2015 at 17:43, Georgios Dimitrakakis wrote: Hi all!! What is the meaning of public_network in ceph.conf? Is it the network that OSDs are talking and transferring data? I have two nodes with two IP addresses each. One for internal network MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 192.168.1.0/24 [1] and one external 15.12.6.* I see the following in my logs: osd.0 is down since epoch 2204, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6826/33094 [2] osd.1 is down since epoch 2206, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6817/32463 [3] osd.2 is down since epoch 2198, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6843/34921 [4] osd.3 is down since epoch 2200, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6838/34208 [5] osd.4 is down since epoch 2202, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6831/33610 [6] osd.5 is down since epoch 2194, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6858/35948 [7] osd.7 is down since epoch 2192, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6871/36720 [8] osd.8 is down since epoch 2196, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6855/35354 [9] I ve managed to add a second node and during rebalancing I see that data is transfered through the internal 192.* but the external link is also saturated! What is being transferred from that? Any help much appreciated! Regards, George ___ ceph-users mailing list ceph-users@lists.ceph.com [10] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com [11] -- Andrija Panić Links: -- [1] http://192.168.1.0/24 [2] http://15.12.6.21:6826/33094 [3] http://15.12.6.21:6817/32463 [4] http://15.12.6.21:6843/34921 [5] http://15.12.6.21:6838/34208 [6] http://15.12.6.21:6831/33610 [7] http://15.12.6.21:6858/35948 [8] http://15.12.6.21:6871/36720 [9] http://15.12.6.21:6855/35354 [10] mailto:ceph-users@lists.ceph.com [11] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com [12] mailto:gior...@acmac.uoc.gr ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Public Network Meaning
This is how I did it, and then retart each OSD one by one, but monritor with ceph -s, when ceph is healthy, proceed with next OSD restart... Make sure the networks are fine on physical nodes, that you can ping in between... [global] x x x x x x # ### REPLICATION NETWORK ON SEPARATE 10G NICs # replication network cluster network = 10.44.251.0/24 # public/client network public network = 10.44.253.0/16 # [mon.xx] mon_addr = x.x.x.x:6789 host = xx [mon.yy] mon_addr = x.x.x.x:6789 host = yy [mon.zz] mon_addr = x.x.x.x:6789 host = zz On 14 March 2015 at 19:14, Georgios Dimitrakakis gior...@acmac.uoc.gr wrote: I thought that it was easy but apparently it's not! I have the following in my conf file mon_host = 192.168.1.100,192.168.1.101,192.168.1.102 public_network = 192.168.1.0/24 mon_initial_members = fu,rai,jin but still the 15.12.6.21 link is being saturated Any ideas why??? Should I put cluster network as well?? Should I put each OSD in the CONF file??? Regards, George Andrija, thanks a lot for the useful info! I would also like to thank Kingrat at the IRC channel for his useful advice! I was under the wrong impression that public is the one used for RADOS. So I thought that public=external=internet and therefore I used that one in my conf. I understand now that I should have specified in CEPH Public's Network what I call internal and which is the one that all machines are talking directly to each other. Thanks you all for the feedback! Regards, George Public network is clients-to-OSD traffic - and if you have NOT explicitely defined cluster network, than also OSD-to-OSD replication takes place over same network. Otherwise, you can define public and cluster(private) network - so OSD replication will happen over dedicated NICs (cluster network) and thus speed up. If i.e. replica count on pool is 3, that means, each 1GB of data writen to some particualr OSD, will generate 3 x 1GB of more writes, to the replicas... - which ideally will take place over separate NICs to speed up things... On 14 March 2015 at 17:43, Georgios Dimitrakakis wrote: Hi all!! What is the meaning of public_network in ceph.conf? Is it the network that OSDs are talking and transferring data? I have two nodes with two IP addresses each. One for internal network MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 192.168.1.0/24 [1] and one external 15.12.6.* I see the following in my logs: osd.0 is down since epoch 2204, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6826/33094 [2] osd.1 is down since epoch 2206, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6817/32463 [3] osd.2 is down since epoch 2198, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6843/34921 [4] osd.3 is down since epoch 2200, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6838/34208 [5] osd.4 is down since epoch 2202, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6831/33610 [6] osd.5 is down since epoch 2194, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6858/35948 [7] osd.7 is down since epoch 2192, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6871/36720 [8] osd.8 is down since epoch 2196, last address MAILSCANNER WARNING: NUMERICAL LINKS ARE OFTEN MALICIOUS: 15.12.6.21:6855/35354 [9] I ve managed to add a second node and during rebalancing I see that data is transfered through the internal 192.* but the external link is also saturated! What is being transferred from that? Any help much appreciated! Regards, George ___ ceph-users mailing list ceph-users@lists.ceph.com [10] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com [11] -- Andrija Panić Links: -- [1] http://192.168.1.0/24 [2] http://15.12.6.21:6826/33094 [3] http://15.12.6.21:6817/32463 [4] http://15.12.6.21:6843/34921 [5] http://15.12.6.21:6838/34208 [6] http://15.12.6.21:6831/33610 [7] http://15.12.6.21:6858/35948 [8] http://15.12.6.21:6871/36720 [9] http://15.12.6.21:6855/35354 [10] mailto:ceph-users@lists.ceph.com [11] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com [12] mailto:gior...@acmac.uoc.gr ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Andrija Panić ___ ceph-users mailing list ceph-users@lists.ceph.com