You understood my question correctly, thanks for the explanation. Boris, I was able to force the traffic to go out only through the cluster network by making the first machine have OSDs only of the Primary type and the other machines only Secondary. It worked as intended on writing, but reading only occurred from the primary OSDs, obviously this would be the correct behavior for Ceph, but is there any way to split this workload at read time? Em qui., 29 de set. de 2022 às 04:27, Boris Behrens <bb@xxxxxxxxx> escreveu: > Hi Murilo, > > as far as I understand ceph: > You connect via NFS to a radosgw. When sending data to the rgw instance > (uploading files via NFS), the RGW instance talks to the primary OSDs for > the required placementgroups through the public network. To primary OSDs > talk to their replicas via the cluster network. > > Hope I understood ceph correctly and your question :) > > Cheers > Boris > > Am Do., 29. Sept. 2022 um 04:11 Uhr schrieb Murilo Morais < > murilo@xxxxxxxxxxxxxx>: > > > Good evening everyone. > > > > I setup a cluster with three machines, each with two network interfaces, > > one for the public network and one for the cluster network ( > 172.25.50.0/24 > > for public and 10.10.10.0/24 for cluster). All machines see each other > and > > are communicable in their respective networks. So far everything is > > correct. > > > > I'm not properly understanding how the traffic is working. I know there > are > > countless places explaining what each network in Ceph is for, but I'm > still > > confused. > > > > For example: I created an NFS on the first machine and mounted it on > > another machine and started writing a lot of data, at the same time the > > other machines in the cluster received the replicas. > > > > What I'm not understanding is that part of the traffic for all machines > in > > the cluster goes out through the public network and also through the > > cluster network. I'm getting all the initial traffic on the port of the > > first machine and some of this traffic is going out the same port and > also > > the port of the cluster network. > > > > Is this behavior correct? Wouldn't it be correct for this traffic that is > > going out through the public network to leave through the cluster > network? > > Is this traffic going out over the public network replication traffic? > > _______________________________________________ > > ceph-users mailing list -- ceph-users@xxxxxxx > > To unsubscribe send an email to ceph-users-leave@xxxxxxx > > > > > -- > Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend im > groüen Saal. > _______________________________________________ > ceph-users mailing list -- ceph-users@xxxxxxx > To unsubscribe send an email to ceph-users-leave@xxxxxxx > Em qui., 29 de set. de 2022 às 04:27, Boris Behrens <bb@xxxxxxxxx> escreveu: > Hi Murilo, > > as far as I understand ceph: > You connect via NFS to a radosgw. When sending data to the rgw instance > (uploading files via NFS), the RGW instance talks to the primary OSDs for > the required placementgroups through the public network. To primary OSDs > talk to their replicas via the cluster network. > > Hope I understood ceph correctly and your question :) > > Cheers > Boris > > Am Do., 29. Sept. 2022 um 04:11 Uhr schrieb Murilo Morais < > murilo@xxxxxxxxxxxxxx>: > > > Good evening everyone. > > > > I setup a cluster with three machines, each with two network interfaces, > > one for the public network and one for the cluster network ( > 172.25.50.0/24 > > for public and 10.10.10.0/24 for cluster). All machines see each other > and > > are communicable in their respective networks. So far everything is > > correct. > > > > I'm not properly understanding how the traffic is working. I know there > are > > countless places explaining what each network in Ceph is for, but I'm > still > > confused. > > > > For example: I created an NFS on the first machine and mounted it on > > another machine and started writing a lot of data, at the same time the > > other machines in the cluster received the replicas. > > > > What I'm not understanding is that part of the traffic for all machines > in > > the cluster goes out through the public network and also through the > > cluster network. I'm getting all the initial traffic on the port of the > > first machine and some of this traffic is going out the same port and > also > > the port of the cluster network. > > > > Is this behavior correct? Wouldn't it be correct for this traffic that is > > going out through the public network to leave through the cluster > network? > > Is this traffic going out over the public network replication traffic? > > _______________________________________________ > > ceph-users mailing list -- ceph-users@xxxxxxx > > To unsubscribe send an email to ceph-users-leave@xxxxxxx > > > > > -- > Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend im > groüen Saal. > _______________________________________________ > ceph-users mailing list -- ceph-users@xxxxxxx > To unsubscribe send an email to ceph-users-leave@xxxxxxx > _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |