ISCSI disks and Virtual IPs
stevehughes
10 Posts
May 18, 2020, 5:25 amQuote from stevehughes on May 18, 2020, 5:25 amIt seems that each path for each iSCSI disk gets a unique Virtual IP. So if we have say 10 disks (correlating to 10 vSphere datastores) with 3 paths each we need 30 VIPs, which all need to be configured in the ESXi initiator on each host.
Is it essential that every path have its own IP? For a small system with 3 nodes it would seem enough to have one VIP on each node and use that for all disks. Mind you I can also see that in a larger system with say 20+ nodes there would be great benefit in being able to spread the VIPs for each disk across the entire cluster of hosts.
Also, at what point does the benefit of additional paths start to drop off? If a disk has two VIPs on PetaSAN and we have two iSCSI initiator NICs on the host that gives us 4 paths to the disk. Is there much point in going more than this?
It seems that each path for each iSCSI disk gets a unique Virtual IP. So if we have say 10 disks (correlating to 10 vSphere datastores) with 3 paths each we need 30 VIPs, which all need to be configured in the ESXi initiator on each host.
Is it essential that every path have its own IP? For a small system with 3 nodes it would seem enough to have one VIP on each node and use that for all disks. Mind you I can also see that in a larger system with say 20+ nodes there would be great benefit in being able to spread the VIPs for each disk across the entire cluster of hosts.
Also, at what point does the benefit of additional paths start to drop off? If a disk has two VIPs on PetaSAN and we have two iSCSI initiator NICs on the host that gives us 4 paths to the disk. Is there much point in going more than this?
admin
2,930 Posts
May 18, 2020, 12:02 pmQuote from admin on May 18, 2020, 12:02 pmWe allow many paths per disk to support active/active, where a disk is served actively by several nodes, most iSCSI SANs do not support this and an active iSCSI disk is served by only 1 server at a time. It also better in case of failover, when a node fails you can distribute its load to many other nodes and not just to a single node.
If you have 20 nodes, you will note create disks with 20 paths each, maybe 2 or 4 each. For scale-out, you do not want the servers to each serve all disks
We allow many paths per disk to support active/active, where a disk is served actively by several nodes, most iSCSI SANs do not support this and an active iSCSI disk is served by only 1 server at a time. It also better in case of failover, when a node fails you can distribute its load to many other nodes and not just to a single node.
If you have 20 nodes, you will note create disks with 20 paths each, maybe 2 or 4 each. For scale-out, you do not want the servers to each serve all disks
Last edited on May 18, 2020, 2:30 pm by admin · #2
stevehughes
10 Posts
May 18, 2020, 10:21 pmQuote from stevehughes on May 18, 2020, 10:21 pmThanks for the info.
Thanks for the info.
ISCSI disks and Virtual IPs
stevehughes
10 Posts
Quote from stevehughes on May 18, 2020, 5:25 amIt seems that each path for each iSCSI disk gets a unique Virtual IP. So if we have say 10 disks (correlating to 10 vSphere datastores) with 3 paths each we need 30 VIPs, which all need to be configured in the ESXi initiator on each host.
Is it essential that every path have its own IP? For a small system with 3 nodes it would seem enough to have one VIP on each node and use that for all disks. Mind you I can also see that in a larger system with say 20+ nodes there would be great benefit in being able to spread the VIPs for each disk across the entire cluster of hosts.
Also, at what point does the benefit of additional paths start to drop off? If a disk has two VIPs on PetaSAN and we have two iSCSI initiator NICs on the host that gives us 4 paths to the disk. Is there much point in going more than this?
It seems that each path for each iSCSI disk gets a unique Virtual IP. So if we have say 10 disks (correlating to 10 vSphere datastores) with 3 paths each we need 30 VIPs, which all need to be configured in the ESXi initiator on each host.
Is it essential that every path have its own IP? For a small system with 3 nodes it would seem enough to have one VIP on each node and use that for all disks. Mind you I can also see that in a larger system with say 20+ nodes there would be great benefit in being able to spread the VIPs for each disk across the entire cluster of hosts.
Also, at what point does the benefit of additional paths start to drop off? If a disk has two VIPs on PetaSAN and we have two iSCSI initiator NICs on the host that gives us 4 paths to the disk. Is there much point in going more than this?
admin
2,930 Posts
Quote from admin on May 18, 2020, 12:02 pmWe allow many paths per disk to support active/active, where a disk is served actively by several nodes, most iSCSI SANs do not support this and an active iSCSI disk is served by only 1 server at a time. It also better in case of failover, when a node fails you can distribute its load to many other nodes and not just to a single node.
If you have 20 nodes, you will note create disks with 20 paths each, maybe 2 or 4 each. For scale-out, you do not want the servers to each serve all disks
We allow many paths per disk to support active/active, where a disk is served actively by several nodes, most iSCSI SANs do not support this and an active iSCSI disk is served by only 1 server at a time. It also better in case of failover, when a node fails you can distribute its load to many other nodes and not just to a single node.
If you have 20 nodes, you will note create disks with 20 paths each, maybe 2 or 4 each. For scale-out, you do not want the servers to each serve all disks
stevehughes
10 Posts
Quote from stevehughes on May 18, 2020, 10:21 pmThanks for the info.
Thanks for the info.