|
| 1 | +--- |
| 2 | +marp: true |
| 3 | +theme: metal-stack |
| 4 | +paginate: true |
| 5 | +footer: Gerrit Schwerthelm – x-cellent technologies GmbH — metal-stack Training |
| 6 | +backgroundImage: url("https://metal-stack.io/images/shape/banner.png") |
| 7 | +--- |
| 8 | +<!-- _class: cover lead --> |
| 9 | + |
| 10 | + |
| 11 | + |
| 12 | +--- |
| 13 | +<!-- _class: cover lead --> |
| 14 | + |
| 15 | +# Multi-Partition-Layout |
| 16 | + |
| 17 | +--- |
| 18 | +<!-- |
| 19 | +_class: lead |
| 20 | +_backgroundColor: #1f1f1f |
| 21 | +_backgroundImage: |
| 22 | +_footer: "" |
| 23 | +--> |
| 24 | + |
| 25 | + |
| 26 | +--- |
| 27 | +<!-- |
| 28 | +_class: lead |
| 29 | +_backgroundColor: #1f1f1f |
| 30 | +_backgroundImage: |
| 31 | +_footer: "" |
| 32 | +--> |
| 33 | + |
| 34 | + |
| 35 | +--- |
| 36 | +<style>section { font-size: 30px; }</style> |
| 37 | + |
| 38 | +# Multi-Partition-Layout Properties |
| 39 | + |
| 40 | + |
| 41 | +- Fully independent locations with own storage and own node networks |
| 42 | +- Clusters can only be created independent in every location |
| 43 | + - Failover mechanism for deployed applications requires duplicated deployments, which can serve indepedently |
| 44 | + - Failover through BGP |
| 45 | +- If cluster nodes are spread across partitions (not implemented yet), nodes will not be able to reach each other |
| 46 | + - Would require an overlay network for inter-node-communication |
| 47 | + |
| 48 | +--- |
| 49 | +<!-- _class: cover lead --> |
| 50 | + |
| 51 | +# Single-Partition-Layout |
| 52 | + |
| 53 | +--- |
| 54 | +<!-- |
| 55 | +_class: lead |
| 56 | +_backgroundColor: #1f1f1f |
| 57 | +_backgroundImage: |
| 58 | +_footer: "" |
| 59 | +--> |
| 60 | + |
| 61 | + |
| 62 | +--- |
| 63 | +<style>section { font-size: 30px; }</style> |
| 64 | + |
| 65 | +# Single-Partition-Layout Properties |
| 66 | + |
| 67 | +- Multiple groups of racks at multiple locations but connected to same CLOS topology |
| 68 | +- All racks can connect to the same storage network |
| 69 | +- Nodes in private networks can communicate |
| 70 | +- When creating a cluster, nodes will be randomly spread across the racks |
| 71 | + - Possible improvement of this situation, see `MEP-12: Rack Spreading` |
| 72 | + |
| 73 | +--- |
| 74 | + |
| 75 | +# MEP-12: Rack Spreading |
| 76 | + |
| 77 | +- Instead of selecting a machine from a machine pool randomly |
| 78 | +- Get all existing machines in the same project and count to which rack they belong |
| 79 | +- Place machine on the rack with the least amount of machines already allocated |
| 80 | +- Best effort only |
0 commit comments