Networking with a Purpose (#4) – Final Wrap Up
Mon, 09 Sep 2024 05:00:00 -0000
|Read Time: 0 minutes
Introduction
Welcome to final installment of the Networking with a Purpose blog series. The intention of this series was to provide information for making decisions before designing and deploying a PowerStore appliance. In this series, we’ve discussed topics ranging from PowerStore hardware basics to advanced high availability features, and storage and replication networks. While the blogs were written for new deployments, the information presented may help justify changes to existing deployments also. This may happen with the new features included in PowerStoreOS 4.0. For those of you that are still unsure where to start and need some additional guidance in your design choices, the information below will hopefully answer any remaining questions you have.
What we do know is that every environment is different, so I can’t just say do x, y, and z to have the best configuration. There are just too many unique factors to consider. With this said, there are common areas that apply to every configuration. Before writing this final blog, I asked my peers where they would start if they needed to plan the connectivity of a PowerStore. The most common response I received was to start with special configurations. While this is a broad statement, it really comes back to the must haves and things that need to be planned around, such as the network, replication, and file connectivity. Because everyone seems to love lists, below are the top six things to review when planning the network configuration of a PowerStore.
#1 Network Environment
The first thing to review when planning connectivity of a PowerStore is the existing infrastructure and what the PowerStore will connect to. This is not just fibre channel or ethernet and speeds that will be used, but how many different networks that are present. If connectivity to multiple networks on different/dedicated hardware is needed, as shown in the figure below, then connections to each node from each network is required. If there is just a single network, then there is less to worry about.
#2 Fibre Channel Connectivity
The next decision is if the PowerStore will require fibre channel connectivity or not. Fibre channel within a PowerStore appliance is used for block host access and migrations (data import). Immediately, this answer determines if an optional IO module slot needs to be reserved for a 32Gb FC module. If the environment is primarily FC, then maybe both IO module locations will include FC. This decision entirely depends on the mix of connectivity types and how many FC hosts there are compared to ethernet-based hosts. Utilizing both IO module locations for FC still leaves the four-port card for ethernet connectivity. Another thing to keep in mind is that PowerStore can import data over fibre channel from a source system. If this required even once, then an FC IO module is required and it must reside in slot 0. If no FC will be used, then it’s one less technology to worry about.
#3 Ethernet Connectivity
Ethernet connectivity is used for block and file access, along with replication, migration (import), and remote backup integration with Storage Direct, PowerStore has multiple connectivity options available including 100GbE optical, 25/10GbE optical (supporting SFP+ optical connections and twinax copper connections), and 10GbE copper. The front-end connectivity needs to match what is required to connect to the network. If there is a single network that has multiple connectivity options to access it, then go with the fastest connectivity option available. For instance, in my lab we have 10GbE copper and 25GbE optical switches on the same physical network. This network setup allows us to configure access from any of our hosts to any of our current and legacy storage systems, regardless if they use 10Gb or 25Gb. If I were to order a new PowerStore for this network as if it was production, I would go with 25Gb front-end ports and skip adding both 10GbE and 25 GbE, although that may not be possible for everyone. If there are separate networks, then multiple front-end connectivity types may be required.
When choosing ethernet connectivity, don’t forget that the four-port card supports either 25/10GbE optical or 10GbE copper. These cards come in multiple versions, four-part card and IO module, unlike 100GbE optical and fibre channel which only come in the IO module variant. 100GbE and fibre channel connectivity should be planned and decided first, and then the addition of 25/10GbE optical or 10GbE copper connectivity.
#4 Replication
When planning the connectivity of a system, replication should also be reviewed. When synchronous or metro replication is used, it is suggested for performance reasons to use a dedicated port for replication as all writes must be written to both the local and remote system before the host is acknowledged. This helps separate host and replication traffic and removes contention. In some instances, remote sites are connected using a dedicated network between them. The connections between the sites may just be used for replication, and host connectivity would occur over a different network. In this case, a port on each node with the correct connectivity type must be dedicated for replication. Even if the network connection to the remote system isn’t dedicated, the decision to dedicate one of more ports to PowerStore replication traffic is an option.
#5 File Connectivity
Another consideration is with file. If this is the first appliance within a cluster, the appliance can be Unified, which allows the appliance to host both file, block, and Virtual Volume (vVol) workloads. The other option is Block Optimized, which allows hosting of just block and vVol workloads. If file will be utilized, a few items need to be reviewed. First, is the network physically separated into a block network and a file network. If there are multiple networks, then port connectivity needs to be appropriately planned and configured for each network. If there is a single network, the configuration is easier, except that file connectivity needs to occur over a bond or fail-safe network configuration. If minimal file resources are being created, then utilizing the default bond may be enough. If file needs more ports, then one or more bonds or FSNs can be created. The configuration of the bonded ports also needs to be planned as they can be in an active/active or active/passive configuration as discussed in Networking with a Purpose (#2) – HA To Save The Day. It is suggested that you use different physical ports for both file and block IO when possible, but it isn’t required. NAS server interfaces also require a VLAN that is not in common with any block storage network on the cluster.
#6 High Availability
A resilient network and storage design eliminates issues caused by single points of failure. You don’t want a faulty component or something else to cause an outage in the environment. As a best practice, each individual network should have multiple paths to each node. For example, if the network has two switches, then each switch should have a path to Node A and a path to Node B at a minimum, as outlined in the figure below. This ensures that any single failure or issue in the data path will not impact access to a node. When multiple physical networks are involved, this can quickly use up many front-end ports, which is not a problem. It is just a byproduct of the configuration. It is also a best practice to cable and utilize all ports within a PowerStore appliance from the start. Spreading the traffic across all ports of the system evenly helps to spread out the workload and achieve the best performance.
In Networking with a Purpose (#2) – HA To Save The Day (see link below), bonds and fail-safe networks were discussed. While bonds are required for file, iSCSI and replication storage networks can also be mapped to a bond. Creating multiple user-defined bonds on the system can also help spread out NAS servers and therefore spread-out file workloads. When configuring bonds on PowerStore, it is suggested to configure link aggregation across the corresponding switch ports to ensure the ports are active/active. It is also suggested to choose PowerStore ports that come from different IO modules. This is also done for redundancy. One thing to note however, user-defined bonds do not support NVMe/TCP. If NVMe/TCP will be used, then bond0 or individual ports tagged with a storage network must be utilized.
When creating storage networks, user-defined bonds, or fail-safe networks, PowerStore will always create the same configuration on the peer node. Say you are tagging a storage network on Node A IOM 0 FEPorts 0 and 1. The system will automatically tag the same network onto the exact same ports on Node B. This helps to keep the configuration symmetrical across the nodes.
Conclusion
And so, you have reached the end of the Networking with a Purpose series. Whether this is your first look at PowerStore, or you already have it deployed, I hope the information provided in this series helped in whatever part of the process you are in. While this is planned to be the final installment of the series, who knows, maybe future enhancements will require a revival and expansion of the series. And for those of you still reading, you’re still here? It’s over. Go home!
Resources
- Networking with a Purpose (#1) – Let’s Get Physical
- Networking with a Purpose (#2) – HA To Save The Day
- Networking with a Purpose (#3) – What’s My Purpose?
- PowerStore InfoHub
- Dell PowerStore: Best Practices Guide
Author: Ryan Poulin, Principal Engineering Technologist