 Hello, and welcome to the Windows Server Summit. This session is on new storage features in Windows Server 2025. My name is Dan Kumo, and I'm a principal program manager on the Core OS Storage and File Systems team. In this session, we'll run through the storage features you can expect in Windows Server 2025. There are some new improvements we've been building over the past few years that you may have seen if you're using Azure Stack HCI. If not, this will be all new to you, so we encourage you to review the features you see here and consider if Windows Server 2025 is right for you and your business. Of course, please follow up with us if you have any questions. Now, on to the agenda. Before we talk about the improvements, we'll recap the available storage architectures so that you can visualize how you'll apply these new features in your environment. Next, we'll talk about some of the latest innovations, including new storage protocols that lower latency with NVMe storage and NVMe over fabric's network transports. Then, we'll talk about improvements in storage spaces, such as deduplication and compression on our resilient file system, also known as REFS. We'll talk about thin provisioning to more efficiently provision a volume without allocating the storage space until it's needed. And finally, we'll talk about how to optimize the storage repair and resynchronization on your storage space's direct system to ensure your workload's performance. If you're not familiar with our platform, there are a variety of architectures you can choose from to fit your particular scenario. Let's take a look at some of the scenarios that we already support. Then we'll cover some of the improvements that enhance these scenarios in Windows Server 2025. You can run Hyper-V on a standalone server without a cluster. This could be a tower or rack mount server, and you'll have a full complement of storage technologies available, so you can choose the best option that fits you and your business. This includes a standalone server with local storage where the drives are inside the tower or rack mounted server chassis. These could be hard disk drives, solid state drives, or NVMe SSDs. If you want to centralize your storage for several standalone Hyper-V systems, we support file server access for virtual machines, whether that be an SMB share from another Windows Server, or a network attached storage system from your preferred storage vendor. Lastly, you can use traditional SAN storage from an iSCSI, Fiber Channel, or NVMe Target. This allows you to use an existing investment in SAN technology centralizing your data storage if you choose. Now on to the clustered options. For multi-node deployments where you want to improve workload availability, you can also cluster your systems using a variety of architectures. Let's take a look at the cluster storage options. As you can see, we have several unique options to fit your particular scenario. The first is a design called hyperconverged, which couples virtual machines on the same server as the physical storage resides. The storage is then replicated between the nodes in the cluster to ensure storage resiliency similar to traditional RAID options you're likely familiar with. This hyperconverged architecture can scale from one to 16 nodes, enabling you to grow your cluster in a modular form alongside your business growth. Of course, your business sometimes grows asymmetrically. That is, memory and CPU requirements do not always grow at the same rate as your storage. This scale-out architecture allows you to scale the resources you need independently, whether that be at the compute or storage tiers. You can safely scale one tier without needing to also expand the other. This architecture can use different file-based protocols depending on the needs of your application, like SMB over TCP or SMB Direct using RDMA for workloads needing the highest performance guarantees. This approach can enable low-cost, centralized storage for many virtual machine clusters which simplifies storage administration, including backup. Finally, this disaggregated architecture lowers licensing costs as well, particularly when compared to other subscription-based models which required licensing all CPU cores used for virtual machines. For example, the top row is the compute tier that runs your virtual machines and would leverage a data center server license that simultaneously licenses all Windows virtual machines. Whereas the bottom row is the storage tier and would use a lower-cost, standard Windows server license and doesn't need the virtualization benefits afforded to the data center license. Of course, if you've invested in a sand, you'll no doubt want to continue using your investment. Microsoft embraces customer choice from any vendor and provides several options for connection, including iSCSI, Fiber Channel, and Sune NVMeover Fabrics. Like the scale-out option, this allows you to connect multiple compute clusters to the same sand storage. It also leverages benefits of centralized storage like backup. That was a high-level overview of the possible architectures you can use with Windows Server 2025. As a recap, you can run your workloads in a standalone tower or rack mount system or choose from one of several clustered options for a high availability experience. Now let's take a look at some of the new or improved features in Windows Server 2025. The first we'll talk about today is NVMe and NVMeover Fabrics. As a client or initiator of storage, Windows has supported the SCSI and iSCSI protocols for a long time. Now, a new standard for extremely fast solid-state drives called NVMe has emerged. By 2027, Gartner estimates that at least 25% of enterprises will deploy NVMeover Fabric solutions. In Windows Server 2025, we're introducing full NVMe support in the storage stack to ensure that you can leverage the full performance benefits of your storage infrastructure. This brings a 70% improvement in IOPS over Windows Server 2022. And that's not all. We're continuing to innovate and improve our NVMe support that will eventually bring that IOPS improvement to over 90% of that of Windows Server 2022. In addition, we're also going to bring an NVMeover Fabrics initiator for connecting your Windows Server 2025 system to highly performant but disaggregated storage infrastructure. When we officially launch our NVMe support following Windows Server 2025 launch, our NVMeover Fabric client will support TCP, which will enable you to connect to your NVMe-based sands. Eventually, we plan to optimize this with an RDMA-based network support for your highest performing applications. Now, let's take a look at deduplication and compression features in Windows Server 2025. Deduplication and compression enable significant storage savings for workloads like virtualization, backups, and other file server workloads. This includes Azure Virtual Desktop on Azure Stack HCI. We've made this easy to manage through Windows Admin Center or PowerShell Commandlets for whatever experience you might prefer. Let's take a look at this demo to see how easy it is to manage deduplication and compression with Windows Admin Center. On the Volumes tab in Windows Admin Center, you can click on Settings to open up the RFS deduplication and compression options. Here you can enable it and set a schedule, including days of the week and the start date. You can also set a maximum duration for the job to run. Note on the screen that the use capacity is at 1.79 terabytes and the available space is 19.2 terabytes. After the job runs for the first time and the page is refreshed, you can see that we saved around 75% of the previously consumed space and returned it to the available pool space. Deduplication works in a cluster and it consumes very low overhead, adding less than one millisecond latency to storage access times. In addition, we only re-scan new or changed data to ensure we don't negatively add to the deduplication times which would otherwise adversely affect VM performance without yielding any additional savings. We enable several options for you to choose how deduplication and compression runs, including deduplication only, compression only, or both, based on your personal needs. If you choose compression, you have two modes of operation allowing you to toggle how aggressively to compress the files. Since this feature does incur minor performance overhead, this does require an administrator to enable the feature on the volumes. Now on to thin provisioning. You can now thin provision volumes on storage spaces direct in Windows Server 2025. This allows you to specify the size of a volume even if it exceeds the size of the available disk capacity in the pool at the time. This allows you to add disks and expand the pool later without having to recreate the volume. You also have the option to set thin provisioning when you're creating a new volume or configure it as the default setting for the pool. If you've already configured your volumes as fixed on storage spaces direct, don't fret. You can easily change the volume to thin provisioned whenever you'd like. Now let's talk about adjustable repair and resync. No matter what happens, at some point you will need to repair or resynchronize data in your storage spaces direct cluster. In the event of a drive replacement, the missing data on the failed disk would need to be repopulated on the new disk. This is an example of a repair. Another such event that might require a resynchronization would be a reboot following updates in your monthly maintenance cycle. In Windows Server 2025, storage repair or resynchronization operations can be tuned to either optimize performance for the VM workloads or repair and resynchronization operations. This capability has been improved in Windows Server 2025 and can be configured using Windows Admin Center or PowerShell, again depending on your preference. There are five unique settings that determine how aggressively to complete the repair and resynchronization operation. By setting this to the lowest setting, the VM workloads will be given priority over the repair and resynchronization. As you select a more aggressive setting, the repair and resynchronization efforts will be prioritized. Overall, you stay in control of your workloads and ensure that you balance your resiliency requirements with the performance of your workloads. I want to thank you for spending some time learning about the new storage improvements in Windows Server 2025. We hope this session was valuable for learning about the storage architectures you can use with Windows Server, our new NVMe and NVMe over fabrics capabilities, deduplication and compression, thin provisioning, and tuning the storage resynchronization operations. Thank you again, and we hope you love Windows Server 2025.