The correct answer is B. The HPE ProLiant Compute DL384 Gen12 is an AI-optimized server platform built around NVIDIA Grace Hopper Superchip architecture. In this class of design, the CPU/GPU complex and high-bandwidth memory are integrated as part of the Superchip module rather than being expanded like conventional DIMM-based system memory. The HPE DL384 Gen12 data sheet describes the platform as using NVIDIA GH200 NVL2 Superchips and highlights coherent memory and memory bandwidth characteristics, which is exactly why the memory configuration is treated as tied to the Superchip selection. Option A is incorrect because the DL384 Gen12 architecture is not upgraded by freely adding more GB200 accelerators; it is based on a defined Grace Hopper Superchip platform. Option C is incorrect because EDSFF drive support is a platform/backplane architecture consideration, not simply a RAID-controller dependency. Option D is also not the governing rule for 400Gb NIC support. Reference/topics: AI server architecture, DL384 Gen12, NVIDIA GH200/NVL2, Superchip memory architecture, hardware validation.

The correct answer is D because the requirement is not simply additional raw capacity; it is a write-performance target of 500K IOP/s. For HPE Alletra Storage MP B10000 designs, controller-node model selection, node count, shelf count, drive population, and media layout collectively determine whether the configuration can satisfy the performance envelope. Merely increasing the number of drives, as suggested in option C, does not guarantee that the front-end controller resources, CPU, cache, and back-end connectivity can sustain the required write workload. Options A and B use lower or inappropriate node selections for this stated requirement. The B10230 model, combined with additional shelf and drive resources, is the configuration change that aligns the controller and media resources with the required write I/O target. HPE’s course outline places HPE Alletra Storage MP B10000 in both general enterprise solution design and database workload solution design, which is exactly the kind of workload-sensitive sizing exercise represented here. The proposal must be validated against the customer’s required I/O profile, not only against usable capacity.

The correct answer is D. HPE Alletra Storage MP B10000 supports mission-critical Peer Persistence designs using synchronous Remote Copy, and FC can be used as an underlying Remote Copy transport. HPE material for Active Peer Persistence states that the solution uses Remote Copy synchronous mode, while HPE B10000 training material describes the requirement for synchronous RC links using RCFC or RCIP and cross-site Fibre Channel or iSCSI host SAN connectivity. Options A and B are wrong because Alletra 6000 is associated with Nimble/Alletra 6000 Peer Persistence, not the same Active Peer Persistence model used with Alletra Storage MP B10000, Primera, and Alletra 9000-class arrays. Option C is not the best answer because Active Peer Persistence is a two-array high-availability model; one-to-many Remote Copy is a replication topology and should not be confused with the active-active peer-persistence protection model. The key validated design differentiator here is that B10000 supports FC-based synchronous replication for high-availability block-storage use cases. Reference/topics: HPE Alletra Storage MP B10000, Active Peer Persistence, Remote Copy, mission-critical storage design.

The correct answer is C. RPO 0 is the decisive requirement. A zero recovery point objective means the customer cannot lose acknowledged writes if a site fails. That requirement cannot be satisfied by asynchronous replication because asynchronous copy always introduces replication lag. The storage volumes backing the VMware datastores must be synchronously replicated between the two sites so that committed data exists at both locations before the write is considered protected. HPE Alletra Storage MP B10000 is designed for mission-critical block workloads and supports high-availability replication models, including synchronous Remote Copy / Peer Persistence designs for VMware environments. HPE documentation for Active Peer Persistence describes synchronous Remote Copy as the mechanism used for high-availability protection across arrays. Option A is incorrect because stretching or joining HPE Synergy frames across sites is not the mechanism that provides RPO 0. Option B can help with VMware policy-driven placement, but SPBM alone does not replicate data. Option D fails the RPO requirement. Reference/topics: DR architecture, RPO 0, synchronous replication, B10000 Remote Copy, VMware datastore protection.

The correct answer is B. RoCEv2 is the appropriate choice when RDMA traffic must traverse routed Layer 3 networks, multiple VLANs, and multiple switches. RoCEv1 is Layer 2 only and is therefore unsuitable when the design requires traffic to move between different routed segments or logical enclosures through the data-center network. A lossless RoCEv2 design must be planned end to end: priority flow control, ECN, QoS classification, MTU consistency, and routing paths must be aligned across the server adapters, HPE Synergy Virtual Connect domain, and Aruba CX fabric. Option A is incorrect because RoCEv1 cannot satisfy the stated Layer 3 traversal requirement. Option C is incorrect because Aruba CX 8325 is a data-center switch family capable of participating in advanced data-center designs; replacing it with CX 8400 is not inherently required just to support DCBX-style fabric behavior. Option D creates artificial separation and does not solve the multi-switch lossless-fabric requirement. Reference/topics: RoCEv2, lossless Ethernet, Aruba CX data-center switching, HPE Synergy Virtual Connect networking, workload-driven network design.