Storage-Compute Q.A.xlsx
XLSX spreadsheet 24 KB Posted
- Attached to
- PAO Storage and Compute Solution
- Solicitation number
- RFQ-25-PHX-059
About this file
This document is a Questions and Answers (Q&A) spreadsheet for a Phoenix Area Indian Health Service (PAIHS) Storage and Compute Solution solicitation. The requirement is a 100% Buy Indian set-aside for an advanced storage and compute infrastructure, seeking a comprehensive solution that supports up to 11 primary sites with scalable, AI-capable computing and storage platforms. Key technical requirements include NVIDIA SuperPOD validation, GPU Direct Storage capabilities, global namespace functionality, and support for complex query engines.
The solicitation (RFQ-25-PHX-059) is issued by the Department of Health and Human Services Indian Health Service, with a standard 5-year period of performance and a solicitation close date of 08/27/2025. Vendors are expected to provide pricing for small, medium, and large solution sets, with a minimum of three compute servers for small sites and four compute servers for medium and large sites. The solution must include 10-year product support and licensing, two-socket servers, InfiniBand connectivity, and optional rack, PDU, and UPS components. Pricing should accommodate potential expansion from 8 to 11 sites, with a focus on supporting future AI and healthcare data analysis workloads.
View the file
Other files
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
Sheet1
PHOENIX AREA INDIAN HEALTH SERVICE (PAIHS)
STORAGE & COMPUTE QUESTIONS AND ANSWERS
| Q&A # | Reference (if provided) | Requirement | Vendor Questions | PAIHS Answer | |
| 1 | PWS Section 3.1, Catalog Functionality | • | The system must have a built-in catalog with a built-in query interface to search the filesystem for specific users, create time, file size, extension, etc. | ||
| • | The system must support data pushdowns from complex query engines such as Trino, Spark, and Vertica | ||||
| • | The built-in catalog shall support the following: ○ _Up to 100 billion rows per table | ||||
| o | Functions - Insert, update, delete, filtered queries, and aggregates | ||||
| o | Sparse Indexes | 1) | What is IHS’ use case for this the catalog functionality requirements? In our experience with similar customers, including VA and DHA – these requirements are not necessary for day-to-day operations and adds additional cost for minimal value. |
2) As written, these requirements can only ‘inherently’ be met within a storage platform by one vendor and there is no brand name justification attached to the solicitation.
| Recommended Change: We respectfully request these requirements be removed to be commensurate with standard enterprise hospital requirements and IHS scope statement “The contractor will be required to provide a standard server solution for …[computing, storage and networking solution]…to each local area network.” | The utilization of advanced analytics, coupled with optimized query analysis within a data federation architecture, is imperative for harnessing historical healthcare data to advance preventative care strategies. This methodology facilitates the efficient management of extensive healthcare datasets, significantly mitigating I/O overhead and empowering governmental entities to discern and address operational shortcomings, thereby streamlining workflows. | |||
| 2 | PWS Section 3.1, Requirement for Fabric Switches | • | The solution must be SuperPod validated by Nvidia (validation report to be attached) | |
| • | The specified solution must meet the Superpod BEST guidance | |||
| • | The specified solution must support Nvidia GPU Direct Storage over multiple paths to multiple nodes. | 1) | What is IHS’ use case for this the catalog functionality requirements? This is an over-architected requirement – only the largest data consumers for AI LLM (e.g., OpenAI, Meta, Amazon, etc.) require this level of capability. |
This requirement adds unreasonable cost for minimal value.
| Recommended Change: We respectfully request these requirements be removed to be commensurate with standard enterprise hospital requirements and IHS scope statement “The contractor will be required to provide a standard server solution for …[computing, storage and networking solution]…to each local area network.” | SuperPOD technology will allow a powerful and scalable AI computing platform essential for complex healthcare workloads. Its robust architecture will support AI model development and training, enabling faster and more accurate analysis of patient data, including symptoms, medical histories, and images, leading to quicker diagnosis and early intervention. This platform can also automate routine tasks for clinicians, optimizing care delivery and freeing them to focus on patient interaction. Moreover, the SuperPOD's flexibility and scalability ensure it can adapt to evolving patient needs and advancements in AI applications within healthcare. | |||
| 3 | PWS Section 3.1, Global Namespace Capabilities | The global namespace must provide a decentralized locking mechanism (IE not over the WAN) for remote site data access and consistency. | Our solution can meet the requirement for a global namespace for remote site data access and consistency in a different technical manner. This is a vendor specific requirement. |
Request IHS remove “must provide a decentralized locking mechanism” to allow additional cost-efficient vendors to respond.
| Recommended change: “The solution must support a global namespace for remote site data access and consistency.” | Implementing decentralized locking mechanisms is required to prevent a single point of failure associated with a centralized authority. | |||
| 4 | PWS Section 3.1, SMB Support Requirements | Server failover using SMB 3 resilient handles | Our solution can meet the requirement for server failover in a different technical manner. |
Request IHS revise this requirement to “Support Server failover” to allow additional cost-efficient vendors to respond.
| Recommended change: “Solution must support server failover.” | SMB 3 Resilient handles are a mandatory requirement to guarantee robust reliability and continuous availability for file shares. | ||
| 5 | Are there environmental constraints for the locations (e.g., cooling, rack space, power density)? | Yes and we will work with contractor to meet those challenges in our facilities, we are budgeting for new server racks, PDU's and UPS if needed. | |
| 6 | What is the purpose of the Nexus switches called out in the RFP? | The switch will facilitate the integration of the platform into the existing network architecture to mitigate bandwidth bottlenecks. | |
| 7 | Are compute nodes with 2-sockets with 16 cores each acceptable or is the intent to have less than 16 cores per node total? | 2-Sockets are acceptable. | |
| 8 | Is native InfiniBand required for all components, or only for GPU Direct Storage (GDS) paths? | Infiniband connectivity is mandatory for all direct paths between storage and analytic nodes. | |
| 9 | Will the fabric be Ethernet, InfiniBand, or both for GPU Direct Storage traffic? | The fabric will require both for direct storage traffic. | |
| 10 | Should the compute systems will be GPU-enabled? If so, is there any estimate on the type or GPU compute capacity per node. | Computing systems with NPUs and GPUs are recommended to address future AI workload demands. Server requirements in the PWS should be considered as a minimum requirement. | |
| 11 | Can you provide detailed information about the expected workloads for these solutions, including workload types (e.g., AI/ML, VDI, databases, file services), performance requirements (e.g., IOPS, latency, throughput), and any anticipated growth or scaling considerations? | ||
| "Is AMD acceptable for processors as long as they are comparable or exceed the Intel Xeon-G 6326 or Better" | Infrastructure development prioritizes meeting the immediate needs of hospital clinician services while ensuring future compatibility with AI workloads. The inclusion of NPUs in hardware proposals is encouraged for optimal performance. | ||
| 12 | Is InfiniBand or RoCEv2 a strict requirement, or is NVMe over TCP an acceptable alternative? | Infinband required to reduce CPU overhead. | |
| 13 | Are there any defined backup and restore requirements between sites, such as disaster recovery or replication objectives (e.g., RTO/RPO)? | Vendors are encouraged to propose innovative software solutions that can enhance our existing backup and disaster recovery capabilities. Our current system utilizes Rubrik technology for both site-specific backups and disaster recovery hosted in our Albuquerque data center. Proposed solutions must demonstrate the ability to integrate seamlessly with this architecture. | |
| 14 | Have the primary and secondary sites been formally identified? | This project is for 8 primary sites, and allows for the flexibility of adding up to 3 additional sites, reaching a total maximum of 11 primary sites. No secondary sites are required. | |
| 15 | Can you provide the WAN bandwidth and latency (e.g., RTT) between the sites, if available? | Bandwidth between sites varies and circuits can be anywhere from 100Mbps to 1Gbps. | |
| 16 | Is there a minimum required storage capacity per compute node or across the hyperconverged cluster, or is it acceptable for all storage to reside on an external storage array? | ||
| Can you clarify whether IHS is seeking an NVIDIA SuperPOD, BasePOD, or a more general AI/ML-capable compute and storage solution? | Minimum storage for compute node is on PWS. Minimum 400 GB redundant Flash storage for compute operating system per node. IHS is seeking Nvidia SuperPOD. | ||
| 17 | BasePOD is a reference design for small to mid-scale AI workloads. It includes validated GPU servers, networking (typically RoCE or InfiniBand), and AI-optimized storage—ideal for pilot projects or moderate AI deployments as opposed to general compute / storage workloads. |
SuperPOD is a full-scale, production-ready architecture designed for large, enterprise AI workloads. It includes large GPU clusters, high-speed networking, and integrated storage, all validated under NVIDIA’s highest standards. It is all Nvidia Hardward and must be 32+ nodes.
| The choice depends on IHS’s target use cases, performance needs, and expected scale. Let us know your goals so we can recommend the best-fit architecture. | Phoenix Area IHS requires Nvidia SuperPOD fabric switches for the purpose of optimizing storage node I/O. | ||
| 18 | Will IHS allow a site visit to all locations at the Call level prior to vendors providing IHS with a firm fixed price? | Ample power/space/cooling is maintained at all locations, and collaboration will be made with the facilities management team at each site to address potential deficiencies. | |
| 19 | The SOW requires the solution to be NVIDIA SuperPOD validated and to support GPU Direct Storage over multiple paths to multiple nodes. SuperPOD is a complete, integrated reference architecture built specifically for large-scale AI workloads and includes DGX GPU systems, NVLink, high-speed networking, and NVIDIA AI software. In contrast, NVIDIA BasePOD is a flexible, validated reference architecture that enables organizations to build scalable AI infrastructure using a broader set of vendor components. HPE’s BasePOD-certified solution includes GreenLake for File (GL4F) for storage, which supports GPU Direct Storage, and can be paired with HPE ProLiant GPU servers as part of solutions like Private Cloud for AI (PCAI). Given that the IHS use case appears to focus on scalable and modular storage with standard servers—not large-scale AI compute—would the government consider acceptance of a solution leveraging BasePOD-certified components (specifically GL4F with GPU Direct Storage support) in lieu of full SuperPOD validation? | SuperPOD technology will allow a powerful and scalable AI computing platform essential for complex healthcare workloads. Its robust architecture will support AI model development and training, enabling faster and more accurate analysis of patient data, including symptoms, medical histories, and images, leading to quicker diagnosis and early intervention. This platform can also automate routine tasks for clinicians, optimizing care delivery and freeing them to focus on patient interaction. Moreover, the SuperPOD's flexibility and scalability ensure it can adapt to evolving patient needs and advancements in AI applications within healthcare.We are open to any proposal that can meet our technical needs. | |
| 21 | What is the required support/licensing term for the storage? | 10 years | |
| 22 | What is the period of performance for the requested BPA? | Standard 5 Year POP for BPA. | |
| 23 | Can the govt please explain how they would like to see pricing as the min. required solution set is eight (8), but the location sheet lists eleven (11) sites. | ||
| • | Does the govt want to see a QTY=1 for Small/Medium/Large solution sets + Optional Rack/PDU/UPS Pricing |
Or/And
• Does the govt want to see a QTY=1, per site solution set + Optional Rack/PDU/UPS Pricing This project is for 8 primary sites, and allows for the flexibility of adding up to 3 additional sites, reaching a total maximum of 11 primary sites.
Pricing should reflect a solution for a small site, a solution for a medium site, and a solution for large site, with options for Rack/PDU/UPS with each, all meeting minimum specifications.
| 24 | The govt outlines a chart of deliverables under section 5 of the posting. | |
| Is that just an example as the totals listed don’t align to the requested number of servers for Large/Medium sites requiring a min. of 4 nodes each (Yes/No)? | The proposed solutions should demonstrate a scalable architecture capable of supporting operations for the initial five years. Specifically, small sites mandate a deployment of three compute servers for every storage solution implemented. For large and medium sites, the requirement increases to four compute servers per storage solution. |
See Attachment 2. Locations. Storage.pdf
| 25 | If the optional Rack/PDU/UPS Solution is the same for all Solution Sets (Small/Medium/Large), does the govt want to see pricing per solution size and/or per site? | |||
| • | No, QTY=1 is fine | |||
| • | Yes, Per Solution Size (Small/Medium/Large) | |||
| • | Yes, Per Site (See Attachment 2. Locations. Storage.pdf) | |||
| • | Yes, Per Solution Size and per Site? | Yes, per site (small/medium/ large). 2 UPS/PDU sets per rack. | ||
| 26 | The Govt lists a sever count per solution set and max core count per processor. | |||
| • | How many processors does the govt require per server (QTY=1 or QTY=2)? | Two (2) | ||
| 27 | Does the govt want to see optional Layer 3 and Security Licensing for the TOR Switching requested (Yes/No)? | Yes | ||
| 28 | Given the govt has asked for the rack solution to be optional, should the offeror provide optional cabling lengths to accommodate differences in deployment scenarios per site (Yes/No)? | Yes | ||
| 29 | The Govt lists an upfront support/maintenance duration for TOR switching. | |||
| • | What duration of support is required for the servers requested (3 Years or 5 Years)? | |||
| o | Does the Govt require disk retention for the servers (Yes/No)? | |||
| • | What duration of support is required for the optional rack/pdu/ups solution (3 Years or 5 Years)? | |||
| • | What duration of licensing/support is required for the storage solution requested (3 Years or 5 Years)? | 5 years support on all. |
Yes, for disk retention.
30 The solution requested by the govt will require a management switch to maintain/operate/upgrade the solution.
Does the govt want to see pricing for a management switch?
| • | No, Govt will use an existing switching platform | |||
| • | Yes, include a management switch with the solution, 5 Year Support Contract, and installation/cabling/configuration services per solution set | |||
| • | Yes, include an optional management switch with the solution, 5 Year Support Contract, and installation/cabling/configuration services per solution set | No, Govt will use an existing switching platform. | ||
| 31 | Does the Govt want to see optional transceiver and fiber options for connecting the requested TOR Switches to the existing LAN (Yes/No)? | Yes | ||
| 32 | The Govt lists a SuperPod validation requirement for the Storage Fabric Switches. | |||
| Is BasePod certified acceptable since SuperPod certification is not needed or applicable to this solution set (Yes/No) | No, SuperPod is required. | |||
| 33 | The Govt lists a software licensing and support requirement that varies from traditional industry standards of EoS + 5 Years. | |||
| To guarantee multiple solution proposals are received, will the Govt accept a licensing and support requirement of EoS + 5 Years? | We are looking for vendors that can certify their product for 10 years. | |||
| 34 | Would you be open to considering a “brand name or equal” approach for this requirement? We believe this flexibility could enable us to propose a more cost-effective and competitive solution—either by evaluating alternatives to the current specified components or by maintaining the core (e.g., VAST) while optimizing other elements of the configuration. | Vast or similar technology. We will consider all proposals that can meet our minimum technical needs for storage and compute. |
File details come from the government source that posted it. Updated .