ACT HPC Cluster BNJ_Redacted.pdf

J&A Number: N00167-26-

SIMACQ-CD-66-0021

Code: 664 P.R. Number: N/A

JUSTIFICATION AND APPROVAL

FOR USE OF BRAND NAME ONLY

JUSTIFICATION

1. Contracting Activity

The Naval Sea Systems Command (NAVSEA), Naval Surface Warfare Center, Carderock Division (NSWCCD), Contracting & Acquisitions Department (Code 02).

2. Description of the Action Being Approved

NSWCCD intends to procure hardware, equipment, and associated services support produced by the following Original Equipment Manufacturer (OEM) on a brand name only basis:

Advanced Clustering Technologies, Inc.

3148 Roanoke Road Kansas City, MO 64111

The requisite brand name only hardware, equipment, and associated support services identified in Section 3 of this Justification are necessary for NSWCCD to expand its existing high-performance computing (HPC) clusters infrastructure, which are used to perform analysis of the performance of various ship and submarine platforms.

This Justification is in accordance with Revolutionary FAR Overhaul (RFO) 12.102(b).

The Government anticipates establishing a Blanket Purchase Agreement (BPA), which will allow for the placement of firm-fixed price calls.

The period of performance/ordering period of the anticipated BPA is expected to be sixty (60) months.

The Government will use the combined synopsis and solicitation procedure set forth in RFO 12.202(b), and the notice will be posted

DEPARTMENT OF THE NAVY

NAVAL SURFACE WARFARE CENTER

CARDEROCK DIVISION

9500 MACARTHUR BOULEVARD

WEST BETHESDA MD 20817-5700

J&A Number: N00167-26-SIMACQ-CD-66-0021 to the SAM.gov. The total estimated value of this proposed acquisition is approximately

3. Description of Supplies/Services

The proposed acquisition is for the purchase of the quantities of the supplies/services described below:

Cabinet Enclosure:

The HPC cabinet enclosure(s) shall contain all hardware for the HPC system:

Be compliant with EIA-310 or Open Rack Standards.

Have a minimum 40 Open Rack Unit (OU) height.

Have adequate capacity for:

o Power Distribution Units (PDUs) o Cluster Management Server o Login Node Server o Compute Nodes o 4-Port Keyboard Video Mouse (KVM) o Network Attached Scratch Storage Server o Network Attached Cold Storage Server o Administrative Network o InfiniBand Fabric (NDR) Network Blanking panels to cover unused spaces

Advanced Cluster Technologies (ACT) Cluster Management Appliance:

There shall be a minimum of one cluster management Appliance. The cluster management server shall have all of the following attributes:

EIA-310 Standard or Open Rack Standard Integrated solution 2x 32-core Genoa 9334 2.7GHz or similar processors approved by NSWCCD.

o Same processor type and same processor speed as the compute node servers o Configured to have a high ratio of peak CPU memory bandwidth to peak CPU floating point capability.

o Hyper Threading disabled.

Minimum of 384 GB RAM memory

Separate boot mirrored solid state boot drives.

Minimum 2 TB storage space

Dual network adapters; Must include both InfiniBand (NDR) at 200 GB or better and Ethernet 10 GB connections.

Redundant power supply, 208V Operating System requirements:

o Preload, RHEL cluster version 9 or higher, approved by

NSWCCD.

o OS will need to connect to currently existing Windows Active Directory

Remote iKVM with in-band Management Cluster Management Software Suite o Web or Application based GUI for infrastructure management for all aspects of the server Slurm Queuing Controller Must be compatible with the rest of the cluster management hardware and software requirements

Login Node Server:

There shall be a minimum of one login node server. The login server shall have all the following attributes:

EIA-310 Standard or Open Rack Standard Integrated solution 2x 32-core Genoa 9334 2.7GHz or similar processors approved by NSWCCD.

o Same processor type and same processor speed as the compute node servers o Configured to have a high ratio of peak CPU memory bandwidth to peak CPU floating point capability.

O Hyper Threading disabled.

Minimum of 384 GB RAM memory Separate boot mirrored solid state boot drives.

Minimum 2 TB storage space Hardware redundant array of independent disks (RAID) controller Dual network adapters; Must include both Infiniband (NDR) at 200 GB or better and Ethernet 10 GB connections.

Redundant power supply, 208V Operating System requirements:

o Preload, RHEL cluster version 9 or higher, approved by

NSWCCD.

o OS will need to connect to currently existing Windows Active Directory

Remote iKVM with in-band Management Web or Application based GUI for infrastructure management.

Slurm scheduling Must be compatible with the rest of the cluster management hardware and software requirements.

ACT Compute Nodes:

The compute node server shall have all the following attributes:

EIA-310 Standard or Open Rack Standard Integrated solution 2x 32-core Genoa 9334 2.7GHz or similar processors approved by NSWCCD.

o Configured to have a high ratio of peak CPU memory bandwidth to peak CPU floating point capability.

o Hyper Threading disabled.

Minimum of 384 GB RAM memory Dual network adapters; Must include both Infiniband (NDR) at 200 GB or better and Ethernet 10 GB connections.

Redundant power supply, 208V Operating System requirements:

o Preload, RHEL cluster version 9 or higher, approved by

NSWCCD.

o OS will need to connect to currently existing Windows Active Directory

Remote iKVM with in-band Management Web or Application based GUI for infrastructure management.

Slurm scheduling

Must be compatible with the rest of the cluster management

Scratch Storage Server:

The scratch storage array will need to have the following capabilities listed below and be available for the cluster nodes and external user network:

EIA-310 Standard or Open Rack Standard Integrated solution 16-core Genoa 9124 3.0 GHz or similar processors approved by NSWCCD 192 GB of RAM memory 2x 1TB NVMe SSD for OS in RAID 1 configuration 10x 3.2TB NVMe for scratch in RAID 0 configuration (32TB

Total) Hardware RAID controller Redundant power supply Dual network adapters; Must include both Infiniband (NDR) at 200 GB or better and Ethernet 10 GB connections Operating System o Preload, RHEL cluster version 9 or higher, approved by

NSWCCD

o OS will need to connect to currently existing Windows Active Directory

Remote iKVM with in-band Management

Web or Application based GUI for infrastructure management hardware and software requirements.

Cold Storage Server:

The attached storage array will need to have the following capabilities listed below and be available for the cluster nodes and external user network:

EIA-310 Standard or Open Rack Standard Integrated solution 16-core Genoa 9124 3.0 GHz or similar processors approved by NSWCCD 192 GB of RAM memory 2x 1TB NVMe SSD for OS in RAID 1 configuration

3.5 inch SAS/SATA bays o 12x 18TB SAS HDDs 12GB/s 7200RPM w/ 512MB cache installed with RAID 6 configuration to allow for 160TB of usable storage.

Able to mount /home to the login node and the compute nodes.

Contain two distinct file systems (way in which files are named and where they are placed logically for storage and retrieval).

o /work — a single file system with 90% of the total formatted usable disk storage and expandable.

o /home — a single file system (user accounts and system files) with at least 10% of the total formatted usable disk storage.

At least 6GB/s of read, write, and full duplex bandwidth (respectively) between the compute nodes and /home.

At least 6GB/s of read, write, and full duplex bandwidth

(respectively) between the compute nodes and /work.

Hardware RAID controller Redundant power supply Dual network adapters; Must include both Infiniband (NDR) at 200 GB or better and Ethernet 10 GB connections.

Operating System o Preload, RHEL cluster version 9 or higher, approved by

NSWCCD

o OS will need to connect to currently existing Windows Active Directory.

Remote iKVM with in-band Management Web or Application based GUI for infrastructure management Must be accessible if the compute nodes are offline.

Administrative Network:

The Administrative Network infrastructure will be used to manage and monitor all the aspects of the cluster environment locally and remotely through a 10 GB network. The Administrative Network shall contain the following:

EIA-310 Standard or Open Rack Standard Integrated solution 10GB switches and supporting infrastructure The network will have the ability to:

o manage file system traffic, cluster utility communication, system administration, and maintenance o remotely monitor temperatures, voltages, fans, and power supplies o allow for remote power cycling of each node

Tightly integrated Cluster Expansion and Infrastructure Interconnect reliability, availability, and serviceability (RAS).

InfiniBand Fabric Network:

The InfiniBand fabric will be set up to have the best reliability, availability within the cluster environment. The InfiniBand fabric shall have the following minimum requirements:

NDR or similar switch approved by NSWCCD and supporting infrastructure Need to have peak bandwidth between nodes at least 100GB/s Enough ports to accommodate expansion in the near future for another 25% t0 50% additional nodes as needed

ACT Cluster Management Software Suite The cluster management software shall provide management for the hardware, the operating system, HPC software, and users in a unified framework of tools to manage the cluster. The cluster management software shall also have the following minimum requirements:

Provide a batch scheduler/workload management system (WMS) that shall work across all nodes within the cluster.

Pre-bundled OS and network optimizations for OS settings, drivers, and InfiniBand (NDR), environments that are necessary for proper performance and functionality inside an HPC environment.

Allow compute nodes (as a collective whole) to be powered down without impacting the power to other system components.

Support diskless node deployment

Web or Application based GUI of the Cluster Management Software.

Ability for web interface to be accessed by standard user to monitor all resources associated with running jobs with no administrative capability.

Integrate seamlessly into an active directory environment all the way down to the nodes Web or Application based GUI for the complete deployment and management of the cluster environment.

Installation:

Complete system installation with interface to the existing power supply within Multiple classification facility.

Ensure Clustervisor features are tested against the exact hardware being deployed.

Preform cluster-wide benchmarking before delivery.

Maintenance:

Maintenance and Warranty on all furnished equipment and software for a period of no less than five years from the time of installation.

On-site training The contractor shall provide training as needed.

This agreement will be awarded in FY26 and have five ordering years from the time of the award.

The Government’s minimum needs have been verified by the certifying technical and requirements personnel.

4. Statutory Authority Permitting Other Than Full and Open Competition

10 U.S.C. §3204(a)(1). “[T]he property or services needed by the agency are available from only one responsible source or only from a limited number of responsible sources and no other type of property or services will satisfy the needs of the agency.” (RFO Overhaul 12.102(b)).

5. Rationale Justifying Use of Cited Statutory Authority

NSWCCD Code 664 requires the procurement of ACT branded hardware, equipment, and associated support services to expand its existing HPC clusters infrastructure. The current modeling software necessitates a highly consistent, homogenous operating environment to function reliably. Because the existing infrastructure uses ACT branded HPC clusters, only ACT branded components possess the exact form, fit, and function required to seamlessly integrate into the current system architecture. For instance, the ACT branded HPC clusters generate images that are essentially used for modeling, and each image contains the operating system and drivers that are specifically built for all hardware components that make up the nodes. Using a different brand of node would require the construction of a new image in order for all of the nodes within the cluster to operate properly, and the use of a heterogenous operating environment using different brands of nodes would adversely impact computational time and the ultimate results of the end users, meaning the ability to create jobs on the login nodes that rely on the images specific to the nodes.

Furthermore, the proprietary ACT Graphical User Interface (GUI) is integral to current operations, providing engineers with real-time job management capabilities that are critical for executing time-sensitive, mission-essential analysis. Moreover, NSWCCD uses more than one software that was developed by the Navy and specifically engineered and optimized for use on the existing ACT branded HPC cluster infrastructure. If different brands of nodes are used in the existing ACT branded HPC cluster infrastructure, NSWCCD would need to develop new software for those different brands of nodes to operate properly, and ultimately will impact NSWCCD’s modeling actions, such as resulting in unacceptable delay in fulfilling mission-critical requirements, as well as a substantial duplication of cost to the Government that is not expected to be recovered through competition.

6. Description of Efforts Made to Solicit Offers from as Many Offerors as Practicable

Through market research, NSWCCD Code 664 identified multiple brands of HPC clusters-related hardware, equipment, and associated support services that were comparable to the ACT branded items identified herein; however, the ACT branded items are the only brand of hardware, equipment, and associated support services that meet the technical requirements for interacting with the existing HPC cluster infrastructure. NSWCCD previously procured ACT branded HPC clusters (DO No. N00167-25-F-0098), which are currently utilized in the existing computational cluster infrastructure. This prior acquisition was competitive and resulted in the purchase of the existing ACT branded HPC clusters.

Pursuant to RFO 12.202(b), NSWCCD intends to use the combined synopsis and solicitation procedure, and the notice will be posted to the SAM.gov. Additionally, in accordance with RFO 6.301(c), the Government will make the Justification publicly available at the

Government Point of Entry (GPE)will make this justification available on the Government Point of entry (SAM.gov), and (3) for a minimum of 30 days. The Government will ensure that the contract file documents any responses received from by making the justification available at the GPE.

7. Determination of Fair and Reasonable Costs

In accordance with RFO 12.204(a), the Procuring Contracting Officer (PCO) must determine the price to be fair and reasonable. The resultant solicitation contemplated with this Justification will require interested offerors to submit a formal priced offer with sufficient information to support the accuracy and reliability of the estimate. The PCO shall determine whether the proposed price of each offer represents the best value (as defined in RFO 2.101) to meet the Government's needs. The PCO will also rely on Code 631 experienced technical experts, and, use past procurements from the Navy and other Department of War (DoW) agencies that have acquired ACT branded HPC clusters-related hardware, equipment, and associated support services (or comparable products/services), when conducting the price analysis.

This information will form the basis for a fair and reasonable price determination.

8. Actions to Remove Barriers to Competition

The Government will continue to canvass and maintain open communication with ACT and other OEMs of comparable HPC clusters-related hardware, equipment, and associated support services (via symposiums, conferences, publications, product advertisements, vendor demos, etc.) to seek out any alternative parts suppliers, if they exist.

CERTIFICATIONS AND APPROVAL

TECHNICAL/REQUIREMENTS CERTIFICATION

I certify that the facts and representations under my cognizance, which are included in this justification and its supporting data, except as noted herein, are complete and accurate to the best of my knowledge and belief.

LEGAL SUFFICIENCY REVIEW (NMCARS 5206.104-90)

CONTRACTING OFFICER CERTIFICATION (RFO 6.104-1(a)(12)

I certify that this justification is accurate and complete to the best of my knowledge and belief.

APPROVAL BLOCK (FAR 6.104-2(a)for a proposed contract exceeding $900,000 - $20,000,000, the Advocate for Competition approval)

Upon the basis of the above justification, I hereby approve, as the solicitation of the proposed procurement(s) described herein using other than full and open competition, pursuant to the authority of 10 U.S.C. 3204(a)(1).