Attachment_L-2_Price_Scenarios.pdf

Attachment L-2 1 Joint Enterprise Defense Infrastructure (JEDI) Cloud 2

Price Scenarios 3 Updated 26 July 2018 4

1.0 For all scenarios: 6

a. Offerors should assume that the required solution is for an unclassified JEDI Cloud 8 requirement in accordance with the SOO and JEDI Cloud Cyber Security Plan, unless the 9 scenario explicitly states otherwise. For CONUS scenarios that do not specify an exact 10 location, the Offeror shall price the most expensive CONUS “region” or “availability 11 zone” proposed for JEDI Cloud. 12

b. Offerors should assume that all services and resources are utilized continuously, and that 13 all storage and data is retained for the duration of the order, unless the scenario explicitly 14 states otherwise. 15

c. Offerors should assume that the migration of any applications described into JEDI Cloud 16 is an instantaneous operation that takes place on day 1 of the order unless explicitly stated 17 otherwise; this migration of any applications is not to be priced. 18

d. Holidays, fiscal year end, and calendar year end have no effect on traffic patterns or 19 scenario requirements. Leap years are not to be observed. 20

e. Offerors must propose a solution that is consistent with the solicitation, including all 21 terms and conditions, requirements, and attachments, unless otherwise specified in the 22 scenario. 23

f. No classified information is required, nor should it be provided, in any response. 24

g. If Cloud Support Package offerings are required by the scenario, the Offeror shall clearly 25 identify the applicable tier of support and relevant category of services, including any 26 service constraints. If meeting the specified requirement in the price scenario is 27 dependent on ordering other Cloud Support Package offerings, then the Offeror shall also 28 identify and price all dependent Cloud Support Package offerings. 29

h. The performance characteristics of virtual CPU (vCPU) and GPU cores in JEDI Cloud, 30 including on tactical edge devices, must match the vendor’s commercial cloud offering. 31

i. Assume all applications in all scenarios are fully deployed and running in a production 32 environment. 33

j. Assume for all Ruggedized devices that the system does not need to be tested and 34 certified as meeting the standard. 35

Price Scenario 1 Sample Unclassified Application Stack 37 A scalable application has the following traffic patterns in production: 39

● Averages 4,000 requests/min during normal business hours (Monday to Friday, 0900 – 40

1700 ET) 41

● Monday from 0900ET to 1200ET traffic jumps to 400,000 requests/min 42

● Tuesday through Friday from 0900ET to 1000ET traffic jumps to 20,000 requests/min 43

● Averages 500 requests/min outside the “normal business hours” specified above 44

A globally distributed JEDI Cloud CDN routes all incoming web requests to the application 46 stack’s load balancer. Static file requests are routed directly to the static file store by the CDN 47 and account for 5% of all requests, averaging 40 KB in size. The CDN then caches the static file 48 requested based on the expiration header information configured in the static file store (assume 49 30 days). All web requests are routed to a JEDI Cloud load balancer and average 10 KB in size. 50 Load is then distributed evenly across healthy application servers with 10% of all requests 51 resulting in an insert or update to the database. The system utilizes DNS Zone Hosting services 52 offered by the JEDI Cloud vendor and averages 1000 DNS requests/hour. 53 The main application resides on multiple nodes that are evenly distributed among the available 55 zones in the region, with a minimum of two nodes at all times. Each application node is hosted 56 on a moderate performance compute instance (8 vCPUs and 32 GB of RAM) and can handle up 57 to 500 requests per minute. During traffic spikes the number of required application instances is 58 expected to grow dynamically using the auto-scaling capabilities available from the JEDI Cloud 59 provider. Once a traffic spike has ended, the number of application nodes is expected to scale 60 back down to appropriate baseline levels. All requests are authenticated using transport layer 61 security (TLS). 62 Application activity will be pushed through a JEDI Cloud event stream (also known as a 64 message queue) with functionality similar to Apache Kafka. Assume 90% of application requests 65 result in events pushed to the queue, an average of 2 events per such requests, and each event is 66 200 KB in size. Eventually each event is read off the stream and results in an update or write to 67 the database cluster. Users may also upload files to the application, where they will be stored in 68 the static file store and retrieved dynamically according to an application specific access control 69 mechanism. Additionally, the application will store short-term session information in a JEDI 70 Cloud caching service. This session storage will start off at 1 GB in size on day 1 (do not price 71 any costs associated with the migration of all of this data to JEDI Cloud), with 100 MB added 72 each day and another 100 MB that expires each day. 73 The database cluster is hosted on six JEDI Cloud IaaS compute instances. The database contains 75 750 GB of data on day 1 (do not price any costs associated with the migration of all of this data 76 to JEDI Cloud). The nodes are evenly distributed among the available zones in the region, and 77 each node requires a compute instance with 24 vCPUs, 256 GB of RAM, and 500 GB of normal 78 SSD storage. Every minute, 500 MB of the 750 GB are updated. The data in the relational 79 database will grow in size by 0.2% per month. Each day the reads from the application and 80 analytics nodes total 1.5 TB in size. Automated snapshots are taken of the relational database 81 daily using a JEDI Cloud service and rotated every 7 days. Weekly full database backups are 82 created on Saturdays (starting the week the order is placed) and stored online for 4 weeks before 83 being rotated to offline storage. 84 Four nightly analysis jobs are run on eight high performance RAM optimized compute instances 86 requiring 32 vCPUs and 400GB of RAM each. Each instance should be powered down except 87 for the two hours they run each night. These jobs will each read 700 GB of data from the static 88 file store and 300 GB of data from the database cluster with no data preparation. Each job 89 produces 5 GB of new results which are stored in a separate, redundant, multi-zone JEDI Cloud 90 relational database service which requires 8 vCPU, 32 GB of RAM, and 300 GB of storage. All 91 previous analysis job results are retained. This database service is separate from the database 92 cluster mentioned in the previous paragraphs. 93 The static file store resides in a single region, will contain 1 TB of static files on day 1, and will 95 grow as necessary (do not price any costs associated with the migration of all of this data to JEDI 96 Cloud). Every minute, 700 MB of the 1 TB are updated with new content (each update request is 97 500 KB). The static file store content will grow 0.2% per month. Each day the reads by the 98 application and analytic nodes total 2 TB in size (each read request averages 500 KB). 99 Of the files uploaded to the static file store, 33% require PKI encryption. Assume this application 101 has one master key, which is managed by the JEDI Cloud provider. The application will request 102 separate data keys based on that master for segments of its user population. Assume 5 new keys 103 are created every day. The application will use these keys to encrypt 1000 files per day and 104 decrypt 30,000 per day. 105 Every action that changes infrastructure configuration or alters infrastructure state is logged 107 through a JEDI Cloud logging service (assume there are 20 such actions a day). Additionally, 108 every data action taken for the managed data services is also logged through the same JEDI 109 Cloud service. The infrastructure components running this application are continuously 110 monitored through a JEDI Cloud service offering for performance degradation, which sends out 111 an email alert if defined thresholds are exceeded for any infrastructure performance metrics. 112 Additionally, there are 50 separate custom application metrics being monitored by this JEDI 113 Cloud service, 25 of which will send out an email alert if defined thresholds are exceeded. In 114 total, the JEDI Cloud service monitoring for this application sends 3 email notifications per day. 115

a. Please price this scenario: An order for a single application is placed on January 6, 117 2020, that meets the above technical requirements operating for 365 days continuously. 118

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 119 technical requirements for a single application. 120

ii. Please separately price services under the Cloud Support Package line item to 121 support the application, including at a minimum: 24x7 support by phone, web, 122 and email; less than 1 hour response time for critical issues; access to detailed 123 online training materials; and guidance on application and infrastructure 124 architecture by phone using the lowest applicable tier. 125

b. Please price this scenario: The order is placed on June 7, 2021, for 50 applications total 127 across 25 separate accounts, meeting the above technical requirements operating for 365 days 128 continuously. 129

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 130 technical requirements for 50 applications across 25 accounts. 131

ii. Please separately price services under the Cloud Support Package line item to 132 support the application, including at a minimum: 24x7 support by phone, web, 133 and email; less than 1 hour response time for critical issues; less than 4 hours 134 response time for moderate issues; online instructor-led and on-demand online 135 training for a total of 20 hours for 30 people; and remote in-depth architectural 136 support for refactoring applications and configuring cloud infrastructure totaling 137 40 hours. 138

c. Please price this scenario: The order is placed on June 3, 2024, for 1,000 applications 140 total across 750 separate accounts meeting the above technical requirements operating for 365 141 days continuously. 142

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 143 technical requirements for 1000 applications across 750 accounts. 144

ii. Please separately price services under the Cloud Support Package line item to 145 support the application, including at a minimum: 24x7 support by phone, web, 146 and email; less than 1 hour response time for critical issues; less than 4 hours 147 response time for moderate issues; online instructor-led and on-demand online 148 training for a total of 60 hours for 200 people; and remote in-depth architectural 149 support for refactoring applications and configuring cloud infrastructure totaling 150 80 hours. 151

d. Please price this scenario: The order is placed on June 7 2027, for 1,000 applications 152 total across 750 separate accounts meeting the above technical requirements operating for 365 153 days continuously. 154

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 155 technical requirements for 1000 applications across 750 accounts. 156

ii. Please separately price services under the Cloud Support Package line item to 157 support the application, including at a minimum: 24x7 support by phone, web, 158 and email; less than 1 hour response time for critical issues; less than 4 hours 159 response time for moderate issues; online instructor-led and on-demand online 160 training for a total of 60 hours for 200 people; and remote in-depth architectural 161 support for refactoring applications and configuring cloud infrastructure totaling 162 80 hours. 163

Price Scenario 2 Sample Classified COTS deployment + Tactical Edge 165 A Military Service utilizes an Enterprise Resource Planning (ERP) tool at the Secret level with a 167 relational database included as part of the ERP to conduct supply, maintenance, and 168 transportation services. The ERP has approximately 30,000 active user accounts across all 169 commodities previously mentioned. This application is deployed to CONUS data centers, but has 170 local, synced mirrors in garrison and aboard ship and in numerous austere environments, 171 including a high likelihood of intermittent disconnected state. The ERP utilizes DNS Zone 172 Hosting services offered by the cloud provider hosting the ERP. The hosted DNS service will 173 receive 100 DNS requests per hour. The nature of this environment is such that a single user 174 conducting maintenance can submit multiple service requests each hour. 175 Supply transactions consist of: generating reports, which vary in size but involve querying into 177 the ERP’s database, submitting requisitions for parts, obligating funds, and receiving 178 requisitions. 179 Transportation transactions consist of managing shipments and movements of military 181 equipment providing logistical support to an area. Units submit movement requests for each 182 desired movement, identifying the requirement. The unit providing transportation then assigns 183 organic equipment, forwards the request for any of the requests they cannot support to other 184 units. Once a movement is completed a service request is completed for each vehicle used. 185 Emails are regularly sent to the requesting and approving users confirming the transportation 186 request status. Emails are sent from a hosted messaging service provided by the JEDI Cloud 187 service provider. The ERP system sends a total of 500 messages per day. 188 Across all of the areas above combined, the main CONUS ERP system deployment handles a 190 total of 1000 requests per minute with about three-quarters of those performing write operations 191 in the database. Assume each OCONUS garrison ERP system deployment handles a total of 200 192 requests per minute and each field ERP system deployment handles a total of 50 requests per 193 minute, regardless of the state of connectivity, with the same write operation ratio (three-194 quarters). Assume the size of each request for any ERP system deployment is 512 B and the 195 associated response is 128 KB. Each request is received by a simple, JEDI Cloud provided load 196 balancer which terminates SSL and then passes the request on to an available application node. 197 The JEDI Cloud virtual machines that the ERP system is running on require 8 vCPUs and 32 GB 198 of RAM per application node. Each virtual machine instance also requires 20 GB of simple Solid 199 State Disk storage. There are 50 application nodes in the main CONUS ERP system deployment, 200 20 application nodes per OCONUS garrison ERP system deployment, and 10 application nodes 201 per field ERP system deployment. Once per day, a snapshot of the entire provisioned block 202 storage for each virtual machine is taken. These snapshots are kept for 7 days. Each ERP system 203 deployment requires some JEDI Cloud virtual desktop working environments for use by the 204

Database Administrators (DBAs). These virtual desktops run Ubuntu, that requires the 205 equivalent of 2 vCPUs and 16GB of RAM, have 10 GB of storage space, and are solely used for 206 database access and tuning (they do not use or require any file sharing or additional systems or 207 services). The main CONUS ERP system deployment requires 10 virtual desktops, each 208 OCONUS garrison ERP system deployment requires 3 virtual desktops, and each field ERP 209 system deployment requires 1 virtual desktop. 210 The ERP system is deployed in OCONUS garrisons using static, modular, rapidly deployable 212 data centers. The ERP system is deployed in the field using ruggedized, portable edge devices 213 that support disconnected operations and automatic resync of all data with the rest of the ERP 214 system when possible. 215 To access the CONUS ERP system deployment while personnel are using non-government 217 communications networks, a site-to-site virtual private network (VPN) tunnel to the JEDI Cloud 218 must be utilized. 5 separate sites will each utilize a 1 Gigabit per second VPN connection 24 219 hours a day. The ERP will be expected to operate as both a completely self contained capability 220 per specific performance characteristics as defined in this scenario and also be capable of re-221 integrating (on the fly), to all appropriate external systems that require data exchange with the 222

ERP. 223

The database is currently 1 TB. A database dump will be uploaded to JEDI Cloud and then 225 imported into the new JEDI Cloud database on day 1 of the order (network usage, data storage, 226 and transaction fees associated with the transferring and importing of this data set should be 227 priced; assume the uploaded data is deleted after the import is completed; as with all scenarios, 228 do not price any migration support services). The database is expected to grow 0.5% per month. 229 The highly available database cluster for the main CONUS ERP system deployment will be 230 hosted in a JEDI Cloud service offering and requires 64 vCPUs and 400 GB of RAM. Each 231 OCONUS ERP system deployment and field ERP system deployment has a local database that 232 will use 2 compute nodes with 8 vCPUs, 64 GB of RAM, and 1TB of storage per node. Syncing 233 data between the main JEDI Cloud database and the local databases is handled by the ERP 234 system. Each OCONUS ERP system deployment writes 500 GB of data to JEDI Cloud once per 235 week, downloading any other changes that have been made. Each field ERP system deployment 236 writes 50 GB of data to JEDI Cloud once per week, downloading any other changes that have 237 been made. 238 Every action that changes infrastructure configuration or alters infrastructure state is logged 240 through a JEDI Cloud logging service, assuming there are 20 such actions a day. Additionally, 241 every data action taken for the managed data services is logged through a JEDI Cloud service. 242 Automated snapshots of the CONUS database occur daily and are kept online for 30 days before 243 rotation to offline storage along with weekly full backups taken every Saturday (starting the 244 week the order is placed) that are rotated to offline after 30 days. 245

a. Please price this scenario: The order is placed on January 6, 2020, for the above technical 247 requirements operating for 365 days continuously. In addition to the CONUS ERP system 248 deployment, assume 4 garrison OCONUS ERP system deployments and 30 field ERP system 249 deployments in ruggedized equipment. 250

i. Please separately price all IaaS and PaaS offerings in JEDI Cloud, the modular 251 data centers, and portable edge devices required to satisfy the above technical 252 requirements. 253

ii. Please price separately any non-consumption based fees or other charges 254 associated with the number of modular data centers and portable edge devices 255 required to meet the tactical edge storage and compute requirements for the 256 specified OCONUS and field ERP system deployments. 257

iii. Please separately price services under the Cloud Support Package line item to 258 support all instances of the ERP system described in the pricing scenario, 259 including at a minimum: 24x7 support by phone, web, and email; less than 1 hour 260 response time for critical issues; less than 4 hours response time for moderate 261 issues; instructor-led and on-demand online training for a total of 20 hours for 30 262 people; and remote in-depth architectural support for refactoring applications and 263 configuring cloud infrastructure totaling 40 hours. 264

iv. On January 30, 2020, a request is placed to retrieve a single offline backup of the 265 database for use in forensic investigation. Assume the backup size is 950 GB. 266 The retrieval shall be done as fast as possible, and the request is expedited. 267

b. Please price this scenario: The order is placed on June 7, 2027, for the above technical 269 requirements operating for 365 days continuously. In addition to the CONUS ERP 270 system deployment, assume 4 garrison OCONUS ERP system deployments and 30 field 271 ERP system deployments in ruggedized equipment. 272

i. Please separately price all IaaS and PaaS offerings in JEDI Cloud, the modular 273 data centers, and portable edge devices required to satisfy the above technical 274 requirements. 275

ii. Please separately price services under the Cloud Support Package line item to 276 support all instances of the ERP system described in the pricing scenario, 277 including at a minimum: 24x7 support by phone, web, and email; less than 1 hour 278 response time for critical issues; less than 4 hours response time for moderate 279 issues; instructor-led and on-demand online training for a total of 20 hours for 30 280 people; and remote in-depth architectural support for refactoring applications and 281 configuring cloud infrastructure totaling 40 hours. 282

iii. On April 30, 2027, a request is placed to retrieve a single offline backup of the 283 database for use in forensic investigation. Assume the backup size is 950 GB. 284

The retrieval shall be done as fast as possible, and the request is expedited. 285

Price Scenario 3 Sample Tactical Edge + Peered Query 288 A set of sensors on a government owned device captures 12 GB of High Definition Audio and 290 Video data per hour. This data is collected 24 hours a day. There are 40 of these government 291 devices deployed in the field for a specific operation. All of the sensor data from all of these 292 government owned devices is streamed to a dedicated cluster of JEDI Cloud ruggedized portable 293 edge devices in the field that has intermittent network access to a WAN. The cluster of one or 294 more JEDI Cloud ruggedized portable edge devices must be able to store up to 2 weeks of sensor 295 data. 296 The data must be processed daily by the portable edge devices regardless of connection; the 298 devices will use JEDI Cloud PaaS offerings to conduct image recognition and audio analysis on 299 the device itself. The analysis of video and audio will utilize a two stage, ensemble model. The 300 first stage of the model will evaluate audio and video separately. All of the models used by the 301 portable edge devices will be trained in JEDI Cloud and then synced to the devices before they 302 are taken into the field. The first stage video model will be trained on 100,000 210 KB labeled 4k 303 video frames. The audio model will be trained on 100,000 2 MB WAV files. The first stage 304 models will classify and tag data for triage in which triaged data will be forwarded to the second 305 stage of the ensemble model. The second model will take into account the output of both 306 previous first stage models and will have been trained on a separate set of 100,000 labeled and 307 paired audio and video files (assume this set also consists of 210 KB video frames and 2 MB 308 WAV files). 309 Twice a day, an operator in the field will run a real-time prediction analysis on the portable edge 311 devices using the ensemble model above. Results of the analysis must be accessible to, and 312 consumable by, commanders in the field using a separate viewing application that the Offeror 313 should not address as part of this scenario. Once connected to the WAN, all processed data is 314 securely transferred to the user’s JEDI Cloud account (i.e., not on the portable device), including 315 any meta-data and analysis results. On a weekly basis the raw sensor data is transferred back to 316 the user’s JEDI cloud account and stored in nearline storage for a period of 4 weeks before being 317 rotated to offline storage. Offerors may assume that the weekly raw sensor data transfer occurs 318 when a stable, high bandwidth connection is established between each portable device and JEDI 319 Cloud. 320 Assume that the scenario described in the previous paragraphs is occurring for 10 separate 322 operations simultaneously. 364 days after the initial order date, and once each month thereafter, 323 advanced data analysis (referred to as the periodic advanced data analysis) using JEDI Cloud 324 business intelligence PaaS offerings must be performed against a subset of the datasets across all 325 10 accounts. Each monthly analysis requires 32 vCPUs, 400 GB of RAM, and takes 36 hours to 326 complete. There are 100 users with access to the analytics platform. As part of this cross-account 327 advanced data analysis, each of the 10 JEDI Cloud accounts also has a single JEDI Cloud 328 relational database containing 10 TB of corroborating data. Each database is highly available, 329 requires 32 vCPUs, and has 200 GB of RAM. The cross-account analysis must include data from 330 these 10 traditional relationship databases (1 for each account, so 10 total). The cross-account 331 analysis will be looking for specific patterns as well as anomalies. The total data to be analyzed 332 is 10 PB (assume 2 PB of processed data, 500 TB of raw sensor data in online storage, and 7.5 333 PB of raw sensor data in offline storage that is loaded with a normal, non-expedited request). 334 Predictions from this analysis are performed in batch (not real-time) and must be stored in 335 another JEDI Cloud account (do not price any costs associated with the transferring or storing of 336 the prediction results). 337

a. Please price this scenario: The order is placed on September 2, 2019, for the above 339 technical requirements and operating for 15 months continuously where the classification 340 level of the source data and the analysis results is Unclassified. 341

i. Please price separately the described number of tactical edge storage and compute 342 devices. 343

ii. Please price separately all IaaS and PaaS offerings required to transmit and collect 344 the data (across all of the accounts specified in the scenario). 345

iii. Please price separately all IaaS and PaaS offerings required to perform the 346 advanced data analysis using cloud hosted, Offeror provided services across all of 347 the accounts specified in the scenario. 348

iv. Please price separately all costs such that for every operation, there is an 349 additional 10% identical spare portable tactical edge devices on-hand (minimum 350

2) at the base running the operation in case of failure (but that are not being 351 utilized). 352

v. Identify and price any fee required upon return of the device to the vendor. 353

vi. Please price separately secure destruction of all classified storage media 354

(applicable to paragraphs c, d, e, and f below in this price scenario) used upon 355 mission termination in accordance with the Cyber Security Plan. 356

vii. Include any fees should the devices be entirely destroyed 364 days after the initial 357 order is placed while in Government possession. 358

viii. Please separately price services under the Cloud Support Package line item to 359 support the applications described in the pricing scenario for the initial 12 months 360 only, including at a minimum: 24x7 support by phone, web, and email; less than 1 361 hour response time for critical issues; less than 4 hours response time for 362 moderate issues; instructor-led and on-demand online training for a total of 40 363 hours for 50 people; and remote in-depth architectural support for refactoring 364 applications and configuring cloud infrastructure totaling 80 hours. 365

ix. Identify any additional fee(s) if the Government retains the devices for an 366 additional 12 months beyond the one year initial period of performance without 367 having placed another order. For purposes of this price scenario, assume that the 368 ID/IQ option ordering period is exercised. 369

b. Please also price this scenario in accordance with the entirety of paragraph (a) 371 above, but assume the order is placed on June 7, 2027. 372

c. Please also price this scenario in accordance with entirety of paragraph (a) above, 374 but assume: the order is placed on June 7, 2021, the classification level of the source 375 data and data analysis on the portable devices at the tactical edge is unclassified, a copy 376 of all of the data transferred from all portable devices to any JEDI Cloud account is also 377 simultaneously transferred to a single JEDI Cloud account operating at Secret 378 classification, and the periodic advanced data analysis applications and results are 379 classified at Secret. 380

d. Please also price this scenario in accordance with entirety of paragraph (a) above, 382 but assume: the order is placed on September 2, 2024, the classification level of the 383 source data is Secret, and all analysis applications and results are classified at Secret. 384

e. Please also price this scenario in accordance with entirety of paragraph (a) above, 386 but assume: the order is placed on September 2, 2024, the classification level of the of 387 the source data is Top Secret/SCI, and all analysis applications and results are classified 388 at Top Secret/SCI. 389

f. Please also price this scenario in accordance with entirety of paragraph (a) above, 391 but assume: the order is placed on July 5, 2027, the classification level of the entire 392 program has been deemed a SAP at the TS/SCI level. 393

Price Scenario 4 Large Data Storage, Analysis, and Archiving 396 The following is an example of large data storage and follow up data analysis that retrieves and 398 processes chunks of data from a large offline set. 399 A collection of systems across multiple accounts (assume 100) produces a large number of log 401 entries every day. There are two different sources for the logs: (1) logs from the JEDI Cloud IaaS 402 and PaaS offerings and (2) logs from the applications deployed in those accounts. Assume a total 403 of 500 GB of logs in aggregate from all of the accounts each day. These log entries must be 404 stored in a separate JEDI Cloud account (referred to as log collection account), which the 405 individual account users have no access to (do not price the collecting, aggregating, or 406 transferring of these log files into the log collection account). The team managing the log 407 collection account only has read access to the storage. All data in the log collection account is 408 rotated to nearline storage after 30 days, and then to offline storage after 90 days from creation 409 and retained in perpetuity. Regular scans of these logs occur nightly and on-demand as described 410 below. 411 The log collection and analysis team also has another JEDI Cloud account (referred to as the 413 analysis account) for the analysis application used to analyze the data in the log collection 414 account. The analysis application stack consists of a set of ten (10) high performance GPU nodes 415 performing the analysis jobs (At Least 2 GPUs, 10 vCPUs, and 200GB of RAM per node), a set 416 of four (4) moderate performance compute nodes hosting a clustered queueing and messaging 417 system (At Least 8 vCPUs and 32GB of RAM per node), and a set of four (4) low performance 418 compute nodes hosting a web-based graphical user interface (GUI) (At Least 1 vCPU and 1GB 419 of RAM per node). The high performance GPU nodes each require an additional 50 GB of 420 simple solid state disk (SSD) block storage to be provisioned. The web application is behind a 421 JEDI Cloud load balancer that receives 100 requests per minute (512B request, 40KB response). 422 The web application requires a single, low performance (but highly available) relational database 423 that is provided as a JEDI Cloud PaaS offering (At Least 2 vCPU, 8 GB of RAM, and 100 GB of 424 storage). Assume that there is already 2 PB of offline storage, 30 TB of nearline storage, and 15 425 TB of online storage in the log collection account on day 1 of the order (do not price any costs 426 associated with the migration of all of this data to JEDI Cloud). For pricing purposes, “At Least” 427 in this paragraph means that if the proposed JEDI Cloud service(s) do not identically match the 428 specified minimum technical requirements, then the Offeror must propose the service(s) that 429 satisfy those minimum technical requirements even if the level of service exceeds what is 430 required by the scenario; in no event may an Offeror propose a service that does not meet 431 minimum technical requirements). 432 Assume that 10 on-demand analysis jobs are run each week, with 4 pulling from online data, 3 434 from nearline data, and 3 from offline data. For each online on-demand analysis job assume the 435 average target dataset is 5 TB. For each nearline on-demand analysis job assume the average 436 target dataset is 10 TB. For each offline on-demand analysis job assume the average target 437 dataset is 50 TB and that offline data may be brought online with slow access (not expedited). In 438 each case, the data is first copied in bulk to online storage (if necessary) and then analyzed. Any 439 appropriate tags or markings identified by the analysis are applied to the original source data. 440 The tags will be used in identifying that raw data at a later time for security review. Once the 441 analysis is complete, if the data being analyzed was an online copy of nearline or offline data, the 442 online copy is deleted. Nightly analysis jobs (assume 20 each night) run over online data only, 443 but may process log files from any time during the 30 day rotational period in which data is kept 444 online. Assume these nightly jobs process an average of 500 GB of online data each. 445 In total, analysis results consume 200 GB of storage per day on average. Analysis result data are 447 tagged based on their source data, method of initiation, and other key markers and those tags are 448 used in technical policy review and for billing purposes. Analysis results are saved in online 449 storage for 45 days, nearline storage for another 90 days, and then archived into offline storage. 450 Results stored offline are discarded after four (4) years. Assume the application has 4.5 TB of 451 online result data, 4.5 TB of nearline result data, and 72 TB of offline result data on day 1 of the 452 order (do not price any costs associated with the migration of all of this data to JEDI Cloud). 453 A user of the analysis application must be able to override the default lifecycle configuration set 455 for the results such as setting alternate expiration dates and moving analysis results from one 456 storage class to another. The application will execute API commands against the JEDI Cloud 457 provider to accomplish this. Assume that users will transition 500 GB of storage from nearline to 458 online each week and 200 GB of offline to online each week (not expedited). 459

a. Please price this scenario: The order is placed on January 6, 2020, for the above 461 technical requirements operating for 365 days continuously. 462

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 463 technical requirements. 464

ii. Please separately price services under the Cloud Support Package line item to 465 support the applications described in the pricing scenario, including at a 466 minimum: 24x7 support by phone, web, and email; less than 1 hour response time 467 for critical issues; access to detailed online training materials; and guidance on 468 application and infrastructure architecture by phone using the lowest applicable 469 tier. 470

b. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 472 placed on January 6, 2022. 473

c. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 474 placed on January 6, 2025. 475

d. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 476 placed on January 6, 2028. 477

e. Please price this scenario: The order is placed on October 31, 2023, for a complete 479 exportation of all log data from the log collection account and all analysis results 480 associated with the analysis application account described in this scenario. Assume log 481 data totals of 15 TB of online data, 30 TB of nearline data, and 3 PB of offline data. 482 Assume analysis result data totals of 5 TB of online data, 5 TB of nearline data, and 90 483 TB of offline data. Assume this request is not time-sensitive (no rush request). Assume 484 the data must be transferred to an on-premise data storage solution provided by the 485 Department that is capable of supporting the relevant data types in this scenario. 486

f. Please also price paragraph (c) above, but assuming that the order for paragraph (c) is 488 placed on April 5, 2026. 489

g. Please also price paragraph (c) above, but assuming that the order for paragraph (c) is 491 placed on April 5, 2028. 492

Pricing Scenario 5 Rapidly Deployed, Static Data Center 494 There is a forward operating base (FOB) that is processing large quantities of data and regularly 496 engages in various activities that need large, elastic computing power. They need to rapidly get 497 access to such computing power through static, modular, rapidly deployable data center(s) in 498 short order. 499 This ruggedized data center solution must be delivered to a CONUS U.S. Military base within 501 the number of days specified in performance metric 31 in Table 5.1 of the SOO from the order 502 date. The ruggedized data center must be able to fit and be properly secured inside of a U.S. 503 military cargo aircraft or ship and commercial shipping vessel. The data center should be able to 504 operate fully (as specified below) in a fully disconnected state. Assume a solid pad for placement 505 and clean power connections are provided by the FOB. The data center must be modular 506 meaning that additional computing and storage resources can be added at a later date. 507 At minimum, each FOB site requires data center(s) to have: 509

(a) Usable storage capacity for 100 PB of data, spread across all forms of storage (objects, 510 files, databases, file systems, etc.). All storage must be redundant (three copy minimum); 511

(b) Computing power of 2000 virtual CPU cores and 200 virtual GPU cores; 512

(c) Multiple data uplink options to include fiber optic, low and high bandwidth ethernet, and 513 compatibility with standard satcom systems; and 514

(d) Implementation that meets CNSSAM TEMPEST/01-13: Red/Black Installation Guidance 515 with regard to physical separation of environment where necessary. 516

a. Please price this scenario: The order is placed on November 4, 2019, for the above 518 technical requirements operating for 365 days continuously for 10 FOB military 519 deployment sites where there is an equal split of Unclassified, Secret, and Top Secret 520 workloads at each site, including maintaining logical and physical isolation, as 521 appropriate. 522

i. Please price separately all 10 FOB sites. 523

ii. In pricing the 10 FOB sites, for any consumption-based charges, assume a single 524 copy of the application stack described in Pricing Scenario 1 is running in each 525 modular data center throughout the duration of the order. Assume any nearline or 526 offline storage remains online storage for the duration of the order. 527

iii. Please separately price services under the Cloud Support Package line item to 528 support the data center deployments described in the pricing scenario, including at 529 a minimum: 24x7 support by phone, web, and email; less than 1 hour response 530 time for critical issues; less than 4 hours response time for moderate issues; 531 instructor-led and on-demand online training for a total of 60 hours for 200 532 people; and remote in-depth architectural support for refactoring applications and 533 configuring cloud infrastructure totaling 80 hours. 534

iv. Identify and price any fees required upon return of the data centers to the vendor. 535

v. Identify and price any fees for secure destruction of all classified storage media 536 used upon mission termination in accordance with the Cyber Security Plan. 537

vi. Include any fees should the data centers be entirely destroyed at any point while 538 in Government possession. 539

vii. Identify any additional fee(s) if the Government retains the data centers for an 540 additional 5 months beyond the one year initial period of performance without 541 having placed another order. 542

b. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 543 placed on October 4, 2021. 544

c. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 545 placed on October 7, 2024. 546

d. Please also price paragraph (a) above, but assuming that the order for paragraph (a) is 547 placed on October 4, 2027. 548

Pricing Scenario 6 Containerized Data Analysis Framework 550 A military Service runs flight operations at a given base that produce logs on system function 552 and maintenance needs. Separately, a maintenance system is used by the Service to enter reports 553 on maintenance actions and scheduling. Each flight operation results in a large amount of data 554 that is streamed after flight completion to a data warehouse hosted in JEDI Cloud. Maintenance 555 records are stored in a separate system not hosted in JEDI Cloud. The Service will perform near 556 real-time predictive analysis on both the flight data and maintenance records to determine future 557 maintenance needs. 558 Each flight mission produces 500 GB of structured binary data per aircraft. Assume 20 flights 560 per 24-hour period at the base. This data is streamed into the JEDI Cloud data warehouse directly 561 post-flight. Assume the data warehouse storage backend has 1 PB of data in it on day 1 (do not 562 price any costs associated with the migration of all of this data to JEDI Cloud). A total of 250 563 maintenance records are created per day at this base in the maintenance system (this system is 564 not to be priced in this scenario). The maintenance system converts each record into a 500 KB 565 structured record, which is sent via API call to a highly available JEDI Cloud serverless function, 566 which requires 2.5GB of RAM to run. This serverless function parses the incoming data for 567 validity and stores it in a highly available JEDI Cloud NoSQL document-based data store. 568 Assume this NoSQL data store is 100 GB on day 1 (do not price any costs associated with the 569 migration of all of this data to JEDI Cloud). The run time for each execution of this function is 1 570 second. In addition, to format validity of the maintenance record, the identity of the maintenance 571 worker and their authorization to submit records for that aircraft is validated through a JEDI 572 Cloud directory service. Assume there are a total of 10,000 identity objects and 1,000 other 573 directory objects in the directory service for the base. 574 The flight operations system will push an event to a separate, highly available JEDI Cloud 576 serverless function for each flight mission once that flight’s operational data has been uploaded. 577 The serverless function requires 512MB of RAM to run. This serverless endpoint will in turn add 578 1000 events to a message queue in a JEDI Cloud message service (each message is 50KB). The 579 run time for each execution of this function is 500 milliseconds. Each message will be consumed 580 by a data analysis application, which will start a data analysis job, which is hosted in JEDI 581 Cloud. Each analysis job constructs and submits a query to the data warehouse service, which 582 returns 1 GB of data. The results of the query are analyzed alongside the maintenance record 583 data from the document store. The results of this analysis consume 5 MB of data and are stored 584 in a highly available simple NoSQL key-value based data store. Assume this NoSQL data store 585 is 500 MB on day 1. There are 10 such analysis applications in use within the overall system. 586 Each data analysis application stack is a collection of containers in a microservices architecture 588 managed by a JEDI Cloud container orchestration service. Assume each application consists of 4 589 distinct microservices, hosted in 12 separate containers, and each container requires 2 vCPU, 2 590 GB of RAM, and 1 GB of storage. The application code for each microservice for each analysis 591 job is version controlled in its own JEDI Cloud hosted code repository. On code commit to any 592 master branch a set of tests are run using continuous integration hosted by JEDI Cloud. Each 593 build for each of the code repositories requires that the build service utilize 2 vCPUs and 2 GB of 594 RAM for 10 minutes. Assume a total of 10 builds are run per day. Following successful passing 595 of the tests, a build artifact is produced and the resulting container is scheduled for zero-596 downtime redeployment. Assume 50 containers, that are each 1 GB in size, are generated and 597 uploaded to a JEDI Cloud container registry every 7 days. The container registry contains 144 1-598 GB containers on day 1 (do not price any costs associated with the migration of all of this data to 599 JEDI Cloud), and old containers are removed at the same rate that new containers are added (the 600 total number remains constant). 601 A JEDI Cloud web application firewall is employed to protect the analysis applications from 603 unwanted traffic. There are 30 custom rules in addition to any standard rules applied by the JEDI 604 Cloud vendor. In total, the data analysis applications receive 100,000 requests per day between 605 both valid and malicious actions. Automated security scans of the data analysis applications and 606 data is performed daily, which includes threat identification, reporting, and real-time notification 607 to the system owner. 608

a. Please price this scenario: The order is placed on January 6, 2020, for the above 610 technical requirements operating for 365 days continuously where the classification level 611 of the entire system is unclassified. This order is for operations on a single base. 612

i. Please separately price all IaaS and PaaS offerings required to satisfy the above 613 technical requirements. 614

ii. Please separately price services under the Cloud Support Package line item to 615 support all of the applications and services described in the pricing scenario, 616 including at a minimum: 24x7 support by phone, web, and email; less than 1 hour 617 response time for critical issues; less than 4 hours response time for moderate 618 issues; instructor-led and on-demand online training for a total of 20 hours for 30 619 people; and remote in-depth architectural support for refactoring applications and 620 configuring cloud infrastructure totaling 40 hours. 621

b. Please also price this scenario in accordance with entirety of paragraph (a) above, 622 but assume the order is placed on April 4, 2022, and the scenario is replicated at 100 623 bases. 624

c. Please also price this scenario in accordance with entirety of paragraph (a) above, 626 but assume the order is placed on September 5, 2022, and the system has been classified 627 secret. 628

d. Please also price this scenario in accordance with entirety of paragraph (a) above, 630 but assume the order is placed on September 5, 2022, the system has been classified 631 Secret, and the scenario is replicated at 100 bases. 632

e. Please also price this scenario in accordance with entirety of paragraph (a) above, 634 but assume the order is placed on July 1, 2024, and the system has been classified Top 635 Secret / SCI. 636

f. Please also price this scenario in accordance with entirety of paragraph (a) above, 638 but assume the order is placed on July 1, 2024, the system has been classified Top Secret 639 / SCI, and the scenario is replicated at 100 bases. 640

g. Please also price this scenario in accordance with entirety of paragraph (a) above, 642 but assume the order is placed on January 3, 2028, and the system has been classified as a 643 SAP at the TS/SCI level. 644

h. Please also price this scenario in accordance with entirety of paragraph (a) above, 646 but assume the order is placed on January 3, 2028, the system has been classified as a 647 SAP at the TS/SCI level, and the scenario is replicated at ten (10) bases. 648