)]}'
{"/COMMIT_MSG":[{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":7,"context_line":"FPGA resource use cases"},{"line_number":8,"context_line":""},{"line_number":9,"context_line":"This is a proposal how FPGA type of accelerators might be managed within"},{"line_number":10,"context_line":"OpenSstack."},{"line_number":11,"context_line":""},{"line_number":12,"context_line":"Change-Id: I835dd8bc5dd56635e86f23d53f60e83d7c70fffe"}],"source_content_type":"text/x-gerrit-commit-message","patch_set":3,"id":"bab6814e_7866552a","line":10,"range":{"start_line":10,"start_character":0,"end_line":10,"end_character":10},"updated":"2016-05-25 12:37:05.000000000","message":"s/OpenSstack/OpenStack","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"}],"specs/backlog/approved/fpga-use-cases.rst":[{"author":{"_account_id":2750,"name":"Sean Dague","email":"sean@dague.net","username":"sdague"},"change_message_id":"72db0b791c15d3fe2d20293e3e9505004641c6bb","unresolved":false,"context_lines":[{"line_number":69,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"For use cases in first and second point, there should be an automatic way for"},{"line_number":72,"context_line":"discovery, programming and managing such resource within Nova resource manager."},{"line_number":73,"context_line":""},{"line_number":74,"context_line":"Alternatives"},{"line_number":75,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":2,"id":"bab6814e_8be2852d","line":72,"updated":"2016-05-20 14:45:57.000000000","message":"I think that it is fine to say we all agree Nova should use FPGA topology information in placement. I think saying that Nova must be responsible for FPGA programming is a bridge too far.","commit_id":"2321f92599b25d96dbfcb03a2205452a78084db0"},{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":22,"context_line":""},{"line_number":23,"context_line":"Furthermore, modern FPGA provide us with additional features like ability to"},{"line_number":24,"context_line":"divide array to slots/regions, which might me (re)programmed separately, and"},{"line_number":25,"context_line":"each programmed slot/region may be use by several VMs (sometimes it is called"},{"line_number":26,"context_line":"context)."},{"line_number":27,"context_line":""},{"line_number":28,"context_line":"To systematize, FPGA can be resource in context of:"},{"line_number":29,"context_line":""}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_b301fe6c","line":26,"range":{"start_line":25,"start_character":0,"end_line":26,"end_character":9},"updated":"2016-05-25 12:37:05.000000000","message":"IMO this should be emphasized - that\u0027s the main reason why passing an FPGA device as blank PCI device is inefficient.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ea23da488a45dd89a7d2c44f5b6d62fe5ded15b0","unresolved":false,"context_lines":[{"line_number":22,"context_line":""},{"line_number":23,"context_line":"Furthermore, modern FPGA provide us with additional features like ability to"},{"line_number":24,"context_line":"divide array to slots/regions, which might me (re)programmed separately, and"},{"line_number":25,"context_line":"each programmed slot/region may be use by several VMs (sometimes it is called"},{"line_number":26,"context_line":"context)."},{"line_number":27,"context_line":""},{"line_number":28,"context_line":"To systematize, FPGA can be resource in context of:"},{"line_number":29,"context_line":""}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_8115614a","line":26,"range":{"start_line":25,"start_character":0,"end_line":26,"end_character":9},"in_reply_to":"bab6814e_b301fe6c","updated":"2016-05-31 12:24:05.000000000","message":"Ok.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":53,"context_line":"from the user."},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"1. As an User, I want to run specific workload, which which needs certain"},{"line_number":56,"context_line":"   hardware or acceleration, so I specify a flavor or image so that scheduler"},{"line_number":57,"context_line":"   would pick up right host with available hardware for me."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_f359c65c","line":56,"range":{"start_line":56,"start_character":47,"end_line":56,"end_character":48},"updated":"2016-05-25 12:37:05.000000000","message":"Actually flavors are normally provided by admins.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ea23da488a45dd89a7d2c44f5b6d62fe5ded15b0","unresolved":false,"context_lines":[{"line_number":53,"context_line":"from the user."},{"line_number":54,"context_line":""},{"line_number":55,"context_line":"1. As an User, I want to run specific workload, which which needs certain"},{"line_number":56,"context_line":"   hardware or acceleration, so I specify a flavor or image so that scheduler"},{"line_number":57,"context_line":"   would pick up right host with available hardware for me."},{"line_number":58,"context_line":""},{"line_number":59,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_dc958c87","line":56,"range":{"start_line":56,"start_character":47,"end_line":56,"end_character":48},"in_reply_to":"bab6814e_f359c65c","updated":"2016-05-31 12:24:05.000000000","message":"Right. But I didn\u0027t state anywhere that user is creating any flavor :)","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":59,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"},{"line_number":60,"context_line":"   FPGA on board, to make them manageable by Nova resource tracker."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"3. As an Operator, I\u0027d like to be able to program FPGA accelerators and shared"},{"line_number":63,"context_line":"   among the VMs using SR-IOV."},{"line_number":64,"context_line":""},{"line_number":65,"context_line":"4. As a Developer, I want to be able to use VM with entire FPGA exposed to it,"}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_3354ee1f","line":62,"range":{"start_line":62,"start_character":77,"end_line":62,"end_character":78},"updated":"2016-05-25 12:37:05.000000000","message":"share?","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ea23da488a45dd89a7d2c44f5b6d62fe5ded15b0","unresolved":false,"context_lines":[{"line_number":59,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"},{"line_number":60,"context_line":"   FPGA on board, to make them manageable by Nova resource tracker."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"3. As an Operator, I\u0027d like to be able to program FPGA accelerators and shared"},{"line_number":63,"context_line":"   among the VMs using SR-IOV."},{"line_number":64,"context_line":""},{"line_number":65,"context_line":"4. As a Developer, I want to be able to use VM with entire FPGA exposed to it,"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_fc27b0c7","line":62,"range":{"start_line":62,"start_character":77,"end_line":62,"end_character":78},"in_reply_to":"bab6814e_3354ee1f","updated":"2016-05-31 12:24:05.000000000","message":"Done","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"cea80d564dce05bf4498526c3a58816dd99bac8f","unresolved":false,"context_lines":[{"line_number":69,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"For use cases in first and second point, there should be an automatic way for"},{"line_number":72,"context_line":"discovery and managing such resource within Nova resource manager. Other than"},{"line_number":73,"context_line":"that, external service should be responsible for (re)programming FPGA. Nova"},{"line_number":74,"context_line":"should be able to send a request to such service, and wait for the response."},{"line_number":75,"context_line":""},{"line_number":76,"context_line":"Alternatives"},{"line_number":77,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_5b5f4b75","line":74,"range":{"start_line":72,"start_character":67,"end_line":74,"end_character":76},"updated":"2016-05-24 12:14:52.000000000","message":"This new service will exclusively take care about preparing (so in the case of FPGA - programming) and will be totally separate thing from Nova. One thing is still missing in this puzzle (assuming, that dynamic accelerator placement is a good thing for operators) is the way for Nova to wait, for selected node to be prepared - there might be an issue to libvirt to assign certain hardware, which is not-yet-available.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ea23da488a45dd89a7d2c44f5b6d62fe5ded15b0","unresolved":false,"context_lines":[{"line_number":69,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"For use cases in first and second point, there should be an automatic way for"},{"line_number":72,"context_line":"discovery and managing such resource within Nova resource manager. Other than"},{"line_number":73,"context_line":"that, external service should be responsible for (re)programming FPGA. Nova"},{"line_number":74,"context_line":"should be able to send a request to such service, and wait for the response."},{"line_number":75,"context_line":""},{"line_number":76,"context_line":"Alternatives"},{"line_number":77,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_d740c736","line":74,"range":{"start_line":72,"start_character":67,"end_line":74,"end_character":76},"in_reply_to":"bab6814e_3ee31535","updated":"2016-05-31 12:24:05.000000000","message":"That\u0027s a good point. I\u0027ll add it as an alternative to the service.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":69,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"},{"line_number":70,"context_line":""},{"line_number":71,"context_line":"For use cases in first and second point, there should be an automatic way for"},{"line_number":72,"context_line":"discovery and managing such resource within Nova resource manager. Other than"},{"line_number":73,"context_line":"that, external service should be responsible for (re)programming FPGA. Nova"},{"line_number":74,"context_line":"should be able to send a request to such service, and wait for the response."},{"line_number":75,"context_line":""},{"line_number":76,"context_line":"Alternatives"},{"line_number":77,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_3ee31535","line":74,"range":{"start_line":72,"start_character":67,"end_line":74,"end_character":76},"in_reply_to":"bab6814e_5b5f4b75","updated":"2016-05-25 12:37:05.000000000","message":"Isn\u0027t an os-brick-like library a simpler option than external service with it\u0027s own REST API? What else will be done by the service other than discovery, reservations, reprogramming and attachment of the devices? It will still be Nova triggering these actions anyway.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":11600,"name":"Michał Dulko","email":"michal.dulko@gmail.com","username":"dulek"},"change_message_id":"2e32fdb24ea03a658c4f6548dc11ebfcd3c8879c","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Security impact"},{"line_number":90,"context_line":"---------------"},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Note, although no security impact to the stack itself may be observed, there"},{"line_number":93,"context_line":"could be potential risks in case of bare metal with allowing Developer to"},{"line_number":94,"context_line":"program it\u0027s own algorithms. The issue here is with algorithm itself, and may"},{"line_number":95,"context_line":"be reveal as abnormal power consumption with risk of hardware malfunction or"},{"line_number":96,"context_line":"even physical damage."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Notifications impact"},{"line_number":99,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"bab6814e_593553d8","line":96,"range":{"start_line":92,"start_character":0,"end_line":96,"end_character":21},"updated":"2016-05-25 12:37:05.000000000","message":"Not only baremetal case poses risk - any situation giving users access to program anything on a FPGA device can cause risk - VMs could do that too in case of device passthrough.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":22063,"name":"Fei Chen","email":"uchen@cn.ibm.com","username":"chenfei"},"change_message_id":"e62f71de21cf1af11499ffd372e1e94fb9574c11","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Security impact"},{"line_number":90,"context_line":"---------------"},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Note, although no security impact to the stack itself may be observed, there"},{"line_number":93,"context_line":"could be potential risks in case of bare metal with allowing Developer to"},{"line_number":94,"context_line":"program it\u0027s own algorithms. The issue here is with algorithm itself, and may"},{"line_number":95,"context_line":"be reveal as abnormal power consumption with risk of hardware malfunction or"},{"line_number":96,"context_line":"even physical damage."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Notifications impact"},{"line_number":99,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_fa24cbc5","line":96,"range":{"start_line":92,"start_character":0,"end_line":96,"end_character":21},"in_reply_to":"9abb7d3a_97447f1b","updated":"2016-06-01 05:12:32.000000000","message":"So is it your opinion that we should stop users programming FPGA inside his VM and bare metal environment because this risk?","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ffb72c9d30ea3acb9282b10b9dbc28f0f2cac412","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Security impact"},{"line_number":90,"context_line":"---------------"},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Note, although no security impact to the stack itself may be observed, there"},{"line_number":93,"context_line":"could be potential risks in case of bare metal with allowing Developer to"},{"line_number":94,"context_line":"program it\u0027s own algorithms. The issue here is with algorithm itself, and may"},{"line_number":95,"context_line":"be reveal as abnormal power consumption with risk of hardware malfunction or"},{"line_number":96,"context_line":"even physical damage."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Notifications impact"},{"line_number":99,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_1d93205c","line":96,"range":{"start_line":92,"start_character":0,"end_line":96,"end_character":21},"in_reply_to":"9abb7d3a_fa24cbc5","updated":"2016-06-02 08:08:46.000000000","message":"No, definitely not. I would leave that decision to cloud admin, through either configuration option or creating or not specific flavor which expose whole FPGA, or both.\n\nI can imagine a situation, where developer needs to perform functional tests, benchmarks or any other activities which requires access to physical device. Administrator of private cloud could prepare couple of FPGA equipped platforms, so that developers could use them if they need to.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"ea23da488a45dd89a7d2c44f5b6d62fe5ded15b0","unresolved":false,"context_lines":[{"line_number":89,"context_line":"Security impact"},{"line_number":90,"context_line":"---------------"},{"line_number":91,"context_line":""},{"line_number":92,"context_line":"Note, although no security impact to the stack itself may be observed, there"},{"line_number":93,"context_line":"could be potential risks in case of bare metal with allowing Developer to"},{"line_number":94,"context_line":"program it\u0027s own algorithms. The issue here is with algorithm itself, and may"},{"line_number":95,"context_line":"be reveal as abnormal power consumption with risk of hardware malfunction or"},{"line_number":96,"context_line":"even physical damage."},{"line_number":97,"context_line":""},{"line_number":98,"context_line":"Notifications impact"},{"line_number":99,"context_line":"--------------------"}],"source_content_type":"text/x-rst","patch_set":3,"id":"9abb7d3a_97447f1b","line":96,"range":{"start_line":92,"start_character":0,"end_line":96,"end_character":21},"in_reply_to":"bab6814e_593553d8","updated":"2016-05-31 12:24:05.000000000","message":"Right, allowing user for (re)programing an FPGA which is exposed via PCI pass through to the VM is the same hazard.","commit_id":"8ffcce7c181ccca539432799beea0b9345d8802b"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a56f3f2b4135ef0d25d1566a0137f36b297c4444","unresolved":false,"context_lines":[{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/fpga-resource"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"Currently, Nova lacks of ability to manage resources, which have mutable"},{"line_number":14,"context_line":"nature, just like the FPGA based accelerators."},{"line_number":15,"context_line":""},{"line_number":16,"context_line":"Problem description"},{"line_number":17,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":5,"id":"5a8bc5a2_5e30e89e","line":14,"range":{"start_line":13,"start_character":0,"end_line":14,"end_character":46},"updated":"2016-09-28 12:28:36.000000000","message":"Is this still true with the resource-providers API in place?","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"0fbf85b9d5ada6b66636da2e3044091b15dcc9b9","unresolved":false,"context_lines":[{"line_number":10,"context_line":""},{"line_number":11,"context_line":"https://blueprints.launchpad.net/nova/+spec/fpga-resource"},{"line_number":12,"context_line":""},{"line_number":13,"context_line":"Currently, Nova lacks of ability to manage resources, which have mutable"},{"line_number":14,"context_line":"nature, just like the FPGA based accelerators."},{"line_number":15,"context_line":""},{"line_number":16,"context_line":"Problem description"},{"line_number":17,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":5,"id":"3a98d1f4_0d4941d0","line":14,"range":{"start_line":13,"start_character":0,"end_line":14,"end_character":46},"in_reply_to":"5a8bc5a2_5e30e89e","updated":"2016-10-03 08:08:52.000000000","message":"Until custom-resource-classes land, there is no way for defining accelerator as a resource for the minimal use case, which means already programmed accelerator on FPGA and manually added to resources.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":1063,"name":"Ed Leafe","email":"ed@leafe.com","username":"ed-leafe"},"change_message_id":"fcd4e51b848f268e8ea22de00120e8a6a7e3af44","unresolved":false,"context_lines":[{"line_number":64,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"},{"line_number":65,"context_line":"   FPGA on board, to make them manageable by Nova resource tracker."},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"3. As an Operator, I\u0027d like to be able to program FPGA accelerators and share"},{"line_number":68,"context_line":"   among the VMs using SR-IOV."},{"line_number":69,"context_line":""},{"line_number":70,"context_line":"4. As a Developer, I want to be able to use VM with entire FPGA exposed to it,"},{"line_number":71,"context_line":"   or boot bare metal with FPGA for programming purposes."}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a629dbe_120de7e7","line":68,"range":{"start_line":67,"start_character":3,"end_line":68,"end_character":30},"updated":"2016-11-10 16:47:42.000000000","message":"The programming will not be a part of Nova. It is not what Nova is in the business of doing. Once programmed by something external, though, the programmed algorithms can be added as resource inventory (once custom resource classes merges, of course).","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":6804,"name":"bruce-benjamin","email":"bruce.benjamin@jhuapl.edu","username":"bruce-benjamin"},"change_message_id":"b00244bee6bb03e39838d71957b91741d195347d","unresolved":false,"context_lines":[{"line_number":64,"context_line":"2. As an Operator, I want to have a configure options for compute hosts with"},{"line_number":65,"context_line":"   FPGA on board, to make them manageable by Nova resource tracker."},{"line_number":66,"context_line":""},{"line_number":67,"context_line":"3. As an Operator, I\u0027d like to be able to program FPGA accelerators and share"},{"line_number":68,"context_line":"   among the VMs using SR-IOV."},{"line_number":69,"context_line":""},{"line_number":70,"context_line":"4. As a Developer, I want to be able to use VM with entire FPGA exposed to it,"},{"line_number":71,"context_line":"   or boot bare metal with FPGA for programming purposes."}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a629dbe_6d87f8b4","line":68,"range":{"start_line":67,"start_character":3,"end_line":68,"end_character":30},"in_reply_to":"9a629dbe_120de7e7","updated":"2016-11-11 19:13:00.000000000","message":"+1 for Ed","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":1063,"name":"Ed Leafe","email":"ed@leafe.com","username":"ed-leafe"},"change_message_id":"fcd4e51b848f268e8ea22de00120e8a6a7e3af44","unresolved":false,"context_lines":[{"line_number":67,"context_line":"3. As an Operator, I\u0027d like to be able to program FPGA accelerators and share"},{"line_number":68,"context_line":"   among the VMs using SR-IOV."},{"line_number":69,"context_line":""},{"line_number":70,"context_line":"4. As a Developer, I want to be able to use VM with entire FPGA exposed to it,"},{"line_number":71,"context_line":"   or boot bare metal with FPGA for programming purposes."},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"Proposed change"},{"line_number":74,"context_line":"\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a629dbe_72e62316","line":71,"range":{"start_line":70,"start_character":3,"end_line":71,"end_character":57},"updated":"2016-11-10 16:47:42.000000000","message":"That would mean adding the FPGA as an inventory item, and not being able to add inventory for any programmed algorithms on that FPGA.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":15334,"name":"Stephen Finucane","display_name":"stephenfin","email":"stephenfin@redhat.com","username":"sfinucan"},"change_message_id":"a56f3f2b4135ef0d25d1566a0137f36b297c4444","unresolved":false,"context_lines":[{"line_number":82,"context_line":""},{"line_number":83,"context_line":"Alternatives"},{"line_number":84,"context_line":"------------"},{"line_number":85,"context_line":""},{"line_number":86,"context_line":"1. Instead of library, which Nova could use, new external service should be"},{"line_number":87,"context_line":"   introduced, which will be responsible for interaction with the FPGA, and"},{"line_number":88,"context_line":"   which will be a source of information about FPGA resources to Nova."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"2. We can do nothing, and leave the burden to the Operator, who will prepare"},{"line_number":91,"context_line":"   accelerators manually, and configure the nodes to expose their hardware via"}],"source_content_type":"text/x-rst","patch_set":5,"id":"5a8bc5a2_23622517","line":88,"range":{"start_line":85,"start_character":0,"end_line":88,"end_character":70},"updated":"2016-09-28 12:28:36.000000000","message":"This seems like a far more viable option, IMO. nova seems to have been moving away from doing stuff that are outside of purely configuring compute resources (think: nova-network -\u003e neutron, removal of various proxy services, EC2 api removal etc.). FPGA is a complicated enough use case that it really seems like something that should be done separately and perhaps exposed to nova via the resource providers API.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":13692,"name":"Roman Dobosz","email":"gryf73@gmail.com","username":"gryf"},"change_message_id":"0fbf85b9d5ada6b66636da2e3044091b15dcc9b9","unresolved":false,"context_lines":[{"line_number":82,"context_line":""},{"line_number":83,"context_line":"Alternatives"},{"line_number":84,"context_line":"------------"},{"line_number":85,"context_line":""},{"line_number":86,"context_line":"1. Instead of library, which Nova could use, new external service should be"},{"line_number":87,"context_line":"   introduced, which will be responsible for interaction with the FPGA, and"},{"line_number":88,"context_line":"   which will be a source of information about FPGA resources to Nova."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"2. We can do nothing, and leave the burden to the Operator, who will prepare"},{"line_number":91,"context_line":"   accelerators manually, and configure the nodes to expose their hardware via"}],"source_content_type":"text/x-rst","patch_set":5,"id":"3a98d1f4_584dc598","line":88,"range":{"start_line":85,"start_character":0,"end_line":88,"end_character":70},"in_reply_to":"5a8bc5a2_23622517","updated":"2016-10-03 08:08:52.000000000","message":"Issue with things like FPGA is, that contrary to resources like CPU, memory, network or storage, it can be something else during the runtime. No one is expecting, that CPU will become something else, while there can be totally different IPs programmed on FPGA on demand.\n\nFurthermore, IPs can be divided into two groups. First one is the group where all of the accelerators are transparent for instances for example accelerators for networking functions. Second group are those accelerators, which have to be exposed to instances (using VF, passing device from /dev or other way) with appropriate software which should be used inside VM.\n\nProviding another service with full infrastructure might also be an option, although I\u0027d like to start with something simple, hence the library instead of service.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":1063,"name":"Ed Leafe","email":"ed@leafe.com","username":"ed-leafe"},"change_message_id":"fcd4e51b848f268e8ea22de00120e8a6a7e3af44","unresolved":false,"context_lines":[{"line_number":87,"context_line":"   introduced, which will be responsible for interaction with the FPGA, and"},{"line_number":88,"context_line":"   which will be a source of information about FPGA resources to Nova."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"2. We can do nothing, and leave the burden to the Operator, who will prepare"},{"line_number":91,"context_line":"   accelerators manually, and configure the nodes to expose their hardware via"},{"line_number":92,"context_line":"   PCI pass through."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"Data model impact"},{"line_number":95,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a629dbe_72c38389","line":92,"range":{"start_line":90,"start_character":3,"end_line":92,"end_character":20},"updated":"2016-11-10 16:47:42.000000000","message":"This is actually what was agreed upon at the Austin summit. There is no way that Nova should ever become an FPGA-management tool. Alternative #1 is a future possibility, but will require a great deal of operator burden to configure that library to distinguish the differences between, say, an FPGA that is a raw resource, and the various algorithms that are also resources. They can\u0027t be both, as that leads to race conditions: someone requests an FPGA, while someone else requests an algorithm that is supplied by the same FPGA.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"},{"author":{"_account_id":6804,"name":"bruce-benjamin","email":"bruce.benjamin@jhuapl.edu","username":"bruce-benjamin"},"change_message_id":"b00244bee6bb03e39838d71957b91741d195347d","unresolved":false,"context_lines":[{"line_number":87,"context_line":"   introduced, which will be responsible for interaction with the FPGA, and"},{"line_number":88,"context_line":"   which will be a source of information about FPGA resources to Nova."},{"line_number":89,"context_line":""},{"line_number":90,"context_line":"2. We can do nothing, and leave the burden to the Operator, who will prepare"},{"line_number":91,"context_line":"   accelerators manually, and configure the nodes to expose their hardware via"},{"line_number":92,"context_line":"   PCI pass through."},{"line_number":93,"context_line":""},{"line_number":94,"context_line":"Data model impact"},{"line_number":95,"context_line":"-----------------"}],"source_content_type":"text/x-rst","patch_set":5,"id":"9a629dbe_231fd15b","line":92,"range":{"start_line":90,"start_character":3,"end_line":92,"end_character":20},"in_reply_to":"9a629dbe_72c38389","updated":"2016-11-11 19:13:00.000000000","message":"+1. I agree that Nova shouldn\u0027t take on this capability of managing the handling of FPGAs. Even if Alternative 1 above is introduced, another issue introducing complexity is when multiple FPGA flash updates for a given feature need to be available.  This may be necessary, for example, if support for several versions of FPGA host boards is required that hosts the same FPGA type.  Unless the use of the FPGAs is scoped very narrowly, the complexity, options and versions that need to be supported will be unwieldy to manage in a service (Alternative 1), and it would be even worse to try and manage in Nova.","commit_id":"3c551dc6b14d29a9b1374338d0c0c1cbcbff8857"}]}
