)]}'
{"specs/backlog/approved/normalized-compute-units.rst":[{"author":{"_account_id":9775,"name":"Sudipta Biswas","email":"sbiswas7@in.ibm.com","username":"sudswas"},"change_message_id":"7d195ce9f07cc4cabf24bbb1c9e6aa5ac8b7d91d","unresolved":false,"context_lines":[{"line_number":88,"context_line":"  that will be relevant in Operator\u0027s use case."},{"line_number":89,"context_line":"* At the start of nova-compute a new process will benchmark the platform and"},{"line_number":90,"context_line":"  record NCU value in DB. This will be used as NCU total.  A new resource"},{"line_number":91,"context_line":"  classes will be created for compute_capacity resource."},{"line_number":92,"context_line":"* On the flavor level a new option will be added to allow specifying"},{"line_number":93,"context_line":"  compute_capacity value, unit being NCU."},{"line_number":94,"context_line":"* compute_capacity value will be passed to newly created NCU filter. At least"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba15a1d1_0fb36799","line":91,"updated":"2015-09-16 18:53:00.000000000","message":"If we start the process only when nova-compute is started - it might as well delay the registration of the compute node? Can you explain a bit more on how this would be achieved?\n\nAlso isn\u0027t it possible that the FP operations or the Intructions per second are lesser in a heavily loaded system as opposed to a system running nothing? In such a case, will this case suffice where we just calculate a NCU at the start of nova-compute?","commit_id":"58ee796fd01023259f657e71c8cfd35cbc48696b"},{"author":{"_account_id":9775,"name":"Sudipta Biswas","email":"sbiswas7@in.ibm.com","username":"sudswas"},"change_message_id":"7d195ce9f07cc4cabf24bbb1c9e6aa5ac8b7d91d","unresolved":false,"context_lines":[{"line_number":93,"context_line":"  compute_capacity value, unit being NCU."},{"line_number":94,"context_line":"* compute_capacity value will be passed to newly created NCU filter. At least"},{"line_number":95,"context_line":"  one socket in host machine must offer the same or more NCUs than requested by"},{"line_number":96,"context_line":"  flavor to pass filtering, as all vCPUs of a VM will use cores from one"},{"line_number":97,"context_line":"  socket."},{"line_number":98,"context_line":"* The compute_capacity value will be passed to nova-compute along with other"},{"line_number":99,"context_line":"  parameters in request spec."}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba15a1d1_cf91eff2","line":96,"updated":"2015-09-16 18:53:00.000000000","message":"This sounds like you\u0027d be filtering the NUMA nodes, much like the NUMA Topology filter does today?\n\nThis probably means another BP. A bit too much for the same one.","commit_id":"58ee796fd01023259f657e71c8cfd35cbc48696b"},{"author":{"_account_id":9775,"name":"Sudipta Biswas","email":"sbiswas7@in.ibm.com","username":"sudswas"},"change_message_id":"7d195ce9f07cc4cabf24bbb1c9e6aa5ac8b7d91d","unresolved":false,"context_lines":[{"line_number":101,"context_line":"  compute_capacity value. For example, 2 NCUs \u003d 2048 shares."},{"line_number":102,"context_line":"* Cgroup shares value will be used in hypervisor (libvirt) to create cgroup for"},{"line_number":103,"context_line":"  a VM."},{"line_number":104,"context_line":"* VMs without defined NCU will be assigned default NCU (configurable item)."},{"line_number":105,"context_line":""},{"line_number":106,"context_line":"The future work, not addressed in this spec:"},{"line_number":107,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba15a1d1_81074b17","line":104,"updated":"2015-09-16 18:53:00.000000000","message":"This is interesting. If we start using NCU, i guess we still don\u0027t mean we want to do away with the idea of vCPUs in the first iteration? I guess you mean systems that cannot even calculate NCUs because of lack of benchmarking tools (or any other reason)?","commit_id":"58ee796fd01023259f657e71c8cfd35cbc48696b"},{"author":{"_account_id":9775,"name":"Sudipta Biswas","email":"sbiswas7@in.ibm.com","username":"sudswas"},"change_message_id":"7d195ce9f07cc4cabf24bbb1c9e6aa5ac8b7d91d","unresolved":false,"context_lines":[{"line_number":107,"context_line":""},{"line_number":108,"context_line":"* Allowing to set a ceiling for NCU for VM to provide assurance that"},{"line_number":109,"context_line":"  constrained VM will not take more CPU than contracted for."},{"line_number":110,"context_line":"* Switching from VCPU to NCU model entirely."},{"line_number":111,"context_line":""},{"line_number":112,"context_line":"Alternatives"},{"line_number":113,"context_line":"------------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba15a1d1_01b81b29","line":110,"updated":"2015-09-16 18:53:00.000000000","message":"Is it applicable always? There can be cases where you\u0027d want your VMs to float/not get throttled based on shares etc.","commit_id":"58ee796fd01023259f657e71c8cfd35cbc48696b"},{"author":{"_account_id":9775,"name":"Sudipta Biswas","email":"sbiswas7@in.ibm.com","username":"sudswas"},"change_message_id":"7d195ce9f07cc4cabf24bbb1c9e6aa5ac8b7d91d","unresolved":false,"context_lines":[{"line_number":196,"context_line":"Assignee(s)"},{"line_number":197,"context_line":"-----------"},{"line_number":198,"context_line":""},{"line_number":199,"context_line":"To be defined."},{"line_number":200,"context_line":""},{"line_number":201,"context_line":"Work Items"},{"line_number":202,"context_line":"----------"}],"source_content_type":"text/x-rst","patch_set":1,"id":"ba15a1d1_a1292f39","line":199,"updated":"2015-09-16 18:53:00.000000000","message":"I am interested in collaborating on this one, if needed.\nBut this sounds like HUGE bit of work to me, guess the approach could be to even break down the BP further?","commit_id":"58ee796fd01023259f657e71c8cfd35cbc48696b"}]}