)]}'
{"specs/in_progress/global_ec_cluster.rst":[{"author":{"_account_id":7233,"name":"Matthew Oliver","email":"matt@oliver.net.au","username":"mattoliverau"},"change_message_id":"a8419a0b32c98bee7c98b3d4f8cea0f092319b55","unresolved":false,"context_lines":[{"line_number":59,"context_line":"performance falls down about 1/3 and other stuffs become a similar trend."},{"line_number":60,"context_line":"That\u0027s problem to build 2 regions EC cluster."},{"line_number":61,"context_line":""},{"line_number":62,"context_line":"1: http://ark.intel.com/ja/products/83356/Intel-Xeon-Processor-E5-2630-v3-20M-Cache-2_40-GHz"},{"line_number":63,"context_line":""},{"line_number":64,"context_line":"Proposed change"},{"line_number":65,"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":1,"id":"1a4dcd0f_a2ed1c27","line":62,"updated":"2015-08-09 13:26:23.000000000","message":"Wow, thats a real drop in encode performance.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":4608,"name":"Kota Tsuyuzaki","email":"bloodeagle40234@gmail.com","username":"tsuyuzaki-kota"},"change_message_id":"3012e83db195e7a9dc038b4adf211fe2bbe1bdfb","unresolved":false,"context_lines":[{"line_number":68,"context_line":"(copied) fragments instead of employing larger ec_m."},{"line_number":69,"context_line":""},{"line_number":70,"context_line":"For example, duplication_factor\u003d2, Swift will encode ec_k\u003d10, ec_m\u003d4 and store"},{"line_number":71,"context_line":"24 fragments (10x2 data fragments and 4x2 parity fragments) in Swift."},{"line_number":72,"context_line":""},{"line_number":73,"context_line":"That need to change PUT/GET and reconstruct sequence to map from the fragment"},{"line_number":74,"context_line":"index in Swift to actual fragment index for PyECLib but IMHO we don\u0027t need to"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_f30f8637","line":71,"updated":"2015-08-13 22:50:51.000000000","message":"This is 28.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":7233,"name":"Matthew Oliver","email":"matt@oliver.net.au","username":"mattoliverau"},"change_message_id":"a8419a0b32c98bee7c98b3d4f8cea0f092319b55","unresolved":false,"context_lines":[{"line_number":73,"context_line":"That need to change PUT/GET and reconstruct sequence to map from the fragment"},{"line_number":74,"context_line":"index in Swift to actual fragment index for PyECLib but IMHO we don\u0027t need to"},{"line_number":75,"context_line":"make an effort to build much modification for the converstaion among"},{"line_number":76,"context_line":"proxy-server \u003c-\u003e object-server \u003c-\u003e disks."},{"line_number":77,"context_line":""},{"line_number":78,"context_line":"I don\u0027t want describe the implementation in detail in the first patch in spec"},{"line_number":79,"context_line":"repository because it should be an idea to improve Swift. Need more discussion"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_42f21808","line":76,"updated":"2015-08-09 13:26:23.000000000","message":"What happens, in theory of course because we have no implementation details, when a region goes down. Does the reconstructor then double up (in the case of duplication_factor\u003d2) on the one region.\n\nOr worse, what if we have 3 regions and dulpication_factor\u003d3. Now a region\u0027s WAN link goes down, will it now push 2x the existing data to itself?\n\nI guess we need to make sure it has enough smarts in it to not duplicate to the same region or something.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":4608,"name":"Kota Tsuyuzaki","email":"bloodeagle40234@gmail.com","username":"tsuyuzaki-kota"},"change_message_id":"001ebec2ed1637440d5ba0aed163fac2f67182e1","unresolved":false,"context_lines":[{"line_number":73,"context_line":"That need to change PUT/GET and reconstruct sequence to map from the fragment"},{"line_number":74,"context_line":"index in Swift to actual fragment index for PyECLib but IMHO we don\u0027t need to"},{"line_number":75,"context_line":"make an effort to build much modification for the converstaion among"},{"line_number":76,"context_line":"proxy-server \u003c-\u003e object-server \u003c-\u003e disks."},{"line_number":77,"context_line":""},{"line_number":78,"context_line":"I don\u0027t want describe the implementation in detail in the first patch in spec"},{"line_number":79,"context_line":"repository because it should be an idea to improve Swift. Need more discussion"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_7166c9bc","line":76,"in_reply_to":"1a4dcd0f_42f21808","updated":"2015-08-16 02:58:55.000000000","message":"Nice thought, that\u0027s why I added an item \"Ring placement strategy\" to Implementation section.\n\nExactly, if there are almost doubled fragments in one side region and the other region is down, it would be scared situation we can not rebuild original objects as well as smaller parity fragments case.\n\ne.g:\n duplication_factor \u003d 2, k \u003d 4, m\u003d2\n\n1st Region:\n [0, 1, 2, 6, 7, 8]\n\n2nd Region:\n [3, 4, 5, 9, 10, 11]\n\n(actual indices to rebuild mapped as index // (k+m))\n\nIn this case, 1st region has only fragments consist of fragment index 0, 1, 2, and 2nd has only 3, 4, 5. The worst case like this will cause significant data loss as well as no duplication factor. i.e. In fact, data durability will be\n\n \"no duplication\" \u003c \"with duplication\" \u003c \"more unique parities\"\n\nIn the future work, IMO, we can take a way to control the fragment index tied up to the region, like \"1st subset should be in 1st Region and 2nd subset should be ...\" but it could be beyond of this spec.\n\nOn the 3 region case, you mean...about reconstuctor work? Basically all of jobs in the backend process like reconstructor   should be same as current behavior. \n\ni.e.:\nif existing data already stored enough over regions and WAN link goes doewn, it would be assumed as just NW failure so reconstructor won\u0027t make more fragments in local.\n\nif PUT request coming in while WAN link failing, proxy will make a duplicated copies according to duplication factor in the local region.\n\nDoes it make you sense?","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":4608,"name":"Kota Tsuyuzaki","email":"bloodeagle40234@gmail.com","username":"tsuyuzaki-kota"},"change_message_id":"3012e83db195e7a9dc038b4adf211fe2bbe1bdfb","unresolved":false,"context_lines":[{"line_number":102,"context_line":"^^^^^^^^^^^^^^"},{"line_number":103,"context_line":"Pros:"},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"- Simple and able to reuse existing stuffs."},{"line_number":106,"context_line":""},{"line_number":107,"context_line":"Cons:"},{"line_number":108,"context_line":""}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_f61d3463","line":105,"updated":"2015-08-13 22:50:51.000000000","message":"Add an item to Pros. Container-Sync can send just an object to sync.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":15932,"name":"Bill Huber","email":"wbhuber@us.ibm.com","username":"wbhuber"},"change_message_id":"398c563a48646bd90891b9183c76056f3fbfde6f","unresolved":false,"context_lines":[{"line_number":103,"context_line":"Pros:"},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"- Simple and able to reuse existing stuffs."},{"line_number":106,"context_line":""},{"line_number":107,"context_line":"Cons:"},{"line_number":108,"context_line":""},{"line_number":109,"context_line":"- Need to re-decode/encode to transfer objects to another region"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_fd916cf2","line":106,"updated":"2015-08-14 15:47:22.000000000","message":"Based on Alistair\u0027s comment yesterday, it takes less data for Container Sync to synchronize between containers than that of EC taking fragments from one region to other region.  This is one pro that Container Sync has (can be looked as a con for EC).\n\n@acoles, correct me if I \"heard\" you wrong.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":4608,"name":"Kota Tsuyuzaki","email":"bloodeagle40234@gmail.com","username":"tsuyuzaki-kota"},"change_message_id":"001ebec2ed1637440d5ba0aed163fac2f67182e1","unresolved":false,"context_lines":[{"line_number":103,"context_line":"Pros:"},{"line_number":104,"context_line":""},{"line_number":105,"context_line":"- Simple and able to reuse existing stuffs."},{"line_number":106,"context_line":""},{"line_number":107,"context_line":"Cons:"},{"line_number":108,"context_line":""},{"line_number":109,"context_line":"- Need to re-decode/encode to transfer objects to another region"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_71b3294e","line":106,"in_reply_to":"1a4dcd0f_fd916cf2","updated":"2015-08-16 02:58:55.000000000","message":"You\u0027re right. When using proposed change Swift will send \u003e\u003d1 data across regions but container-sync should send just 1 copy to the another region proxy. That is why the container-sync has one more pros that less data transfer than proposed change as you write down above.\n\nWill fix.","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"},{"author":{"_account_id":4608,"name":"Kota Tsuyuzaki","email":"bloodeagle40234@gmail.com","username":"tsuyuzaki-kota"},"change_message_id":"001ebec2ed1637440d5ba0aed163fac2f67182e1","unresolved":false,"context_lines":[{"line_number":116,"context_line":""},{"line_number":117,"context_line":"- Proxy-Server PUT/GET path"},{"line_number":118,"context_line":"- Object-Reconstructor"},{"line_number":119,"context_line":"- (Optional) Ring placement strutegy"},{"line_number":120,"context_line":""},{"line_number":121,"context_line":""},{"line_number":122,"context_line":"Assignee(s)"}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a4dcd0f_1181854d","line":119,"updated":"2015-08-16 02:58:55.000000000","message":"strategy","commit_id":"baada50c4022f9e40cfe6f0544b38e35dbf75969"}]}