that it is possible, on current Intel processors, to amplify access
rights quickly when calling into a specially configured
user-level library. Library instances in separate processes
can then share data safely, even in the face of independent
process failures. We have used protected libraries to implement
a new version of memcached in which client threads
execute the code of the server themselves, without the need
to send messages. Compared to the original, our new version
is both significantly simpler, containing 24% less code, and
dramatically faster, with a 11\u201356\u00d7 reduction in latency and
a roughly 2\u00d7 increase in throughput.<\/p>\n\n\n\n
![\"\"](\"https:\/\/www.hajim.rochester.edu\/senior-design-day\/wp-content\/uploads\/2020\/05\/our_model4.png\")
<\/figure>\n","protected":false},"excerpt":{"rendered":"Memcached is a widely used key-value (KV) store. It is structured
\nas a multithreaded user-level server, accessed over socket
\nconnections by a potentially distributed collection of clients.
\nBecause socket communication is so much more expensive
\nthan a single operation on a KV store, much of the client
\nlibrary is devoted to batching of requests. Batching is not
\nalways feasible, however, and the cost of communication
\nseems particularly unfortunate when\u2014as is often the case\u2014
\nclients are co-located on a single machine with the server,
\nand have access to the same physical memory.<\/p>\n
Fortunately, recent work on protected libraries has shown
\nthat it is possible, on current Intel processors, to amplify access
\nrights quickly when calling into a specially configured
\nuser-level library. Library instances in separate processes
\ncan then share data safely, even in the face of independent
\nprocess failures. We have used protected libraries to implement
\na new version of memcached in which client threads
\nexecute the code of the server themselves, without the need
\nto send messages. Compared to the original, our new version
\nis both significantly simpler, containing 24% less code, and
\ndramatically faster, with a 11\u201356\u00d7 reduction in latency and
\na roughly 2\u00d7 increase in throughput.
\n\ufeff<\/p>\n","protected":false},"author":6242,"featured_media":3756,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_coblocks_attr":"","_coblocks_dimensions":"","_coblocks_responsive_height":"","_coblocks_accordion_ie_support":"","_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[96],"tags":[],"coauthors":[8612],"class_list":["post-4546","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-csc-archive"],"acf":[],"yoast_head":"\n