penberg / sphinx Goto Github PK

Sphinx needs to support explicit timestamp-based object expiration and LRU policy, similar to Memcached. However, since we use a log-structured memory allocator, it might be difficult to provide LRU. Instead, we can see if we can approximate LRU with the approach MICA is taking, for example:

https://www.usenix.org/node/179748

Memcached parser accepts messages that terminate with just a linefeed character (default for nc):

$ nc 127.0.0.1 11211
get foo
END

Sphinx, however, replies with an error result:

$ nc 127.0.0.1 11211
get foo
ERROR

Everything works as expected if I pass the -C command line option to nc to make it terminate lines with CR/LF characters.

Memcached:

$ nc -C 127.0.0.1 11211
get foo
END

Sphinx:

$ nc -C 127.0.0.1 11211
get foo
END

Although Sphinx is implemented as per the Memcached protocol specification, it's worth fixing this (minor) compatibility issue.

If the amount of memory configured is not divisible by the number of threads, the sphinx program crashes under load.

Memcached has UDP protocol support, which is missing from Sphinx.

Hi, I found this this publish on Glommio docs.

To better understand the impact of thread-per-core I'd like to reproduce the experiment in the paper. Could you provide some information about how to reproduce?