On Fri, 1 Feb 2002, Jukka Santala wrote: > On 19 Jul 2000 [EMAIL PROTECTED] wrote: > > Changed HASH_BITS from 14 to 8. This reduces the size of the > > cache from 128K to 2K. > I recall there's a paper about doing something with the Linux hash-tables > somewhere, but unfortunately I don't remember where and what ;)
Now that the network's actually working at this end a bit, I put the search-engine to good use, and looked up that reference: "Projects: Linux scalability: Linux kernel hash behaviour" http://www.citi.umich.edu/projects/linux-scalability/reports/hash.html ...except that specific "instrumentation patches" affect the matter being instrumented, due to cache misses, optimizer code re-location etc. so that you're better off using the Linux built-in profiling support if you decide to run tests. Also of interest, "Common hash table implementation in Linux kernel", http://www.uwsg.indiana.edu/hypermail/linux/kernel/0107.2/0256.html Note that if in addition to common API & code you re-use the same hash-tables, hashing several types of objects into one, you can save significant space at little cost, since different kinds of objects typically have different hash-distributions. Only thing you need to worry about is a mechanism to distinquish different types of objects in the table, which can be fairly efficiently done. Maybe I should look at doing this optimization myself at some point ;) -Jukka Santala -- To unsubscribe from this list, send a message to [EMAIL PROTECTED] with the command "unsubscribe linux-embedded" in the message body. For more information, see <http://waste.org/mail/linux-embedded>.
