Thanks for that advise; I trust your judgement!

Do you have an opinion on copying IntHashSet directly to Lucene-core or
some other approach?

On 1/30/13 3:59 PM, "Dawid Weiss" <[email protected]> wrote:

>> IntHashSet and SentinalIntSet.  However it computes a hash of the id
>>using
>> MurMurHash3, which I think is overkill, and the API doesn't have a
>> subclassing opportunity for me to use something else.
>
>I assure you this is not an overkill. The keys need to be hashed,
>otherwise you'll hit pretty bad conflicting chains on real-life data.
>We had that in the past and I know for sure. Murmur hash is pretty
>darn fast on its own, even if you remove it though, I don't think
>you'll gain much -- memory latency is typically the blocking factor,
>not hashing which is performed on the CPU or local hot cache.
>
>Dawid
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: [email protected]
>For additional commands, e-mail: [email protected]
>


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to