> Right, much less GC if app frequently reopens.  But a 32X increase in
> RAM usage is not trivial; I think we shouldn't enable it by default?

Right, the RAM usage is quite high!  Is there a more compact
representation we could use?  Ah well, either way for good RT
performance, there are some users who may want to use this option.

> Have you tested?

The test would be a basic benchmark of queries against BV vs. an int[]
of deletes?

On Tue, Jul 20, 2010 at 12:17 PM, Michael McCandless
<[email protected]> wrote:
> On Tue, Jul 20, 2010 at 1:44 PM, Jason Rutherglen
> <[email protected]> wrote:
>>> Breaking up RT patches into baby steps would be great :)
>>> Actually is the RT branch active (I haven't seen commits going
>>> in).
>>
>> From what we discussed, my impression is that the RT changes
>> will be substantial and the sequence ids seem to be something
>> that can be implemented now in trunk, then at least a small
>> piece of RT would be implemented and tested. A small isolated
>> improvement.
>
> I agree.
>
>>> The biggest downside of sequence IDs is increase RAM usage
>>> right?
>>
>> Yes, however the garbage collection would decrease.
>
> Right, much less GC if app frequently reopens.  But a 32X increase in
> RAM usage is not trivial; I think we shouldn't enable it by default?
>
>> We should make seq id deletes pluggable.
>
> +1 -- make the deletes impl pluggable (ie one impl is sequence IDs;
> another is BitVector).
>
>>> Then, checking if a doc is deleted becomes an int compare
>>> instead of a bit lookup, right?
>>
>> Right it'd be an int compare, I think this'd be ok?
>
> Yeah maybe even more than OK; could be this is faster than looking up
> a bit!  (Though it's much more traffic on the memory bus).  Apps may
> want to use this even outside of NRT.  Have you tested?
>
> Mike
>
> ---------------------------------------------------------------------
> 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