Performance issue with initialSize parameter for SolrCache

[Simo Simov <si...@ontotext.com>]

SolrCache implementations LRUCache, FastLRUCache and LFUCache have a
parameter initialSize, that is passed directly to HashMap initialCapacity
constructor argument. 

For LFUCache this is the code line:

 

map = new LinkedHashMap<K,V>(initialSize, 0.75f, true) .

 

In solrconfig.xml I tried to set initialSize to be size/0.75 + some_more,
but it's impossible to set value greather than size because of this line:

 

final int initialSize = Math.min(str==null ? 1024 : Integer.parseInt(str),
limit);

 

For FastLRUCache there is no such limitation, but still I think
documentation and default values in solrconfig.xml should not use equal
values for initialSize and size.

Re: Performance issue with initialSize parameter for SolrCache

[Shawn Heisey <so...@elyograg.org>]

On 8/4/2014 9:08 AM, Simo Simov wrote:
>
> SolrCache implementations LRUCache, FastLRUCache and LFUCache have a
> parameter initialSize, that is passed directly to HashMap
> initialCapacity constructor argument.
>
> For LFUCache this is the code line:
>
> map = new LinkedHashMap<K,V>(initialSize, 0.75f, true) …
>
> In solrconfig.xml I tried to set initialSize to be size/0.75 +
> some_more, but it’s impossible to set value greather than size because
> of this line:
>
> final int initialSize = Math.min(str==null ? 1024 :
> Integer.parseInt(str), limit);
>
> For FastLRUCache there is no such limitation, but still I think
> documentation and default values in solrconfig.xml should not use
> equal values for initialSize and size.
>

Simo,

I wrote LFUCache, with FastLRUCache as a model. I'm having a hard time
grasping exactly what you've written above, which is probably my failing.

Based on what I think I understand, here's what I can say:

This is the line I can find in LFUCache.java that's similar (but not the
same) as the line you mentioned. The actual line you mentioned is in
LRUCache, not LFUCache:

final int initialSize = str == null ? limit : Integer.parseInt(str);

The initialSize value should never be greater than size, but that line
will not prevent you from setting it larger -- it only determines what
happens if the initialSize parameter is missing from the config, which
is to set it the same as size. I believe this is the correct action. The
similar line in LRUCache (the one that you quoted) could produce
unexpected behavior, because if initialSize is missing, it will be
limited to 1024, even if size is many times larger than that. The
workaround is to explicitly set initialSize, so I don't think it's
actually a bug.

Setting initialSize and size to the same value is a perfectly acceptable
and logical thing to do. You're telling Solr that it should allocate all
of the slots in the underlying Collection object up front, so it won't
be forced to dynamically expand it later -- an operation that could take
time.

On a related front, I actually would like to completely replace the
LFUCache implementation with a new one, because the current
implementation is a very naive and unoptimized way of doing LFU.I just
haven't found time to work on it.

See SOLR-2906 and SOLR-3393.

Thanks,
Shawn


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org