incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Wayne <wav...@gmail.com>
Subject Re: Key Caching
Date Mon, 26 Jul 2010 20:46:06 GMT
If the cache is stored in the heap, how big can the heap be made
realistically on a 24gb ram machine? I am a java newbie but I have read
concerns with going over 8gb for the heap as the GC can be too painful/take
too long. I already have seen timeout issues (node is dead errors) under
load during GC or compaction. Can/should the heap be set to 16gb with 24gb
ram?

On Mon, Jul 26, 2010 at 12:49 PM, Peter Schuller <
peter.schuller@infidyne.com> wrote:

> (the following are best-effort answers resulting from taking the
> opportunity to look at the source; i may be missing something)
>
> > I am trying to use the key cache, and I had some basic questions as to
> how
> > it works. There seem to be settings based on % and total count, but not
> on
> > total cache size like most caching mechanisms. Is the cache stored in the
> > jvm heap or somewhere else?
>
> In the heap (I'm almost but not entirely sure). The key cache seems to
> be backed by a ConcurrentLinkedHashMap, which I think is the same one
> that eventually moved to another Java package and is over here:
>
> http://code.google.com/p/concurrentlinkedhashmap/source/browse/branches/lru/src/java/com/googlecode/concurrentlinkedhashmap/ConcurrentLinkedHashMap.java
>
> > What is the the key cache capacity in cfstats? How is this calculated? I
> > also see the stats get reset often and cache is cleared quite often, what
> > triggers the flushing of key cache? Is it compaction? Given that our keys
> do
>
> I'm not sure about the stats, but looking at SSTableTracker and
> SSTableReader I believe the cache entries are keyed on both row key
> and the path to the sstable in which the data resides. I didn't find
> any code indicating that cached keys are preserved (but modified) as a
> result of sstable replacement (as a result of compaction), so I
> *think* it is expected that cache hit ratio will drop as a result of
> compaction. Can anyone confirm/deny?
>
> --
> / Peter Schuller
>

Mime
View raw message