hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ben Manes (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-15560) TinyLFU-based BlockCache
Date Mon, 07 Nov 2016 06:23:58 GMT

    [ https://issues.apache.org/jira/browse/HBASE-15560?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15643265#comment-15643265
] 

Ben Manes commented on HBASE-15560:
-----------------------------------

{{quote}
Would you want the same dataset loaded too?
{{quote}}

That can't hurt, so unless its more work might as well.

---

In my [simulator|https://github.com/ben-manes/caffeine/wiki/Simulator], I tried to emulate
{{workload c}} using the following configuration,
 * maximum-size = (below)
 * source = "synthetic"
 * distribution = "zipfian"
 * zipfian.items = 1000

I then ran it with small caches to emulate your observation. {{LruBlockCache}} is an SLru
variant, so I'm assuming it behaves similar to the theoretical version.

||Policy||max=5||max=10||max=25||
|Lru|13.10%|20.70%|35.60%|
|SLru|25.90%|29.30|45.00%|
|Caffeine|24.40%|32.30%|46.00%|
|Optimal|35.20%|42.10%|45.50%|

We see that at the smallest size, 5, Caffeine slightly under performs. However whether its
slightly lower, equal, or higher varies on the run. This is due to the distribution generation
and Caffeine's hashing having randomness, so across runs we see it pretty much on par. As
the size increases we see them all stay pretty close. Since SLru is known to be optimal for
Zipf, this at least is a good sign but does not explain your observations.

> TinyLFU-based BlockCache
> ------------------------
>
>                 Key: HBASE-15560
>                 URL: https://issues.apache.org/jira/browse/HBASE-15560
>             Project: HBase
>          Issue Type: Improvement
>          Components: BlockCache
>    Affects Versions: 2.0.0
>            Reporter: Ben Manes
>            Assignee: Ben Manes
>         Attachments: HBASE-15560.patch, HBASE-15560.patch, HBASE-15560.patch, HBASE-15560.patch,
HBASE-15560.patch, HBASE-15560.patch, HBASE-15560.patch, bc.hit.count, bc.miss.count, branch-1.tinylfu.txt,
gets, tinylfu.patch
>
>
> LruBlockCache uses the Segmented LRU (SLRU) policy to capture frequency and recency of
the working set. It achieves concurrency by using an O( n ) background thread to prioritize
the entries and evict. Accessing an entry is O(1) by a hash table lookup, recording its logical
access time, and setting a frequency flag. A write is performed in O(1) time by updating the
hash table and triggering an async eviction thread. This provides ideal concurrency and minimizes
the latencies by penalizing the thread instead of the caller. However the policy does not
age the frequencies and may not be resilient to various workload patterns.
> W-TinyLFU ([research paper|http://arxiv.org/pdf/1512.00727.pdf]) records the frequency
in a counting sketch, ages periodically by halving the counters, and orders entries by SLRU.
An entry is discarded by comparing the frequency of the new arrival (candidate) to the SLRU's
victim, and keeping the one with the highest frequency. This allows the operations to be performed
in O(1) time and, though the use of a compact sketch, a much larger history is retained beyond
the current working set. In a variety of real world traces the policy had [near optimal hit
rates|https://github.com/ben-manes/caffeine/wiki/Efficiency].
> Concurrency is achieved by buffering and replaying the operations, similar to a write-ahead
log. A read is recorded into a striped ring buffer and writes to a queue. The operations are
applied in batches under a try-lock by an asynchronous thread, thereby track the usage pattern
without incurring high latencies ([benchmarks|https://github.com/ben-manes/caffeine/wiki/Benchmarks#server-class]).
> In YCSB benchmarks the results were inconclusive. For a large cache (99% hit rates) the
two caches have near identical throughput and latencies with LruBlockCache narrowly winning.
At medium and small caches, TinyLFU had a 1-4% hit rate improvement and therefore lower latencies.
The lack luster result is because a synthetic Zipfian distribution is used, which SLRU performs
optimally. In a more varied, real-world workload we'd expect to see improvements by being
able to make smarter predictions.
> The provided patch implements BlockCache using the [Caffeine|https://github.com/ben-manes/caffeine]
caching library (see HighScalability [article|http://highscalability.com/blog/2016/1/25/design-of-a-modern-cache.html]).
> Edward Bortnikov and Eshcar Hillel have graciously provided guidance for evaluating this
patch ([github branch|https://github.com/ben-manes/hbase/tree/tinylfu]).



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message