[
https://issues.apache.org/jira/browse/CASSANDRA-13530?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16166091#comment-16166091
]
Yuji Ito commented on CASSANDRA-13530:
--------------------------------------
I drew a diagram to help you understand.
!groupAndBatch.png|thumbnail!
The left one show the behavior of GroupCommitLog.
maybeWaitForSync() of GroupCommitLogService waits for sync.
The sync is executed periodically (commitlog_sync_group_window_in_ms).
The wait time appears as overhead of latency when the concurrency is low.
If other requests come, they also keep waiting for sync. But their commitlog
will be written to the disk at the same sync.
When the sync of commitlog is completed, maybeWaitForSync() returns.
On the other hand, maybeWaitForSync() of BatchCommitLogService (the right one
of the diagram) releases the semaphore of the commitlog sync thread immediately.
So, commitlog_sync_batch_window_in_ms isn't effective. And, if other requests
come, their commitlog might not be written at the same sync.
> GroupCommitLogService
> ---------------------
>
> Key: CASSANDRA-13530
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13530
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Yuji Ito
> Assignee: Yuji Ito
> Fix For: 2.2.x, 3.0.x, 3.11.x
>
> Attachments: groupAndBatch.png, groupCommit22.patch,
> groupCommit30.patch, groupCommit3x.patch,
> groupCommitLog_noSerial_result.xlsx, groupCommitLog_result.xlsx,
> GuavaRequestThread.java, MicroRequestThread.java
>
>
> I propose a new CommitLogService, GroupCommitLogService, to improve the
> throughput when lots of requests are received.
> It improved the throughput by maximum 94%.
> I'd like to discuss about this CommitLogService.
> Currently, we can select either 2 CommitLog services; Periodic and Batch.
> In Periodic, we might lose some commit log which hasn't written to the disk.
> In Batch, we can write commit log to the disk every time. The size of commit
> log to write is too small (< 4KB). When high concurrency, these writes are
> gathered and persisted to the disk at once. But, when insufficient
> concurrency, many small writes are issued and the performance decreases due
> to the latency of the disk. Even if you use SSD, processes of many IO
> commands decrease the performance.
> GroupCommitLogService writes some commitlog to the disk at once.
> The patch adds GroupCommitLogService (It is enabled by setting
> `commitlog_sync` and `commitlog_sync_group_window_in_ms` in cassandra.yaml).
> The difference from Batch is just only waiting for the semaphore.
> By waiting for the semaphore, some writes for commit logs are executed at the
> same time.
> In GroupCommitLogService, the latency becomes worse if the there is no
> concurrency.
> I measured the performance with my microbench (MicroRequestThread.java) by
> increasing the number of threads.The cluster has 3 nodes (Replication factor:
> 3). Each nodes is AWS EC2 m4.large instance + 200IOPS io1 volume.
> The result is as below. The GroupCommitLogService with 10ms window improved
> update with Paxos by 94% and improved select with Paxos by 76%.
> h6. SELECT / sec
> ||\# of threads||Batch 2ms||Group 10ms||
> |1|192|103|
> |2|163|212|
> |4|264|416|
> |8|454|800|
> |16|744|1311|
> |32|1151|1481|
> |64|1767|1844|
> |128|2949|3011|
> |256|4723|5000|
> h6. UPDATE / sec
> ||\# of threads||Batch 2ms||Group 10ms||
> |1|45|26|
> |2|39|51|
> |4|58|102|
> |8|102|198|
> |16|167|213|
> |32|289|295|
> |64|544|548|
> |128|1046|1058|
> |256|2020|2061|
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
