[
https://issues.apache.org/jira/browse/HDFS-15202?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17058076#comment-17058076
]
Danil Lipovoy commented on HDFS-15202:
--------------------------------------
[~leosun08]
Thanks for interest!)
Let me provide more details.
I added some logging:
{code:java}
private BlockReader getBlockReaderLocal() throws InvalidToken {
...
* LOG.info("SSC blockId: " + block.getBlockId());*
ShortCircuitCache cache =
clientContext.getShortCircuitCache(block.getBlockId());
{code}
And run reading test via HBase. We can see the log output:
cat hbase-cmf-hbase-REGIONSERVER-wx1122-02.ru.log.out |grep "SSC blockId"
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256403
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251835
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256236
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251488
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256526
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110252104
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256969
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256965
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256382
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251751
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256871
2020-03-12 18:47:32,447 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251769
2020-03-12 18:47:32,448 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256825
2020-03-12 18:47:32,448 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256241
2020-03-12 18:47:32,448 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256548
2020-03-12 18:47:32,449 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251488
2020-03-12 18:47:32,449 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256808
2020-03-12 18:47:32,449 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110252097
2020-03-12 18:47:32,449 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110257035
2020-03-12 18:47:32,450 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256403
2020-03-12 18:47:32,450 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256779
2020-03-12 18:47:32,450 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256331
2020-03-12 18:47:32,451 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251691
2020-03-12 18:47:32,451 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251521
2020-03-12 18:47:32,451 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251835
2020-03-12 18:47:32,452 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251691
2020-03-12 18:47:32,455 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251769
We can see that last digit is evenly distributed. There are:
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256403 -> 3
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110251835 -> 5
2020-03-12 18:47:32,446 INFO org.apache.hadoop.hdfs.BlockReaderFactory: SSC
blockId: 1110256236 -> 6
Then I collected some statistics:
cat hbase-cmf-hbase-REGIONSERVER-hw2288-32.cloud.dev.df.sbrf.ru.log.out.2 |grep
"SSC blockId"|awk '{print substr($0,length,1)}' >> ~/ids.txt
cat ~/ids.txt | sort | uniq -c | sort -nr | awk '{printf "%-8s%s\n", $2,
$1}'|sort
0 645813
1 559617
2 532624
3 551147
4 484945
5 465928
6 570635
7 473285
8 525565
9 447981
It means the logs the last digit 0 found 645813 times
Digit 1 - 559617 times
etc.
Quite evenly.
When we divide blockId by modulus the blocks will cached evenly too. For
example if clientShortCircuitNum = 3, then last digts will go to:
blockId *0 -> shortCircuitCaches[0]
blockId *1 -> shortCircuitCaches[1]
blockId *2 -> shortCircuitCaches[2]
blockId *3 -> shortCircuitCaches[0]
blockId *4 -> shortCircuitCaches[1]
blockId *5 -> shortCircuitCaches[2]
blockId *6 -> shortCircuitCaches[0]
blockId *7 -> shortCircuitCaches[1]
blockId *8 -> shortCircuitCaches[2]
blockId *9 -> shortCircuitCaches[0]
There a little bit more [0] then [1] or [2], but not too much and works good)
> HDFS-client: boost ShortCircuit Cache
> -------------------------------------
>
> Key: HDFS-15202
> URL: https://issues.apache.org/jira/browse/HDFS-15202
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: dfsclient
> Environment: 4 nodes E5-2698 v4 @ 2.20GHz, 700 Gb Mem.
> 8 RegionServers (2 by host)
> 8 tables by 64 regions by 1.88 Gb data in each = 900 Gb total
> Random read in 800 threads via YCSB and a little bit updates (10% of reads)
> Reporter: Danil Lipovoy
> Assignee: Danil Lipovoy
> Priority: Minor
> Attachments: HDFS_CPU_full_cycle.png, cpu_SSC.png, cpu_SSC2.png,
> hdfs_cpu.png, hdfs_reads.png, hdfs_scc_3_test.png,
> hdfs_scc_test_full-cycle.png, locks.png, requests_SSC.png
>
>
> ТотI want to propose how to improve reading performance HDFS-client. The
> idea: create few instances ShortCircuit caches instead of one.
> The key points:
> 1. Create array of caches (set by
> clientShortCircuitNum=*dfs.client.short.circuit.num*, see in the pull
> requests below):
> {code:java}
> private ClientContext(String name, DfsClientConf conf, Configuration config) {
> ...
> shortCircuitCache = new ShortCircuitCache[this.clientShortCircuitNum];
> for (int i = 0; i < this.clientShortCircuitNum; i++) {
> this.shortCircuitCache[i] = ShortCircuitCache.fromConf(scConf);
> }
> {code}
> 2 Then divide blocks by caches:
> {code:java}
> public ShortCircuitCache getShortCircuitCache(long idx) {
> return shortCircuitCache[(int) (idx % clientShortCircuitNum)];
> }
> {code}
> 3. And how to call it:
> {code:java}
> ShortCircuitCache cache =
> clientContext.getShortCircuitCache(block.getBlockId());
> {code}
> The last number of offset evenly distributed from 0 to 9 - that's why all
> caches will full approximately the same.
> It is good for performance. Below the attachment, it is load test reading
> HDFS via HBase where clientShortCircuitNum = 1 vs 3. We can see that
> performance grows ~30%, CPU usage about +15%.
> Hope it is interesting for someone.
> Ready to explain some unobvious things.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
