[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16286620#comment-16286620
]
stack commented on HBASE-16890:
---
Ok. HBASE-19371 has been resolved as fixed. async WAL works w/ 2.8.x and
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16286480#comment-16286480
]
stack commented on HBASE-16890:
---
128 threads. Did two runs of first and last just to be sure of what I was
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16285351#comment-16285351
]
stack commented on HBASE-16890:
---
12 ycsb threads
||workload||oldFSHLog||async||async+own_eventloop|
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16285168#comment-16285168
]
Chance Li commented on HBASE-16890:
---
Hi Stack, do 24 thread(or 48 threads) mean "threadcount=24" on one
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16285065#comment-16285065
]
stack commented on HBASE-16890:
---
24 threads
||workload||OldFSHLog||asyncfs||acyncfs+own_eventloop||
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16284760#comment-16284760
]
stack commented on HBASE-16890:
---
Here is another run w/ 48 YCSB clients on one machine and then a RS on
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16284373#comment-16284373
]
stack commented on HBASE-16890:
---
That was better.
||workload||current||asyncfs||
|loading|5334 ops/s|7989
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16283130#comment-16283130
]
stack commented on HBASE-16890:
---
No. I didn't have that one. Thanks. Retry.
> Analyze the performance of
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16283109#comment-16283109
]
Duo Zhang commented on HBASE-16890:
---
Does your build have this commit in place?
{quote}
commit
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16283106#comment-16283106
]
stack commented on HBASE-16890:
---
Trying your suggestion [~Apache9] but seems like ycsb hangs. Trying to
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16283020#comment-16283020
]
Yu Li commented on HBASE-16890:
---
I could see some good offline discussion and more testing and maybe we
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16281547#comment-16281547
]
Duo Zhang commented on HBASE-16890:
---
[~stack] There is a config
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16281305#comment-16281305
]
stack commented on HBASE-16890:
---
Comparing one client with a few threads (6) against a remote, single
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16281075#comment-16281075
]
stack commented on HBASE-16890:
---
Let me try w/ YCSB too as per [~carp84] ask. Any other asks?
> Analyze
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16281039#comment-16281039
]
stack commented on HBASE-16890:
---
I ran a compare using WALPE. Setup was multi-DNs. asyncfs does better
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16264067#comment-16264067
]
Duo Zhang commented on HBASE-16890:
---
[~ram_krish] Yes this is expected. Can you please try multiwal?
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16264058#comment-16264058
]
ramkrishna.s.vasudevan commented on HBASE-16890:
The command that I used now is
{code}
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16264057#comment-16264057
]
ramkrishna.s.vasudevan commented on HBASE-16890:
[~chancelq]
The cluster has one DN and 1
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263942#comment-16263942
]
Chance Li commented on HBASE-16890:
---
[~ramkrishna]
bq. One quesiton on the YCSB - so since you measure
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263871#comment-16263871
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I tried this out (still on a single
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263825#comment-16263825
]
ramkrishna.s.vasudevan commented on HBASE-16890:
One quesiton on the YCSB - so since you
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263797#comment-16263797
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I think I need to repeat my test here. I
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263738#comment-16263738
]
Duo Zhang commented on HBASE-16890:
---
The ops metrics are 81435 vs. 77108, FSHLog is 5% more but the run
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263728#comment-16263728
]
Anoop Sam John commented on HBASE-16890:
So as per ur tests the higher percentile latency values
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263521#comment-16263521
]
Duo Zhang commented on HBASE-16890:
---
BTW what do you mean by ‘completely asynchronous’?
> Analyze the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263513#comment-16263513
]
Duo Zhang commented on HBASE-16890:
---
MVCC is assigned before calling consumer. It is an optimization
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16263280#comment-16263280
]
Chance Li commented on HBASE-16890:
---
Thanks, sir. [~Apache9]
The test result is not good enough but I
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16259995#comment-16259995
]
Duo Zhang commented on HBASE-16890:
---
It is designed to be single threaded and we can use multi WAL to
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16259663#comment-16259663
]
Chance Li commented on HBASE-16890:
---
It's very nice code. :) I will work on this.
And I read the code
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16259178#comment-16259178
]
Duo Zhang commented on HBASE-16890:
---
IIRC the problem here is that, on a single node HBase & HDFS,
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16258963#comment-16258963
]
ramkrishna.s.vasudevan commented on HBASE-16890:
bq.No. But I like the numbers you are
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16258949#comment-16258949
]
Yu Li commented on HBASE-16890:
---
Personally I think YCSB result is more persuasive. Could we help get some
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16258838#comment-16258838
]
stack commented on HBASE-16890:
---
bq. Is this test enough?
No. But I like the numbers you are getting. Can
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16258363#comment-16258363
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I got those numbers. Just loaded 25G
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16258336#comment-16258336
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Sorry about the numbers. I don't have
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16257946#comment-16257946
]
Duo Zhang commented on HBASE-16890:
---
So what can we do to close this issue? More tests with different
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16256131#comment-16256131
]
stack commented on HBASE-16890:
---
Whats the difference [~ram_krish]? Did anything change? What numbers were
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16237230#comment-16237230
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Seems in alpha-4 also the performance
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16235228#comment-16235228
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Will test once again with latest alpha-4
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16233661#comment-16233661
]
Anoop Sam John commented on HBASE-16890:
Lets work on this after Alpha-4. This is a very imp
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16227431#comment-16227431
]
Mike Drob commented on HBASE-16890:
---
[~stack] this hasn't seen any activity since Mar - do you still
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15938072#comment-15938072
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Ya sure. Let us check this now.
>
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15937726#comment-15937726
]
Duo Zhang commented on HBASE-16890:
---
{quote}
So can increase the number of writers in AsynWAL also?
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15937704#comment-15937704
]
Anoop Sam John commented on HBASE-16890:
bq. I think the problem is that AsyncFSWAL is single
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15937594#comment-15937594
]
Duo Zhang commented on HBASE-16890:
---
[~stack] [~ram_krish] Let's pick this up again?
The last
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15656848#comment-15656848
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Sorry. Even FSHLog does not call
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15656770#comment-15656770
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I want to ask some dumb questions.
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15656724#comment-15656724
]
ramkrishna.s.vasudevan commented on HBASE-16890:
bq.ut the actual syncs on DFSOutput are
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15656718#comment-15656718
]
Duo Zhang commented on HBASE-16890:
---
{quote}
we can assume that the number of sync futures that gets
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15656705#comment-15656705
]
ramkrishna.s.vasudevan commented on HBASE-16890:
With checksum disabled and with 10G of
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15650734#comment-15650734
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I just ran a PE with 50 threads but
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15649510#comment-15649510
]
Duo Zhang commented on HBASE-16890:
---
It is the actual perf. AsyncFSWAL can finish WALPE with less time.
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15648095#comment-15648095
]
ramkrishna.s.vasudevan commented on HBASE-16890:
But you still get more ops/sec. So which
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15647590#comment-15647590
]
Duo Zhang commented on HBASE-16890:
---
AsyncFSWAL syncs more even after I introduced a
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15644696#comment-15644696
]
stack commented on HBASE-16890:
---
Let me compare packet sizes.
> Analyze the performance of AsyncWAL and
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643323#comment-15643323
]
Duo Zhang commented on HBASE-16890:
---
With 3 DNs then AsyncFSWAL should be faster as we use fan out. And
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643271#comment-15643271
]
Duo Zhang commented on HBASE-16890:
---
We both want to reduce the contention by introducing a ringbuffer,
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643251#comment-15643251
]
stack commented on HBASE-16890:
---
Or, why are we twice as slow? Its reasonable to think that asyncwal should
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643250#comment-15643250
]
stack commented on HBASE-16890:
---
I've not looked... are the two approaches different? Can we unify?
>
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643238#comment-15643238
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Ya that can be made equal. But I avoided
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643235#comment-15643235
]
stack commented on HBASE-16890:
---
Found it over on HBASE-17021. I tried it. Took a little more than the Duo
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643225#comment-15643225
]
ramkrishna.s.vasudevan commented on HBASE-16890:
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15643196#comment-15643196
]
stack commented on HBASE-16890:
---
I tried 90 for ratioio on @duo zhang patch and it took half as long
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15639624#comment-15639624
]
ramkrishna.s.vasudevan commented on HBASE-16890:
bq.Not log roll by too many WAL files,
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15639530#comment-15639530
]
Duo Zhang commented on HBASE-16890:
---
[~ram_krish] Not log roll by too many WAL files, it is a memstore
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15639495#comment-15639495
]
Duo Zhang commented on HBASE-16890:
---
Think of this scenario. 10 thread do append and sync
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15639413#comment-15639413
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I agree to this. In AsyncWAL it is
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638675#comment-15638675
]
stack commented on HBASE-16890:
---
Smile. No worries. I'll try.
> Analyze the performance of AsyncWAL and
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638647#comment-15638647
]
Duo Zhang commented on HBASE-16890:
---
Honestly I do not know... I have never changed it before. You can
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638594#comment-15638594
]
stack commented on HBASE-16890:
---
Takes an int. It defaults 50. You want it 100? [~Apache9]
> Analyze the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638548#comment-15638548
]
stack commented on HBASE-16890:
---
Let me try. Will report back in morning.
> Analyze the performance of
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638480#comment-15638480
]
Duo Zhang commented on HBASE-16890:
---
And for ioRatio, you need to cast the EventLoopGroup in
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638414#comment-15638414
]
Duo Zhang commented on HBASE-16890:
---
The sync request of AsyncFSWAL is asynchronous so theoretically we
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638395#comment-15638395
]
stack commented on HBASE-16890:
---
I'd think that asyncwal would aggregate more than the five threads FSHLog
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638369#comment-15638369
]
Duo Zhang commented on HBASE-16890:
---
If we have more sync request for AsyncFSWAL then no doubt FSHLog
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638213#comment-15638213
]
stack commented on HBASE-16890:
---
48core.
bq. Seems the problem is AsyncFSWAL can not use more CPUs even if
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638208#comment-15638208
]
stack commented on HBASE-16890:
---
We should try and get metrics on packet sizes. FSHLog is making fatter
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638205#comment-15638205
]
Duo Zhang commented on HBASE-16890:
---
So what's the hardware of your machine [~stack] ? Seems the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638206#comment-15638206
]
stack commented on HBASE-16890:
---
{code}
-- Histograms
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638184#comment-15638184
]
stack commented on HBASE-16890:
---
I ran the tests a few times and results consistent. Looking in FSHLog run
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15638140#comment-15638140
]
stack commented on HBASE-16890:
---
I ran WALPE w/ log roll disabled against a single, remote DN. I see that
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15636836#comment-15636836
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Let me see the pattern in log roll here.
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15636822#comment-15636822
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I have observed log rolls in both but I
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15636688#comment-15636688
]
Duo Zhang commented on HBASE-16890:
---
Ah I could also observe the same result with a larger data set.
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15636271#comment-15636271
]
Hadoop QA commented on HBASE-16890:
---
| (x) *{color:red}-1 overall{color}* |
\\
\\
|| Vote || Subsystem
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15636177#comment-15636177
]
Duo Zhang commented on HBASE-16890:
---
Anyway, with my config the performance of the two modified
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635932#comment-15635932
]
Duo Zhang commented on HBASE-16890:
---
But in my PE test the results are almost same for FSHLog and the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635912#comment-15635912
]
ramkrishna.s.vasudevan commented on HBASE-16890:
But with PE single thread
{code}
./hbase
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635907#comment-15635907
]
ramkrishna.s.vasudevan commented on HBASE-16890:
My recent test results
WALPE command
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635789#comment-15635789
]
Duo Zhang commented on HBASE-16890:
---
Another round, the old AsyncFSWAL is still the slowest. The other
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635709#comment-15635709
]
ramkrishna.s.vasudevan commented on HBASE-16890:
Thanks for confirming.
> Analyze the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635708#comment-15635708
]
ramkrishna.s.vasudevan commented on HBASE-16890:
I will update my patch with the change
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635692#comment-15635692
]
Anoop Sam John commented on HBASE-16890:
Did some tests on classic WAL vs async wal using PE tool
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635680#comment-15635680
]
Duo Zhang commented on HBASE-16890:
---
Can you try the latest patch to see if it performs better?
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635654#comment-15635654
]
ramkrishna.s.vasudevan commented on HBASE-16890:
The improved ASyncFSWALs also performs
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635647#comment-15635647
]
Duo Zhang commented on HBASE-16890:
---
Yeah with PE tool, the qps of AsyncFSWAL is about half of the qps
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635619#comment-15635619
]
Duo Zhang commented on HBASE-16890:
---
Fine. Let me run a 10G test first.
Thanks.
> Analyze the
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635615#comment-15635615
]
ramkrishna.s.vasudevan commented on HBASE-16890:
bq. ./hbase
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635609#comment-15635609
]
Duo Zhang commented on HBASE-16890:
---
[~ram_krish] What is your command when running PE tool? I'm going
[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15635522#comment-15635522
]
Hadoop QA commented on HBASE-16890:
---
| (x) *{color:red}-1 overall{color}* |
\\
\\
|| Vote || Subsystem
1 - 100 of 201 matches
Mail list logo