Re: ASM Filter Driver Performance
Date: Fri, 26 Jan 2018 11:45:44 +0000
Message-ID: <CABe10sa_HQk0yCBdRAqiOot26WaYVDNZyMRps8rQvgxPk8kS-g_at_mail.gmail.com>
Hi Jared
The AWR reports look a little odd to me. I'd expect a SLOB Update test to
result in write I/O as well as read I/O. For comparison here's an output
from a 25% update run I did recently.
Load Profile Per Second Per Transaction Per Exec Per
Call
~~~~~~~~~~~~~~~ --------------- --------------- ---------
---------
DB Time(s): 15.9 0.4 0.10
4.98
DB CPU(s): 1.8 0.1 0.01
0.57
Background CPU(s): 0.6 0.0 0.00
0.00
Redo size (bytes): 8,739,600.1 218,657.4
Logical read (blocks): 41,982.9 1,050.4
Block changes: 20,930.0 523.7
Physical read (blocks): 38,938.4 974.2
Physical write (blocks): 10,708.4 267.9
Read IO requests: 38,938.3 974.2
Write IO requests: 9,099.6 227.7
Read IO (MB): 304.2 7.6
Write IO (MB): 83.7 2.1
So far as I can tell there's no write activity in your AWR reports. I'd
further suggest that a useful metric since you are looking at I/O
performance, would be SLOBOps/sec/vCPU - or else duration of a fixed run.
That all said, I see you are using the current (or at least a recent)
version of slob, whereas I'm still using one from 2014 (not yet validated
comparability of results from the newer kit).
On Fri, Jan 26, 2018 at 1:36 AM, Jared Still <jkstill_at_gmail.com> wrote:
> Hi Mladen, nice to hear from you.
>
> Comments inline.
>
> On Thu, Jan 25, 2018 at 5:05 PM, Mladen Gogala <gogala.mladen_at_gmail.com>
> wrote:
>
>>
>>
>
>> - In detail describe machine on which the software was tested. If it
>> was tested on a VM, that makes the statement more than suspicious because
>> the performance of virtual IO is usually much, much slower than the
>> performance of real disk devices.
>>
>> Details were there for the AWW/EC2 instances.
>
> I have also tested on a local small server (not VM) as mentioned in the
> article.
>
> Details for it were not included as the performance difference between
> ASMLib and ASM Filter Driver were consistent regardless of whether run on
> VM or directly on Hardware.
>
>
>>
>> - Describe the test and the load that was used to establish such an
>> incredible difference.
>>
>> If you are familiar with SLOB , the details are in the article.
> ( https://kevinclosson.net/slob/ )
>
>
>>
>> - Explain the mechanism which makes ASM Filter Driver so much faster
>> than ASMLib.
>>
>>
> Interesting that there would be an expectation for me (or anyone) to be
> able to explain just what Oracle Developers did to improve their code.
>
>> As I have already written, ASM Filter Driver is very new. Any bugs in the
>> driver mechanism can bring your instance, and therefore your database,
>> down. Such outlandish claims do not convince me on the merit of using
>> filter driver.
>>
> Outlandish claims?
>
> No claims were made, just results of performance tests.
>
> You are certainly free to ignore them, or even better, run your own
> performance tests and publish the results.
>
> BTW, there will be more ASM details in presentations at Hotsos and IOUG
> Collaborate.
>
>
> Jared Still
> Certifiable Oracle DBA and Part Time Perl Evangelist
> Principal Consultant at Pythian
> Pythian Blog http://www.pythian.com/blog/author/still/
> Github: https://github.com/jkstill
>
>
-- Niall Litchfield Oracle DBA http://www.orawin.info -- http://www.freelists.org/webpage/oracle-lReceived on Fri Jan 26 2018 - 12:45:44 CET