February 14, 2017 at 10:57 pm
Comments posted to this topic are about the item SQL Server vs SQL Server v.Next
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This thing is addressing problems that dont exist. Its solution-ism at its worst. We are dumbing down machines that are inherently superior. - Gilfoyle
February 15, 2017 at 1:41 am
1. Did you do Buffer Pool clean before the second time update operation?
2. I'd like to advice you add another compare to vNext run base on docker container?
That's awesome performance testing compare...
Thanks
February 15, 2017 at 2:00 am
thanks
did you compare vnext on windows vs vnext on linux>?
pini
February 15, 2017 at 6:30 am
I don't think it's fair to say that extra CPU makes up for lacking disk performance. Your test operations are almost purely io, so CPU is almost irrelevant. Also, what exactly is this test proving; I.e. performance of what exactly? In order to say that differences in performance are not related to hardware, you must have exact hardware. I'd even argue that the same machine should be used for both tests. I wouldn't really easy that this set of tests reliably says anything in particular about vNext at all.
Jared
CE - Microsoft
February 15, 2017 at 6:56 am
SQLKnowItAll - Wednesday, February 15, 2017 6:30 AMI don't think it's fair to say that extra CPU makes up for lacking disk performance. Your test operations are almost purely io, so CPU is almost irrelevant. Also, what exactly is this test proving; I.e. performance of what exactly? In order to say that differences in performance are not related to hardware, you must have exact hardware. I'd even argue that the same machine should be used for both tests. I wouldn't really easy that this set of tests reliably says anything in particular about vNext at all.
Fair comments Jared.
The purpose of the testing was to show that out of the box SQL vNext on Linux, important part right here, on Linux, can perform equal and better to that of 2016 on Windows.
As I mentioned, not a very comprehensive test, nor do I claim it to be very scientific, but knowing how the 2 version would compete on simple tasks was something I needed to know.
I've noticed in my immediate community and with my colleagues, SQL on Linux wasn't really received with open arms, so this shows there's more to it and not slacking at all.
Thanks for the input.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This thing is addressing problems that dont exist. Its solution-ism at its worst. We are dumbing down machines that are inherently superior. - Gilfoyle
February 15, 2017 at 7:32 am
... Note: The hardware is very similar, though some might argue the performance of the Toshiba is below that of the Samsung. What the SSD on the Windows instance lacks, it gains with it's faster CPU. ...
Given identical hardware and an identical database as a starting point, I wouldn't expect the performance of SQL Server to differ significantly between Windows versus Linux in a production environment, because performance boils down to things like I/O latency and blocking. As others have pointed out, consider performing each test on the same machine; which can be done using VMWare or VirtualBox.
Also consider leveraging AdventureWorks (Adam Mechanic has a script to enlarge the tables) and also the SQL query stress tool created by Adam Machanic.
http://sqlblog.com/blogs/adam_machanic/archive/2011/10/17/thinking-big-adventure.aspx
https://www.brentozar.com/archive/2015/05/how-to-fake-load-tests-with-sqlquerystress/
"Do not seek to follow in the footsteps of the wise. Instead, seek what they sought." - Matsuo Basho
February 15, 2017 at 2:03 pm
All performance benchmark should be conducted on the same hardware, if possible.
I have ESXI at home with installed VNext 1.5 and 2016SP1. I will allocate the same resourcers for both, to compare.
MS SQL 2008 MCITP x 3
MS SQL 2012 MCSE x 2
February 15, 2017 at 5:55 pm
As a matter of interest -
did you have the same block size disk format?
did you experience file growth?
If so did you have the same size db and log files set up?
and did you have the same auto growth settings?
February 16, 2017 at 6:18 am
RichB - Wednesday, February 15, 2017 5:55 PMAs a matter of interest -
did you have the same block size disk format?
did you experience file growth?
If so did you have the same size db and log files set up?
and did you have the same auto growth settings?
Hey Rich, yes all DB properties were 100% the same.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This thing is addressing problems that dont exist. Its solution-ism at its worst. We are dumbing down machines that are inherently superior. - Gilfoyle
February 16, 2017 at 6:08 pm
Excellent effort. Still, hardware needs to be identical. As other have said, I/O and not memory with the test are perhaps the biggest factors, and you have to very different solid state hard drives. From what I have read, Microsoft is still working on improving the internal processes they are using to abstract the code from the OS layer. In time, it will only get better.
See: SQL Server on Linux - How I think they did it
http://www.centinosystems.com/blog/sql/sql-server-on-linux-internals/
And;
SQL Server on Linux: How? Introduction
https://blogs.technet.microsoft.com/dataplatforminsider/2016/12/16/sql-server-on-linux-how-introduction/
The more you are prepared, the less you need it.
February 22, 2017 at 6:10 am
Are there any benchmarks done yet with server-grade environments?
Br. Kenneth Igiri
https://kennethigiri.com
All nations come to my light, all kings to the brightness of my rising
February 22, 2017 at 6:14 am
Br. Kenneth Igiri - Wednesday, February 22, 2017 6:10 AMAre there any benchmarks done yet with server-grade environments?
Am sure there are, but maybe not published yet.
I might be wrong though, will need to search.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This thing is addressing problems that dont exist. Its solution-ism at its worst. We are dumbing down machines that are inherently superior. - Gilfoyle
Viewing 12 posts - 1 through 11 (of 11 total)
You must be logged in to reply to this topic. Login to reply