Exploring, modeling, predicting and reporting with SAS Visual Analytics and SAS Visual Statistics

VA Performance results/experience with a concurrent workload

Accepted Solution Solved
Reply
Contributor
Posts: 38
Accepted Solution

VA Performance results/experience with a concurrent workload

Hey,

i did not find a paper or documentation about VA Performance results or some user experience.

Maybe anybody can help me with the following questions.

I am searching for an valid statement, how much concurrent workload Visual Analytics running on an Single Machine (SMP) can handle.

Of course this depends on the table size, hardware sizing and many other parameters.

But maybe there is a calculation example?

I only found this paper with some Performance results.

http://h20195.www2.hp.com/V2/GetPDF.aspx%2F4AA5-1601ENW.pdf

[.....]SMP solutions, and allows for larger number of users to access those data sets concurrently. [...]

What is a larger number of users? They tested the performance with 5 light users and 1 heavy user and a 32GB dataset.

If i look at the CPU Utilization there isn`t much tolerance.

They used the following components :

Bildschirmfoto 2014-05-07 um 08.52.30.png

I think we will use  :

Bildschirmfoto 2014-05-07 um 08.37.21.png

So they have more ports to manage the network traffic. This will be probably the biggest faktor?

We won`t have 32 GB but much more concurrent Users.

I think :

max 6 GB

max 20 concurrent

Greetings from Germany

Marius


Accepted Solutions
Solution
‎05-08-2014 01:57 PM
SAS Employee
Posts: 44

Re: VA Performance results/experience with a concurrent workload

I think I was able to find exactly what you are looking for.  I posted the document to the forum as an article:

Hope that helps!

View solution in original post


All Replies
SAS Employee
Posts: 44

Re: VA Performance results/experience with a concurrent workload

,  unfortunately, I can't give you a formula to calculate what kind of load a given number of users will place on the system.  There are just too many variables-- not only does number of users, data size and schema, and hardware play a role, but the activities of the users is a huge factor.  I can create a report with simple tables and graphs that perform little more than summarizations or I could create a report that is exported from VA Explorer with heat maps and decision trees that creates significant more load on the system.  If I then have 20 users just viewing each of those reports, one group puts much bigger load on the system than the other, even though both groups are just "viewing reports".

And although I hope that helps to understand why it's near impossible to provide a formula, it doesn't answer your question.

6 GB isn't much data and 20 users is more than reasonable for that system--  You will be fine.  I will pull up some of our performance testing numbers and post them later today.

Hoffentlich, hilft das etwa!

Solution
‎05-08-2014 01:57 PM
SAS Employee
Posts: 44

Re: VA Performance results/experience with a concurrent workload

I think I was able to find exactly what you are looking for.  I posted the document to the forum as an article:

Hope that helps!

Contributor
Posts: 38

Re: VA Performance results/experience with a concurrent workload

Hello David,

the paper is exactly what i need.

I thank you Smiley Happy

see you soon,

Marius

Post a Question
Discussion Stats
  • 3 replies
  • 638 views
  • 0 likes
  • 2 in conversation