BookmarkSubscribeRSS Feed
titan31
Quartz | Level 8

Hi,

 

I thought I'd find a white paper or something on this online, but having no luck finding anything, so just asking here.

 

Is there a performance benefit for users when upgrading from a non distributed VA to a distributed VA environment, or are any performance increases just on LASR server?

 

Looking to see if we upgrade, would there be benefits on the front end for the users in that reports on VA would load faster etc.

 

Thanks,

8 REPLIES 8
titan31
Quartz | Level 8

Thanks,

 

Read that there, but still wouldn't be fully clear. That page actually says in regards to non distributed that 

 

- As a result, the processing times are very fast and the results are delivered almost instantaneously.

 

I find that VA is very slow when people are using it and results can take a good bit of time to load up.

 

I don't think we'd be using a co-located data provider, so would there be much performance increases for VA users. 

 

If 10 people are using DataSet 1 on a distributed environment, and 10 on a non distributed environment, which one loads faster and is faster to respond, or are they both equal (or negligible difference)?

SASKiwi
PROC Star

If you are having performance issues with your non-distributed SAS VA don't automatically think that going distributed will solve all your problems. If your VA configuration does not have optimal settings for your usage it could give poor performance regardless of your architecture.

 

We experienced poor performance with our non-distributed SAS VA some time ago and went through a lengthy investigation with SAS Tech Support. It turned out some key settings on the mid-tier web applications required changing. Performance improved dramatically and is now quite acceptable. 

DJWanna
Obsidian | Level 7

Do you recall what you corrected to increase performance. One thing that helped us was to use any other browser than IE.

titan31
Quartz | Level 8
Ya,

That's what I'm imagining will likely be the case. We've no test system in place at the moment, so it doesn't allow us to test out any setting changes and I'm not making any changes that I can't test to make sure it doesn't break, but management have decided a new upgraded platform (moving to Distributed and Linux) is better than a test environment.
JuanS_OCS
Amethyst | Level 16

Hello @titan31 ,

 

yours is a good and very interesting question.

 

There are several important and relevant differences from the SMP (non distributed) to the MPP (distributed) versions.

 

The main one is the sizing: the non-distributed has limitations on its hardware sizing, and, then, the operating system has some limitations as well to manage a certain amount or resources above a certain number. For a MPP VA, this limitation does not exist, virtually speaking, and it has virtually speaking, unlimited growth potential.

This bottleneck speaks mainly about the number of users, amount of data and biggest table your VA system can handle while ensuring a minimum amount of performance.

 

Another big one is the MPP capacity, the multi-parallel  processing ability. A single system SMP has also some limitations in this regard, unless you manually code SAS/CONNECT sessions, which won't happen in VA anyways. Also, the way that a SAS Session works, is to take a certain amount of CPUs (by default, 4, for VA sessions, unlimited).

An MPP deployment will allow parallel loads and processing of data, thanks to the undelying TKGrid components.

 

And not only that, certain procedures, related to the high performance (HPA) capabilities of SAS and, in particular, of SAS LASR, are available only to MPP deployments, and in Linux. If you give a look to the VA User documentation, you will be able to see a huge difference between both of them.

 

All in all, if you abstract a little the underlying idea, you might be able to see that LASR is "just" the SAS representation or answer to a general IT and business requirement. And the same you can see in SAS LASR, as in any Hadoop distribution, or other flavours of IT components that can provide High Performance and Availability. The answers, will be also, most likely, the same.

 

Just informationally speaking, that LASR technology which answer to the question, is what now it has evolved to CAS, to give the same answers, but adding the cloud requirements, open languages, and such. 

titan31
Quartz | Level 8
Thanks JuanS.

I can definitely see how a distributed environment will increase load times in LASR, but bar the once a month I reboot the servers after patching (Windows so have to anyway), I wouldn't see the loads in LASR to be a bottleneck anyway as it's quite spread out currently.

We're moving to Linux as well, which I think will probably be a bigger benefit than distributed from what I can see.
JuanS_OCS
Amethyst | Level 16

@titan31 , you will see, yes, Linux will be a great advantage in many ways.

 

If you don't see your requirements to grow, and the system performs well, I see no added value in a distributed environment either.

 

Besides my personal preference on having Linux on the server side as much as possible, I have been experiencing during the past years the fact that SAS software, especially SAS VA, will be always released earlier, and with more functionality, on Linux (in especial RHEL) rather than on other operating systems.

suga badge.PNGThe SAS Users Group for Administrators (SUGA) is open to all SAS administrators and architects who install, update, manage or maintain a SAS deployment. 

Join SUGA 

CLI in SAS Viya

Learn how to install the SAS Viya CLI and a few commands you may find useful in this video by SAS’ Darrell Barton.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 8 replies
  • 2518 views
  • 5 likes
  • 5 in conversation