MAXDOP and HT

Question

MAXDOP and HT

Viewing 3 posts - 16 through 17 (of 17 total)

You must be logged in to reply to this topic. Login to reply

PiMané SSChampion Points: 13965 More actions · Answer 1

I know that nowadays things might be a bit different... but there are some old CPU out there...

And if HT cores are as fast/good/equal to physical cores why don't they just call them "cores"...

In the example I gave the 30% means 30% of power a physical one... So if a physical core take 60 secs to perform an operation a HT one takes 200 secs.

A 4 cores, for numbers, could be 400 points, and a dual core with HT is just 260 points.

But the main question is are HT cores as fast as physical so we can treat them the same way and don't have to make a distinction?

Thanks

If you need to work better, try working less...

Jacob Wilkins One Orange Chip Points: 27976 More actions · Answer 2

There are few considerations here, and I think some of them are running together in the conversation.

1) Assuming the same CPU architecture, sure, 4 logical cores backed by 4 physical cores will be faster than 4 logical cores backed by 2 physical cores, but I didn't think that was ever really in question.

2) Running a query with MAXDOP 2 and then with MAXDOP 4 introduces a bunch of factors that don't necessarily have anything to do with hyperthreading. Even without HT, the same query with higher MAXDOP can be slower.

3) The question as I understood it was this, which isn't at all addressed by the above points/tests: If you have a server with N physical cores and SQL Server's MAXDOP set to N, what is the effect of HT on vs HT off?

You might want to check other tests that have been done that more directly measure this. For example, Joe Chang did some tests at http://sqlblog.com/blogs/joe_chang/archive/2013/04/08/hyper-threading-performance.aspx, as has Linchi Shea at http://sqlblog.com/blogs/linchi_shea/archive/2012/01/17/performance-impact-not-all-is-better-with-hyperthreading.aspx and http://sqlblog.com/blogs/linchi_shea/archive/2012/01/05/performance-impact-hyperthreading-for-reporting-queries.aspx.

With a fixed MAXDOP of N and N physical cores backing the workload in both cases, you'll probably not see a large performance difference either way, and will likely see some workloads perform better with HT on and others better with HT off.

You'll just want to test, and make sure you're testing apples to apples.

Cheers!

PiMané SSChampion Points: 13965 More actions · Answer 3

Thanks all for your help.

I'll look deeper into Microsoft KB about MAXDOP and SQL Server 2014 SP2 and SQL Server 2016 automatic soft NUMA nodes..

If you need to work better, try working less...