Parallel.Foreach taking more time than normal foreach in c#

Question

Below is code which I am using in Parrllel.Foreach. But it is taking same time as normal foreach. Here filingDataItems, nonPeriodic, periodiccells are data tables. Does this happening due to data tables?

System.Threading.Tasks.Parallel.ForEach(tempfilingReferences, T2 => { Filing f = new Filing( T2.Id, T2.DocumentPeriod, T2.FilingDate, T2.VersionId, T2.DocumentId, T2.PrimaryPeriodTypeId, T2.IssuedAsPreliminaryFlag, T2.IssuedAsAmendmentFlag, T2.FilingDetails, T2.CompanyId, GetFilingDataItems(filingDataItems, T2.Id, nonPeriodic, periodiccells, filingToPeriodList, DPTPList, posList), GetPeriods(filingToPeriodList, periodInfoList, T2.Id), GetFilingToPeriods(filingToPeriodList, T2.Id), true, false, AuditedDataType.Financials, T2.FilingTypeList ); lock (filingListnew) { filingListnew.Add(f); } });

You have all the overhead of setting up a parallel loop, but because you are locking to do the actual work, you have few of the benefits. Result = performance decrease. — Matthew Watson
– Matthew Watson, Commented Feb 16, 2015 at 11:17
Things have to actually be parallizable to benefit here, you dont get a magic speed increase just by doing Parallel.ForEach — Jamiec
– Jamiec, Commented Feb 16, 2015 at 11:19
"Does this happening due to data tables?" - we don't know, you didn't show that code. Apart from the lock(), the methods GetFilingDataItems(), GetPeriods() and GetFilingToPeriods() may very well be slower when called in parallel. — CodeCaster
– CodeCaster, Commented Feb 16, 2015 at 11:25
Also note, that for small data sources or simple operations inside the loop the overhead of running stuff in parallel can overshadow the performance benefit — Gerino
– Gerino, Commented Feb 16, 2015 at 11:35

EvilFonti · Accepted Answer · 2015-02-16 11:27:40Z

You are locking in the parallel part. So you wait in each thread for the active thread to remove the lock. So this is nearly the same as the sequencial foreach.

For example:

Parallel.Foreach:

Cycle1

Thread1: filingListnew.Add(Object1);

Thread2: locked

Thread3: locked

Thread4: locked

Cycle2

Thread1: locked

Thread2: filingListnew.Add(Object2);

Thread3: locked

Thread4: locked

Cycle3 ...

"Normal" foreach:

Cycle1

Main Thread: filingListnew.Add(Object1);

Cycle2

Main Thread: filingListnew.Add(Object2);

Cycle3 ...

As you can see in the examples you cannot gain performance the way you are using ParallelForeach.

See Dmitry's answer, but to address that specific question: use concurrent collections (works best if you don't need it to be in any specific order).
If you just did the ParallelForeach for not blocking the main thread a long time you can move the whole foreach into an other thread

Dmitrii Bychenko · Accepted Answer · 2015-02-16 11:32:56Z

You'd rather not lock at each step, but try using LINQ:

 filingListnew.AddRange(tempfilingReference .AsParallel() .Select(T2 => new Filing(...)));

If you have to preserve the order :

 filingListnew.AddRange(tempfilingReference .AsParallel() .AsOrdered() .Select(T2 => new Filing(...)));

You may find useful to create the list, not adding to it:

 filingListnew = tempfilingReference .AsParallel() .AsOrdered() .Select(T2 => new Filing(...))) .ToList();

Linq is easy to switch between sequential/parallel just comment out AsParallel() (or put AsSequential())

Collectives™ on Stack Overflow

Parallel.Foreach taking more time than normal foreach in c#

2 Answers 2

2 Comments

Comments

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

Comments

Related