Comparing 2 SAS servers - program runs 4 hours longer on one server - Block Input Operations ?

[gothaggis]

Have one physical server and one virtual server.  Programs run on the virtual server seem to take longer.  I have enabled performance logging.  One thing I have noticed is that on the slower server, the value of Block Input Operations is always 0 - this is not the case on the faster physical server.  The CPU clock speed is faster on the virtual server (2.3 vs 2.0ghz)

faster physical server:

NOTE: There were 87651 observations read from the data set 
NOTE: PROCEDURE PRINT used (Total process time):
      real time           34:12:29.27
      user cpu time       34:11:31.35
      system cpu time     51.16 seconds
      memory              59511.96k
      OS Memory           72964.00k
      Timestamp           08/10/2018 10:50:41 PM
      Step Count                        2  Switch Count  0
      Page Faults                       95
      Page Reclaims                     29512917
      Page Swaps                        0
      Voluntary Context Switches        2278
      Involuntary Context Switches      56795
      Block Input Operations            418712
      Block Output Operations           1298184

      69         ods csv close;
NOTE: The SAS System used:
      real time           35:07:26.82
      user cpu time       35:06:05.80
      system cpu time     1:05.83
      memory              66841.34k
      OS Memory           72964.00k
      Timestamp           08/10/2018 10:50:41 PM
      Step Count                        2  Switch Count  20
      Page Faults                       229
      Page Reclaims                     29610854
      Page Swaps                        0
      Voluntary Context Switches        47504
      Involuntary Context Switches      60589
      Block Input Operations            923088
      Block Output Operations           1669976

slower virtual server:

NOTE: There were 87651 observations read from the data set 
NOTE: PROCEDURE PRINT used (Total process time):
      real time           37:10:44.68
      user cpu time       37:09:48.54
      system cpu time     52.78 seconds
      memory              61736.18k
      OS Memory           77348.00k
      Timestamp           08/11/2018 01:42:20 AM
      Step Count                        2  Switch Count  0
      Page Faults                       0
      Page Reclaims                     15173
      Page Swaps                        0
      Voluntary Context Switches        2519
      Involuntary Context Switches      41989
      Block Input Operations            0
      Block Output Operations           1298184

67         ods csv close;
NOTE: The SAS System used:
      real time           37:59:02.81
      user cpu time       37:57:21.07
      system cpu time     1:38.17
      memory              70975.18k
      OS Memory           77348.00k
      Timestamp           08/11/2018 01:42:20 AM
      Step Count                        2  Switch Count  20
      Page Faults                       0
      Page Reclaims                     23818
      Page Swaps                        0
      Voluntary Context Switches        2749
      Involuntary Context Switches      48514
      Block Input Operations            0
      Block Output Operations           1639480

any ideas why its so much slower on the virtual server?  Does it have anything to do with Block Input Operations?

[Tom]

Times don't look that different to me. 

34/35 vs 37

[gothaggis]

those are hours - so 34/35 hours vs 37 hours.

[SASKiwi]

Looks like your virtual servers are using more CPU time - nearly 10% more. Is this true of all SAS jobs? Due you have access to SAS Environment Manager? This may give you a better understanding of the differences.

[MargaretC]

If the two underlying systems are the same, (same chipset; same number of cores/RAM; same connections to storage; same storage) this it is probably just the overhead of  virtualization – which can be up to 15%. 

It looks like there is a 10% difference in the times you have listed (2053 minutes compared to 2280 minutes) in the PROC PRINT statements.  So, it looks to be just the overhead of virtualization.

Cheers,

Margaret

[Kurt_Bremser]

You have a proc print to csv of a dataset with less than 100000 observations, and it runs more than 30 hours (and eats more than 30 hours of CPU time)? What kind of dataset is that?

[SASKiwi]

@Kurt_Bremser Looks like the OP is using ODS to create a CVS using PROC PRINT. In my experience ODS is a resource hog when writing out large files. I bet a simple PROC EXPORT using DBMS = CSV would be hugely faster.

[Tom]

Faster than PROC EXPORT is to just use a data step.

data _null_;
  file 'myfile.csv' dsd ;
  set have ;
  put (_all_) (+0);
run;

You can even have the data step write the header row as @Data_null_ has posted multiple places including:

https://communities.sas.com/t5/Base-SAS-Programming/Output-to-delimited-format/m-p/292822#M60848

Here is a link to macro that uses that method:

https://github.com/sasutils/macros/blob/master/csv_vnext.sas

[Kurt_Bremser]

---

@SASKiwi wrote:

@Kurt_Bremser Looks like the OP is using ODS to create a CVS using PROC PRINT. In my experience ODS is a resource hog when writing out large files. I bet a simple PROC EXPORT using DBMS = CSV would be hugely faster.

---

In German-speaking countries, we call this "carrying the church around the cross". Using ODS which creates lots of formatting, just to completely lose that in the csv is - ahhh - well, I'm in a civil mood today, so I won't write it here.

@gothaggis if this is actually the case and you just wanted to avoid the 10 minutes of work it takes to write the data step (or even simpler: the proc export) for a csv export, let this be an object lesson in "do not take shortcuts".

And have a look at Maxim 14.