Hi Tom, So you wrote: "Your first STP starts, runs until finished, and then stops. Therefore, it's not a problem, and needs no changes." Initially I think so, but looks like the STP server response time increase because of this one STP lock one(from a few) of the process that manage all STP requests. I'm not familiar with STP server configuration, I can only suppose how it works from my not very long experience. So, lets suppose you have UI application that has one button:) , and when user click on this button 10 Ajax requests asynchronously sends to STP server, STP server should process all of them and return to UI some Json or XML, but it's not so important now. So Lets say 5 "sas" process serve these 10 STP sessions that should be processed in same time.Using firebug and task manager on server machine easy observe how these processes serves : All 5 "sas" processes serve first 5 STP sessions, then after some of session finished - it serve some another session that before waits in queue  probably. So, lets go back to my problem -  When I didn't run my application(with two long time execute STPs) but run this application with one button - all these 10 Ajax request will finish sequentially successfully similarly as I describe higher. Now, When I run my problematic application that creates 2 STP session that runs a few hours, -  2 from 5 processes become locked by them. Now, when I run second application with one button - logically only 3 processes should serve these 10 STP requests, but in fact what I saw - 3 sas processes serve needed requests normally(finish one, then takes next etc.), but two of requests(that were allocated to 2 locked sas process mantioned higher) can be finished and run for a long time. And these two session will ends only after another , older STPs that runs for a long time will be finished. So due all these I think next, what I do : UI -still run two Ajax to STP, but STP sessions will not executes these macros for a long time - they will just run some x(or call execute) command that will create non STP session that will execute these two macros without impacting another STP users. You wrote: My first question is to ask whether you're sure that it is a resource drain? As far as I know, on any modern capable server a process sitting there mostly in a sleep state won't impede other processes. You are absolutely right, my second STP that observe logs from first STP uses sleep function that doesn't eat almost any cpu time, but anyway STP process is active, it doesn't unlocks on time,when sleep function executes, it lock sas process and keep it on serving only this one concrete session, as I understand. Have you done any testing to measure this effect? If it's very minor (which is what I would predict), just keep your code the way it is. Yes, I made the tests, I described results higher(UI with one button, 10 STPs and another UI with this 2 long-run STPs). Tom, as I wrote higher I am not the specialist on Admin part of STP server configuration, I'm not 100% sure how STP session are distributed between "sas" process that serve them, but even from general logic it looks a little bit not elegant to lock two(lets say from 5) sas processes that manage hundreds of users STP requests(majority of them executes quick, second-two)... I don't think that STP server purpose is manage long time processes, it rather some LSF batch stuff should do... I'm not sure , but I know that "platform process manager" tool gives possibility to run some flow on fly, by some process manage command(jrun for example). So I'll try from STP session create some x command that will send correct command to platform process manager that will run(on demand...) some flow with two jobs that will have UW code with needed macros(one run long executed macro, second - observe it logs)... Paradox of this task is the fact that it's very similar to usual task, with only one difference - macro executes long time and therefore all problems occurs, but anyway it's the interesting things and I hope I'll finish it this year:). Thanks! ... View more

Hi again Amit, Unfortunately I can't share the code but I'll shortly describe you main stages of implementation the macro on SAS. So first of all algorithm is already realized on many programing languages, I used realization on Javascript as template: https://gist.github.com/1440602 You can easy integrate this realization into some simple HTML page and use it for checking /testing your realization on SAS. Regarding SAS realization - I just substitute standard operations on Stack(push, pop etc.) and javascript array functions (contains) by SAS macro with needed parameters and output as sas data set(Stack I replaced by sas table with two values - pos and value, using them easy make push pop etc.). Main challenge was realize recursion - the problem is that SAS recursion isn't standard recursion as in java script, SAS just executes same code but previously set macrovars isn't cleans when new recursive execution starts etc. So I often use macro names like &&i&vertexName etc... I also often used goto statement,I know it bad practice but as I remember in some steps it was almost impossible realize needed functionality without it. Also , to simplify testing before starts algorithm I renamed vertex(nodes) in graph by simple name as 1,2,3 etc., as it is in javascript realization. Testing also takes also a lot efforts - because of recursion using log isn't very helpful, even small graph generates huge log(if you'll use mprint mlogic symbolgen options), so I used html output in needed checkpoints... Results of macro execution I compared with result that output javascript code that I described higher, I also changed a little bit this javascript code – it output not only targrt results but also some checkpoints info(stacks contents after each recursions execution etc.) As I wrote on my previous post  - pure implementation takes not so much time, but testing and integration with business requirement took a lot. If you'll have any question regarding concrete steps –write and I’ll try to answer… Thanks! ... View more

Hi Amit, I've sent request to my Bosses regarding permission of sharing the code, but suppose answer will be negative. In any case I'll be able to provide you simple guidance how to realize the algorithm on SAS. First of all, if you use SAS 9.3 - there are realized solution that very simplify the task,Chang demonstrate how to use it in post higher. On earler version of SAS(9.1.3) it doesn't realized(my case). Secondly Geofferey realized the same algorithm using proc IML, he attached his code in one of the post higher, long time ago I run it(just to compare performance side with my variant), and looks Geofferies code works correct, but again I didn't tested it carefully. Regarding performance - on small graphs execution time is approximately the same, in huge graphs Geofferies algorithm works a little bit fuster, but my solution use recursion instead of hardiocded max iteration amount in Geofferies variant(proc IML doesn't support recursion),so our realizations are very very different. My implementation use SAS macro loops, data manipulation ,recursion etc.,so as I wrote higher -  if I'll not be able post the code I'll share guidance hot to realize it quick and simple. Actually ,as I remember, It takes approximately 4 days(poor realization and tests). More time I spent on collecting source data and loads resutls data etc., but it's another question. Thanks! ... View more

Hi again Arthur, It also works. Thanks a lot for help guys! ... View more

Hi Tom, Thanks, it also works, looks like it is the most qucik solution. I have one more small issue, if few times write put statement it obviosly write text into two separate lines, but I need store them into one line, for example: data _null_;   set sashelp.class;   filename = catx('\','c:\temp',catx('.',sex,'txt'));   file out filevar=filename;   put "txt1";   put "txt2"; run; And result in file will be : txt1 txt2 But I need "txt1txt2". So may be you also know how to write all these lines into one file row? Concatenations two needed character variables isn't ok in  my case becouse they summury length can be more then 32767... Thanks! ... View more

Hi Arthur, Thanks, it's what I need! ... View more

Hi  Astounding , I should write many files from datastep. Offcource I can use few data step or some macro with loop and then file statement for each created file, but it will increaze execution time, and it is critical... In real task I should create ~20-30 files each time from one datastep. May be these functions - fopen fclose etc. were created just for managing(delete,create files) but in this case why it( fopen func) has ,for example, 'A'(append) prameter( modifier) if there is no possibility wite from same datastep into this file... Should be some possibility route any text into created from same data step file(I hope so)... Thanks! ... View more