Followed the instructions I outlined in this post

http://blog.saasinct.com/2010/09/28/speeding-up-web-report-studio-9-13-1-by-jumping-in-puddles/

Was simple enough to setup and get working, only trick was to remember to stop and restart the webserver so WRS/WRV could pick up the changes and switch over from load balanced to pooled.

Other issue was deciding what the minimum number of connections for a puddle to define.  Checked the help, and there was some, but not a lot of insight was gained after reading it.

“Minimum Available Servers

Minimum Available Serversspecifies the minimum number of servers that should be idle and available in the pool at any time. The number of running servers never exceeds the Maximum Clients value specified on the server. If the number of available servers falls below the value specified in this field, additional server connections are created.Minimum Number of Serversspecifies the minimum number of servers (both idle and active) that should be present in the pool at any time. This value also specifies how many servers should be started when the pool is first created.”

Clear as mud eh!

But the key seems to be if you dont have enough pooled sessions available the users waits for one to become free.  So you would create lots then wouldn’t you.  Ah but each one is taking up memory as it sits there waiting for a request, so you dont want to many.

Haven’t found any form of management or monitoring tool that keeps track of how many sessions you have available and  how many requests over time, and of course any mismatch.  Im guessing we are going to have to go log mining yet again.

Some other good sources of info on Pools, Puddles, Sessions and Gumboots can be found here:

Overview of Pooling

Planning the Pooling Security (IOM Bridge only)

Main two issues that constantly arise are:

• Metadata server exceeding addressable memory in 32 bit windows and crashing
• I/O errors intermittently reading/writing SAS datasets

We have tried all the SAS “support” tools to reduce the size of our metadata server.  The final answer seems to be we will have to create a new one from scratch and export/import the object we need to retain.

We have also “tweaked” the windows file cache to reduce the I/O errors due to the windows file buffers being written faster than they can be cleared.  On that note I found this tech support note the other day:

Problem Note 36664: Potential issues with heavy SAS® I/O workloads on Windows 2003, 32- or 64-bit operating systems

Which I paraphrase as saying dont use Windows 2003 if you want to actually process any reasonable amount of data.

Of course the real answer it seems is to move to SAS 9.2 and 64 bit Windows 2008, but it does make me wonder how long it would take before you blew the limits of that operating system?

And shouldn’t the SAS environment at least tell you when it was experiencing these issues, rather than just abending so rudely?

This was to improve performance as pooled workspace servers remove the requirement for sas executables (sas.exe) to start and stop for each query, and therefore speeding up environments with lots of small queries.

SAS issued this tech support note:

Usage Note 39065: Documented best practices and other resources for improving the performance of SAS® Web Report Studio

Which outlines how to setup pooled workspace servers in SAS 9.1.

It also outlines a few other performance related changes you may want to try.

This enables users to run a summary report, and then click on a value and drill to the detailed transactions that make up the total.

One of the issues with this in WRS 3.1 (SAS 9.1) is when a user clicks on a sub-total within the WRS report they often drill to a blank report or a report that doesn’t show valid transactions for the summary value.

We eneded up having to train our users to never click on a sub-total to drill down.

There is a usage note at http://support.sas.com/kb/40/708.html that describes when drilling down on sub-totals will and won’t work.

Good to know when this is going to happen, but of course what is missing is the link to a hotfix that makes it work properly, aaarrrggghhhh.  Of course it is fixed in SAS 9.2 / WRS 4.2 I believe.

As per this tech support note:

Problem Note 20991: SAS® Information Maps with more than 999 data items might fail to open with the SAS Information Map Libname Engine (IMLE)
http://support.sas.com/kb/20/991.html

We have applied the hotfix and checked the config and it all applied as per the tech support note, but we still have the issue intermittently.

What we cant work out is we get this scenario:

• User A opens EG on PC and gets error on info map
• User B uses User A’s open EG client, reconnects to metadata as themselves and it info map opens ok.

• User B opens EG on their PC and info map opens ok
• User A uses User B’s open EG client, reconnects to metadata as themselves and it info map opens ok.

So we deduce that

• it is not the info map because some people can open it
• it sis not the server because some people can open it
• its not users A EG client, because it works for user B
• its not User A’s metadata security because it works on user B’s PC

So only think I can think of is cached files on the EG side, or memory issues on the server?

But as I said it is intermittent across environments, across users.  The only thing we do know is that it only happens on our very large Information Maps.

Any ideas?

Its a complete mare!

One of the issues we have struck is the inability to delete any packages you have published to a channel.  They just stay there forever.

In SAS 9.2 you get a utility that lets you delete these packages:

SAS 9.2 Using the Package Clean-Up Utility to Remove Packages

Luckily the friendly chaps at SAS tech Support mentioned that there is a SAS Toolpool entry for 9.1.3 that does teh same function.

So if you need to delete Archived Packages in SAS 9.1.3 channels, then as your friendly SAS dude for the toolpool entry.

We have all our Warehouse reporting data stored in Star Schema’s in an Oracle Database.

We have found that the users are accessing the Information Maps we have created via SAS AMO, which is a good thing.

But the bad thing is that when a user outs that data into a pivot table, Excel/AMO downloads the data and stores it in Excel to populate the pivot table ans they are ending up with some massive Excel file sizes.

The answer is to build some OLAP Cubes and use these as the source for the pivot table.  But of course then we strike the issues with OLAP in SAS 9.1.3 such as no incremental updates, OLAP locking etc.

So we are trialing SAS OLAP Cubes in ROLAP mode whihc means the data is retained in Oracle and only the cube structure is built in SAS, which bypasses all these issues.

But of course it is not as easy as it seems.  We run a distributed SAS Windows environment so I need the OLAP server to be able to see the oracle instance etc.

In testing I found it a pain to have to login to Excel/AMO and/or EG to test the OLAP Cube.  Also if there are any errors not a lot of detail is provide.

So instead I just use this code:

proc sql;
connect to olap (host=<host> port=5451 user=<user> pass=,pwd>);
select * from connection to olap (
SELECT
{ [Measures].[COUNTSUM] } ON COLUMNS  ,
{ [region].[region].DefaultMember.Level.Members } ON ROWS
FROM
[,olap_cube_name.]
);
disconnect from olap;
quit;

Check the logs and see any errors.

One customer I work with is in the next year camp.  We have just started the rollout of EG across the business and one of the things we encourage/mandate is that before a user gets access to EG they must have attended the SAS EG course. (EG is just so damn powerful/dangerous in a untrained users hands!)

But just been told SAS now only run courses using SAS 9.2 software.  So a dilemma or is it?

If we send our users on the EG 4.2 course I am concerned they will see all the things EG 4.1 doesn’t do and will focus on that, not the things EG 4.1 does very well (compared to other tools, SAS and non SAS)

But then again we need them trained.

EG 4.2 is a little different to EG 4.1 in navigation, functionality etc, but probably not enough that the user will get lost using 4.1 after being trained in 4.2.

(and I can understand SAS need to move training/courses to the latest version etc)

We have the SAS eLearning available to users (funnily enough still only SAS 9.1 based, wonder when that will change to 9.2?), so maybe self learning is enough?

Interested in your views, what are you doing for SAS 9.1 training?

Of course the customer is also currently recruiting BAU DI resource, so DI studio training will be a much bigger issue as DO Studio 3.4 compared to DI Studio 4.2 is a much bigger change.

Lastly if you know of any companies that sell SAS 9.1 training material (these have to have been developed independently of the SAS Education course notes of course and we will check!) that we could use to deliver internal training, let me know.

Also this week I seem to have been shown a lot of tips and tricks I didn’t know you could do.

One of them was found by Rob and passed on to me (thanks Rob).

The issue was in SAS 9.1.3 Management Console and related to some of the screens not having all the relevant fields displayed.  As an example the Libname edit screen might be missing the field where you normally define the file path.

The fix was to go into the Metadata Manager > Resource Templates tree in SMC and delete the relevant template.

Then logout of SMC and log back in and SAS will automagically re-create a new template, and it will be valid.

After the installs were all complete and we started migrating users from our old Citrix server, they started complaining of issues with DI Studio.  These included:

- Nodes missing in the Inventory tree

- Options missing in the custom transform / custom code nodes

Adding the following option to the DI Studio ini file fixed both issues:

-DentityExpansionLimit=1000000

No idea why it fixed it and only reference on support.sas.com is a similar setting for Information Map studio here to fix it not being able to open more than 999 objects.

But as always fix it first, question it later.