How to accelerate data step that running on remote server via EG

[whymath]

I'd like to ask: When I use EG to connect to the server and submit a data-step code,

1. The server executes code slowly, likely constrained by network latency, single-core CPU performance, disk read/write speed (HDD), memory allocation (8GB per user in the workgroup), and memory read/write performance;

2. My local computer significantly outperforms the server in all these hardware metrics;

Besides switching to PC SAS on my local machine, how else can I leverage my local computer's hardware advantages to accelerate code execution?

[SASKiwi]

You can't leverage your local computer's performance advantage unless you switch some of your SAS processing to it. You could parallel process some steps using both your PC and your remote server at the same time.

Have you discussed the SAS server performance with your SAS administrator? In my experience if your PC outperforms your remote SAS server then there is something seriously wrong with the server resources available to you. You might also want to consider doing remote server processing after hours if it performs much better then.  

[whymath]

By using sasfile statement to load input dataset into memory, I observe the time consumed by a data step from 30s decrease to 6s, which is very significant.

So disk read/write speed may be the key. How about add a 4TB SSD to server computer as the work library location? As there are 20 programmers in group, do they share the total 4TB or occupy 200GB maximum for each one?

[SASKiwi]

I suggest you check your server disks IO performance with the SAS IO Test Utility before doing anything.

[Kurt_Bremser · Accepted solution]

Maxim 30. : Analyze, Then Optimize.
You must get a clear picture of the issue first. Run a diagnostic tool on the server while you execute your code (with OPTIONS STIMER for in-depth log analysis, of course). My favorite on AIX was topas, where I could see the detailed CPU load ( sys, user, wait), disk throughput, network activity, top processes all at once. You also see if there‘s paging activity when not enough physical memory is available for the processes, which is a serious performance killer.

You should always use your fastest storage for WORK; unless there‘s a quota set (which any diligent admin will always do), any user can use it all.

A local PC being faster than a server points to a serious mis-conception in the design of the server.