Large discrepancies between processing identical files on Org vs EU site

[jennaj]

Hi @M.r.t

Sorry to hear that you have been having problems.

The tools across servers should be exact. Maybe we can discover what the differences are so you can get the results synced up? Using a workflow would be one way to do this since all of the details would be in the workflow: tool versions, parameter settings, reference data. Then you would control the inputs. You could even move the history with the inputs across servers, intact, then run the workflow against both copies, and then expect identical results.

Now, some tools will have non-deterministic algorithms, but the differences between runs even on the same server would be expected to vary a bit, too, not just the tool running on different cluster nodes, since time would be a variable, no matter where you are working.

I would like to review some of the differences you found. You can share back the histories with the data. If you do not want to share those publicly, we can move into a private message here instead, and only add in other administrators as needed (I’ll ask first!).

This error would be interesting to review: the history share link for both servers. Please note the datasets involved. I’ll be able to see the other details from just that.

@M.r.t

The second crazy discrepancy was when I processed another paired set on the same tool (on the EU site) and it was marked as completed without issue, but upon further inspection I saw that the resulting BAM file was much smaller than the other completed BAM files (2 GB vs the average 10-25 GBs). Again, I decided to process the same paired files on the same exact tool on the org site and the resulting BAM file was a much more reasonable 21 GBs.

And, this one would also be interesting to review.

@M.r.t

However, this one failed and I was presented an error. I repeated this twice and still ended up with an error. I tried to process this paired set using the same exact tool on the org site and to my surprise it was processed without issue.

---

---

How to generate a history share link is in the banner of this forum, also here →

@jennaj

How to get faster help with your question

We love details! Consider Sharing your History or posting screenshots of your Job Information Details page.

Please confirm the Server URL (your web browser at the top!) and any Tutorial links involved.

Let me know if you want to post those into a chat message instead of here. The screenshots will probably not be enough for this one since we are looking at the deeper technical details too. Thanks!