Hi everyone, I am having a small talk/demo in the ...
# random
l
Hi everyone, I am having a small talk/demo in the group I am working with about Nextflow. They have predominantly been using Snakemake so far and want to learn more. Does anyone happen to have any good graphics/plots for comparing both? I think it would be especially useful to perhaps have a benchmark on the performance of both tools and how they perform with input size 🤔 . Unfortunately I have not found any helpful resources for this, but maybe someone knows more. Many thanks for the help.
p
For most workflow managers, performance of the manager isn't very important as they typically do rather little work by themselves (and while most do have some features that allow for processing "inside" the workflow manager, I think most would agree it's an antipattern).
p
It's pretty old now (obviously, given the company 🙄 ) but this 23andme blog post is one I've shared before about characterisation of Nextflow running at large scale: https://medium.com/23andme-engineering/introduction-to-nextflow-4d0e3b6768d1
Here are some stats from our production run and the initial scale we ran at:
• 1 nextflow head job computes 20,000 individuals
• 8–10 nextflow head jobs ran in parallel, 1 kicked off every half hour via AWS Lambda triggered by an AWS CloudWatch rule
• >5000 task jobs per head job running on AWS Spot Instances
• ~2000 task jobs on average in runnable/starting/running state at all times in the AWS Batch dashboard
• >800K succeeded task jobs showing up in the AWS Batch dashboard
• Less than 1.5% of task jobs fail mostly due to spot terminations and all are retryable
• Processed ~1M individuals per day
• The compute environment taking care of the task jobs scaled up and down between 4K and 10K vCPUs automatically based on the request load
This was in 2021, those numbers are actually fairly small now in comparison to some of the workloads that folks are running with Nextflow
l
I appreciate the comments and help from both of you. Thank you very much!
👍🏻 1
p
@Marcel Ribeiro-Dantas @James Fellows Yates @Júlia Mir as a random start list of tags, might have presentation decks kicking around that could help too..
j
I don't have anything specifically comparing the two unfortunatly 🫤
j
I don’t have anything comparing them either 😞 in the last nf-core publication there is a plot comparing different workflow managers by publication, but this is not the technical comparison you are looking for. Tagging @Jose Espinosa-Carrasco just in case he knows about something
l
Really appreciate the effort everyone. It's not a big deal 😄 . I was simply unable to find anything useful online, so i figured this might be a good place to ask.
p
Again a bit old
Something I would emphasise if I were in your situation is that the most common complaint for Nextflow: crappy error messages, has gotten much much better (and continues to improve). Plus the recent VSCode extension with the language server makes writing Nextflow much easier.
If folks have tried Nextflow before a few years ago and not adopted it, these are major improvements in the developer experience since then.
👍 1
m
I don’t usually compare features in Nextflow to other workflow managers, as they change pretty often, and the risk of “you’re wrong, this has been fixed / is different now” is very likely 😆 What I usually do is a macro analysis on what integrations Nextflow supports, the big win of Seqera Platform and our whole ecosystem of tools, who’s behind Nextflow, [active] community size, nf-core and so on.
👍🏻 1