Right now, we use ELB (Elastic Load Balancer) to sit in front of multiple GPU instances.
Is this sufficient or do you suggest adding Celery into this architecture?