I tested in this case below, the xadd/xread performance is not as good as pub/sub 1. m client xadd n client xread, m client pub n client sub 2. Pipeline

Comment From: yihuang

Does this patch make your use-case faster? https://github.com/antirez/redis/pull/6281