Benchmarking Redis on AWS ElastiCache | Hacker News Reader

Benchmarking Redis on AWS ElastiCache

blog.meldium.com

28 pointsbradleybuda13 comments

It is important to realize that these numbers are very, very low. Amazon's architecture has always been particularly terrible at running Redis, and the fact that they gave it a brand name has not changed that.

antirez's "How fast is Redis?" page (http://redis.io/topics/benchmarks) draws some attention to the fact that Redis gets much slower when virtualized. He mentions that a state-of-the-art VMware hypervisor will cut Redis's performance by a factor of two.

EC2's performance is considerably worse than a state-of-the-art hypervisor.

For comparison, I just ran the same benchmark the article describes on the desktop PC I'm currently using to post this. It cost me $800, last year, and that includes the video card. It gets an average of 881,670 OPS, which is over ten times faster than an m2-2xlarge machine in this benchmark.

I recently switched some of our eccomerce caching from a 3 Node ElastiCache Memcache cluster to a m1.large Redis ElastiCache instance for caching and a m1.small for session storage in Redis. So far the performance with Magento I've seen a 30-40% increase.

30-40% increase in what? Latency? Throughput?

I'd be curious to see how using ElastiCache Redis as a session store compares to DynamoDB (on latency, throughput, and price).

App server response time (New Relic measurement). I haven't looked into DynamoDB.

Just to be 100% clear, increase = worse in this context, correct?

One thing that would be interesting to see is how Redis on ElastiCache stands up against other cloud-based Redis offerings that are available in AWS, like redis-cloud.com. Is anyone here using one of these services?

probably around the same, although the winning features for those would rather be the auto-scaling and sharding features, instant failover etc (http://redis-cloud.com/redis/redis-comparison).

I'm always a little dubious about any service whose comparison page is "we do everything! others do nothing!" like that. Anyone got any direct experience with them?

Yep! My company evaluated their service, briefly. We asked how they perform on redis-benchmark. They provided no numbers and said "We have a way to stop users who generate benchmark traffic on our service."

We tried running the actual Redis-using process of our application on their free tier. The server went down.

Would not recommend.

Dear rspeer - we regret learning of your negative experience with our service. We do find it somewhat puzzling that you never reached out to us for assistance after your instance had crashed - we have thousands of active users and each incident is dealt with utmost professionalism regardless the plan type. If you wish, we invite you try our service again and contact us at support@garantiadata.com with any question or issue.

Sorry for the lack of follow-up, but there's a reason we were just evaluating. Nothing was forcing us to use redis-cloud in particular.

In fact, nothing was forcing us to use any sort of Redis in the cloud, which is good, because my conclusion is that nobody has a satisfactory solution for it. It's not just your service. It's not just Amazon either. Nobody has figured out Redis in the cloud.

Redis is designed to be a high-performance, in-memory data store. All "Redis in the cloud" offerings turn it into a low-performance data store that feels more like a database, and that defeats the point.

If you figure it out, let me know... but I'll want to be able to run redis-benchmark.

tl;dr Don't use the t1-micro or m1-small instances for anything serving interactive traffic, m1-large and m2-xlarge are much better than c1-xlarge

and the winners are m1-medium, m1-large, and m2-xlarge

Reply on news.ycombinator.com