Aerospike is well optimized to use SSD performance. The only thing needed is proper capacity planning and your production environment will be pretty stable.We have around 20-30TB data store in Aerospike. Here are below points which is like most about it :-

1. With a peak of 300-400k TPS , 99% of a incoming requests is still finishing in < 1-2ms. I believe it is the fastest Nosql Data store till date.

2. Cluster Setup is very easy.

3. Absolute Zero Maintenance as compared with Cassandra or Hbase

4. Easy to Monitor the cluster health and performance.

5. Enterprise Support is pretty good.

6. Version Upgrades is a very seamless process. Review collected by and hosted on G2.com.

However, there are many things which are work in progress in Aerospike :

1. Migrations : Migration is an activity that comes into action to balance and redistribute data when a node goes down and leaves the cluster. It can happen due to network issues or system level bugs which results in a service down time on that server. It is an expensive operation as it directly impacts the performance with a spike in latencies. This is further complicated as even latest Aerospike version is not smart enough to understand the difference between a manual service restart and a node going down.

2. Namespace Addition Needs Downtime : Namespace in Aerospike is an equivalent term to database in RDBMS. We have many different applications which are using their respective namespaces in production. Any new namespace addition is not dynamic untill now and it requires a service restart on all nodes in the cluster.

3. Isolation at Namespace Level :

For better and stable production environment, we must have better control of all the key metrics that are critical and need special attention. As of now, we can monitor all the TPS at cluster level but we cannot deter- mine how many TPS are happening at namespace level. Also, there is no method by which we can isolate or put constraints at namespace level in terms of number of read/write activiities that can happen at any name- space. We have faced situations in production where TPS spike in one less critical namespace impacts the performance of all other namespaces.

4. IP tracking in Connections : Aerospike provides a security feature in Enterprise Version where we can enable security on our distributed cluster but we cannot track the client hostname or IP from where connection is coming. Having this will help us to narrow down our search to identify hostname which is generating too many connections.

5. XDR Limitations : XDR Replication is a feature available in Enterprise Aero- spike which allows us to replicate from one datacenter to another. But, it is not possible to enable any security while XDR is enabled in the cluster. Also, XDR can not replicate updates at bin level and it replicates the whole record again increasing inefficiency at network level. Review collected by and hosted on G2.com.

The speed. We replaced a MongoDB with this and pound for pound, it was much faster. We performed a couple month evaluation of this versus a few alternative databases. We ended up picking Cassandra after all was said and done because the support was better at the time. It looks like now though Aerospoke has a pretty solid enterprise support plan. Review collected by and hosted on G2.com.

I didn't have anything that I disliked overall. Review collected by and hosted on G2.com.

My use case was to support billions hits a day and with response time of 2-3 ms. Nosql was obvious choice but After exploring several databases we zeroed into aerospike becasue it is :

-Extremely fast (could support >1 Million TPS on a single node)

-Its failover mechanism i.e. Automatic fail-overs

-Support large requests of read and write simultaneously.

-Horizontally scalable. Server side clustering. Sharded & replicated data. Review collected by and hosted on G2.com.

Not much to dislike but if to mention there few things like:

-Maintaining a cluster for early age startup

-In case of very large writes we needed to move to large computing machine Review collected by and hosted on G2.com.

Aerospike is pretty fast. One of the fastest - and easy to configure- tuple-based db. It has great use cases like caching service or storage for small ugc objects. It scales easily and performs like magic on SSD drives! The methods are generic and you find all the options you look for in a database. There are clients written in almost every popular languages. The support is pretty good, documentation is available and community managers are friendly and helpful. Review collected by and hosted on G2.com.

The clients - especially the node one- have been half-assed. There is no JS implementation for the node one nor post-install script to get the binaries. Logs were thrown in stdout freely and DB options were not available. It's a real pain for DevOps teams. I've seen a lot of progress though since I sent my recommendations. Review collected by and hosted on G2.com.

Data can be held on-disk (as well as in-memory) and Aerospike still performs incredibly fast on a single node. This is huge. I can't stress this enough. Review collected by and hosted on G2.com.

The Go library isn't well documented in some areas (especially errors), and while Aerospike's documentation is great, it still needs some work. Review collected by and hosted on G2.com.

easy to use. Not just saying, Aerospike is easy to config and setup.

active official forum support

support server-side script, like HBase co-processor

simple and stable

support TTL

support durable and non-durable

Using hash id as key distributes requests to the whole cluster and the cluster has NO HOT POINT. Adding capacity can simply resolves the throughput problem.

Not like redis, Aerospike supports multi core by default. Review collected by and hosted on G2.com.

the tool is pool, including the web-ui, command line.

more powerful administration support

cluster size is limited to hundreds.

all keys must be loaded to memory Review collected by and hosted on G2.com.

Having performance in mind I like the fact that despite its written in C it can be seamlessly used from the JVM which promotes its use in Big Data solutions. Review collected by and hosted on G2.com.

Key-Value databases are note the best fit for aggregation intensive requirements. Review collected by and hosted on G2.com.

Easy to setup;

Easy to scale;

Pretty well-documented;

Helpful support staff. Review collected by and hosted on G2.com.

No (easy) way to "auto-increment" any integer value.

No security at all on the Community Edition. Review collected by and hosted on G2.com.

Performance is as advertised. With Lua, the flexibility becomes limitless. After a while, everything just works alright in the background, and I do not have to bother the guys. Review collected by and hosted on G2.com.

If it crashes, I don't know how to recover from it. It crashed once, I think I exceeded the max size of a record Review collected by and hosted on G2.com.

The configuration enables you to choose the performance trade-offs that are acceptable. I was originally exposed to aerospike while working on a sub-second response time system (100 millis max). However, I used it a less-strict setting and it was still performant out of the box in testing (still haven't deployed to production yet). Review collected by and hosted on G2.com.

I like the custom query coprocessors, but I don't like how I have to write them in Lua Review collected by and hosted on G2.com.