Nomorerack Achieves 600% Online Sales on Cyber Monday with ClustrixDB, Scales from 48 to 112 Cores to Handle 3x Database Traffic
Nomorerack is among the fastest growing e-commerce companies in US offering deals at deep discounts. Facing a large spike in traffic for the Black Friday and Cyber Monday sales, Nomorerack came to us looking to scale their database to handle the extra load from the herds of sale-hungry holiday e-shoppers. As an AWS customer, Nomorerack is used to on demand capacity. The DBaaS solution from Clustrix is a perfect fit for their relational database scalability needs as it provides flexible scale-up and scale-down. Nomorerack chose Clustrix DBaaS on ClustrixDB Appliance hosted at Equinix (same Data Center as AWS) that is connected to application servers running in AWS via AWS Direct Connect. Nomorerack benefits from the dedicated performance of the Clustrix appliance and networks with the flexibility to add capacity on demand, all the while running their application layer on AWS. A step above in performance and resiliency compared to ClustrixDB on AWS instances.
Provisioning for Expected Growth
During the holiday period, Nomorerack expected increased traffic, but did not know exactly how much. They informed us that marketing impressions would be increased 20-fold, with an expected click-through rate of 0.35%. Using only these two metrics compared with statd (ClustrixDB statistics) data from a previous marketing campaign, we projected this year’s holiday peak to reach 370% the visitors casino online compared to a regular day, which would translate to a theoretical 170% higher CPU usage on the database cluster. Because ClustrixDB scales exceptionally well with higher concurrency, we recommended adding 8 new nodes (64 cores) to the existing 6-node (48 core) cluster (133% added capacity). The 14-node (112 core) cluster would have Nomorerack running smoothly through the two big sales events, a welcomed assurance for the fast growing online retail company.
And smooth holiday shopping it was! Nomorerack reached record high sales on Black Friday and saw even higher sales on Cyber Monday. The database handled the load without any issues. Traffic reached 45,000 TPS with 60% CPU usage on Friday, and 65,000 TPS with 84% CPU usage on Monday. Prior to that, the highest peak was 22,000 TPS. This year, an over 3-fold increase was handled with ease.
Thus, with 2.33x more hardware, ClustrixDB was able to handle 3x the load from 22K TPS to 65K TPS, providing better than linear scaling. The graph below clearly shows this: green is average CPU (right axis), blue is Queries Per Second (QPS, on left axis), and the “X” marks the cluster expansion from 6 to 12 nodes on the 21st, and from 12 to 14 nodes on the 27th. It is immediately clear that cpu usage is cut in half when cluster size is doubled on Nov 21. You can also see the CPU usage move along linearly as TPS increases.
(time is UTC)
In summary, Nomorerack was able to handle a 3x spike in database load by simply adding nodes to their cluster, a very straightforward operation. All we needed to do is estimate how many nodes were required to handle the extra load and add them during off peak hours. No complicated sharding, read-slave replication setup, or load balancing required. They can now choose to downsize the cluster to handle their normal workload and go back to business as usual during the off-season.