Asymmetric Capacity Expansion for Travel Agency Booking

Problem:
An outsource provider of services for various travel booking agencies faced a severe capacity problem each year during the holiday season. Many of its customers wanted to go on vacation and were comparing rates and availability prior to actually booking a trip, generating on average over 700 “looks to one book”. All of this look-to-book activity caused query processing to increase from five to fifty times the normal query load, which greatly impacted the capacity of the HPE NonStop booking system. Recognizing that actual booking activity involved 80%-90% complex queries and 10%-20% simple writes, the provider wanted to replicate the booking system’s master database to NonStop and Windows target systems that support the travel agent queries.

HPE Shadowbase Solution: Asymmetric Capacity Expansions

Internet travel companies send transaction requests to book to the master-booking node in Figure 1. The master-booking node is hosted on an HPE NonStop server, which is connected to any number of scalable NonStop and Windows query nodes via HPE Shadowbase data replication. This read-only function is offloaded from the master-booking node, since travel agents and Internet travel companies use the query nodes to look-up vacancy/seat availability information, etc.

For a full image description, please see the paragraph that starts with "Internet travel companies."

Figure 1 — Asymmetric Capacity Expansion

  • The outsource provider’s application is configured (no application modification was required) to perform all queries against the query nodes, yet still apply the booking transactions to the master booking node.
  • At least one NonStop query node is architected to also act as a disaster recovery standby system for the booking application if that system fails.
  • Additional query node replicates can be added as the query volume increases.
  • The key structure on the master-booking node remains optimized for OLTP access (“skinny” keys), and the key structure on the query node is optimized for query (“fat” keys).
  • Data from the master-booking node is also replicated to Windows servers running SQL, which supports a large Operational Data Store (ODS) application.

Additional Benefits:

  • Each of the NonStop read-only query nodes acts as a disaster recovery backup for the NonStop master node.
  • The entire database resides on each of the read-only nodes, in essence allowing the provider to scale the application across multiple systems to handle unpredictable query volumes.

Please see the article in the Availability Digest for more information.

Please contact us with any questions on this HPE Shadowbase solution.


The above was adapted from the book: Breaking the Availability Barrier, Volume III: Active/Active Systems in Practice by Paul J. Holenstein, Dr. Bruce Holenstein, and Dr. Bill Highleyman.