More and more I hear companies looking into replacing expensive licensed database options with more cost-effective alternatives, for various reasons: simple cost reduction, or avoiding lock-in, or because of other strategic considerations. In order to kickstart the creative thinking, here are a few hardware and software based methods to achieve cost reduction in database options (on top of Oracle Enterprise Edition). I am assuming the base product stays the same (Oracle Enterprise Edition) as replacing EE with alternatives can be very cost effective – but requires serious changes in architecture and management so I’ll leave that for another time.
- Advanced Compression Option – When used for OLTP compression, note that modern All-Flash Arrays (like DellEMC XtremIO) and even some hybrid (flash-based) arrays (like DellEMC VMAX) are capable of inline data compression. In contrast to popular belief, OLTP compression is not really compression (you could regard it more like deduplication within a database block) and the result is that even OLTP compressed tables will usually see further data reduction based on storage compression – but more so if OLTP compression is disabled. As an interesting bonus – storage compression compresses EVERYTHING (including temp tables, indexes, empty database blocks, redo logs, …) and not just the table data. Although the CPU overhead of compression/decompression is minimal, eliminating expensive licensed CPU cycles from the compression/decompression process is another nice bonus. Further reading: Lab Report: Oracle Database on EMC XtremIO. A Compression Technology Case Study. Cost reduction: $11,500 per processor (24% of EE license – 2 x86 cores, list price)
- Active Data Guard – Although DG is a convenient method for replication of a single database, there are some gotcha’s as well (force logging performance overhead, and business consistency issues across multiple databases, middleware and apps to name a few). Data Guard is free but many customers opt for the licensed Active Data Guard option due to being able to use the standby databases for additional purposes. Replacing (Active) Data Guard with some kind of infrastructure replication (SAN based or storage based) can solve business consistency issues, simplify Disaster Recovery procedures and provide better D/R testing due to making separate snapshots of the replicated data (rather than work directly on the D/R data itself). As a nice bonus – the CPU and network overhead on the database server is moved to other layers in the infrastructure stack. Further reading: Data Guard or Storage replication. Cost reduction: $11,500 per processor (24% of EE license – 2 x86 cores, list price, both primary and standby server).
- Oracle RAC (Real Application Clusters) – RAC is often deployed as a “must have” if you need high availability. In my experience however, 98% of all database workloads could happily take an outage of less than 5-10 minutes in extreme scenarios (i.e. CPU failure and the like). Replacing RAC with 3rd party clusterware (i.e. Veritas Cluster, VMware HA/DRS, Linux native clustering or other) simplifies database deployments, eliminates RAC overhead (which is usually pretty high and burns expensive CPU without improving transactional performance), reduces requirements for frequent tuning and optimization and sometimes improves availability (by avoiding RAC configuration problems and clusterware bugs). If you are convinced you can’t live without RAC for a certain application then, IMHO, you should also deploy active/active stretched RAC clusters as standard RAC only protects against server failures, not against datacenter disasters. Further reading: Database Virtualisation: The End of Oracle RAC? and You Probably Don’t Need Rac. Cost reduction: $23,000 per processor (48% of EE license – 2 x86 cores, list price)
- Oracle Multitenant (aka Pluggable Database). Often marketed as a consolidation option (to reduce TCO), it seems counter-intuitive to spend a large amount of extra money just to be able to run more than one database on a server. Be aware that Pluggable Database is mandatory with the latest Oracle releases, however, you don’t need a multitenant license as long as you run one pluggable database per container. Alternatives: Running multi-instance, schema consolidation, and my all-time favourite: One database per virtual machine utilizing server virtualization (i.e. VMware or other hypervisors). Additional bonus: No inter-depencendy of databases, better security, better availability, easier recovery, easier upgrades, no “noisy neighbor syndrome” (cross-db performance influence). Further reading: What’s All the Buzz Around Oracle Multitenant and Five reasons to choose VMware vs Oracle Multitenant. Cost reduction: $17,500 per processor (37% of EE license, 2 x86 cores, list price)
- Database In-memory – An interesting option to speed up large queries. One of the goals is to reduce I/O by keeping large amounts of data in server (SGA) cache and store it in columnar format. If you could massively boost the I/O bandwith and lower response times so that the CPU always gets saturated with data, then a large benefit of in-memory is gone (the columnar store is still beneficial so mileage may vary). New storage protocols like NVMe have the potential to provide 10’s of gigabytes per second per server at a latency of far less than 100 microseconds per IO (expect 30-50 microsec). NVMe PCIe cards are already available but have the problem that it’s local (JBOD) “storage” and has no means of data protection. DellEMC announced NVMe support on existing platforms (XtremIO, VMAX, …) so you can have the best of 4 worlds (shared capacity, extreme performance, data protection and data replication) instead of one. Further reading: Extreme Performance in a Shared Storage Environment (note that the product mentioned was cancelled and the intellectual property will re-emerge in other DellEMC platforms). Cost reduction: $23,000 per processor (48% of EE license – 2 x86 cores, list price)
- Golden Gate – GG is a database replication product that can replicate table data bi-directional across heterogeneous databases (different versions and even different database products). There exist good alternatives with equal features at a fraction of the cost. If you run or consider Golden Gate, you should also seriously give some thought to the alternatives, such as Quest Shareplex, DBVisit, Informatica Data Replication. Further reading: Top Five Reasons to Choose SharePlex over Oracle GoldenGate. Cost reduction: $17,500 per processor (37% of EE license, 2 x86 cores, list price, both primary and target servers)
Getting TCO under control and avoiding lock-in requires an open mind and an appetite for learning about available alternatives. At DellEMC we can provide some but I also pointed out some other products and technologies. There’s much more out there. Google is your friend! But be aware of markitecture, FUD, myths and Oracle “cult”.
This post first appeared on Dirty Cache by Bart Sjerps. Copyright © 2011 – 2017. All rights reserved. Not to be reproduced for commercial purposes without written permission.
5,101 total views