Putting the Survey Results in Context: Webinar Discussion
We had a great discussion with Alex Gorbechev, CTO at Pythian, Tim Coats, Innovation Lead at Trace3, and our own Varun Singh, founder and CTO at ScaleArc on the results of our recent survey “The State of Application Uptime in Database Environments.” The survey canvassed more than 200 enterprises on their database infrastructure and the impact that planned and unplanned database downtime is having on companies’ critical applications.
Each of these panelists brought a lot of customer experience to their perspective on the survey results. Among their more interesting observations:
- low use of cloud: Alex observed the survey data, where 40% of respondents said they don’t use the cloud at all, is consistent with his clients’ usage. He noted his surprise, though, that 15% reported they run in hybrid mode, with apps in the cloud and data on prem. He called out that this deployment is complex to architect, given latency concerns.
- use of secondaries in AlwaysOn: Tim wondered at one of the data points – almost 10% of AlwaysOn users reported not having any secondaries. Who deploys AlwaysOn without secondaries, he wonders! Varun chimed in that ScaleArc occasionally sees customers who haven’t yet taken advantage of the read-only secondaries in AlwaysOn because their apps can’t direct queries appropriately. Automatic read/write split, of course, with no code changes is one of the primary reasons customers deploy ScaleArc to help their SQL Server 2014 upgrades.
- duration of unplanned outages: all the panelists talked about the trade-offs customers make between speeding up a recovery with automatic failover and wanting to take the time to investigate data integrity. For Oracle RAC customers, who had more 1-3 hour outages than the general survey population, Alex speculated that issues with storage were probably the culprit behind the longer time offline.
- frequency of maintenance windows: with almost 35% of respondents noting they perform maintenance one or more times a week on their most critical apps, the full set of panelists talked about the increased frequency of security patches these days. Alex noted that increasingly patches affect multiple elements in the application stack, from the OS on up, so he’s not surprised customers have to perform maintenance so often. The key is to adopt technologies, like database load balancing software, that let customers perform those maintenance tasks without taking down the app.
- developer involvement in maintenance windows: for all respondents, survey results show developers involved in maintenance windows more than 50% of the time. All panelists thought this rate was fairly high and not an optimal use of developers’ time. But when you look at more sophisticated MySQL deployments – with Master/Master or Galera setups – that figure drops, to about 40% for Master/Master and just over 25% for Galera. Why the drop? Tim noted that these implementations require really sophisticated DBAs, so they likely have less of a reliance on developers. He also noted that the apps don’t need to know about the database infrastructure for those deployments, which also reduces dependency on developers.
- deferred security patches: all the panelists were surprised that enterprises are putting themselves at risk so much, with 60% of respondents noting they have deferred or delayed applying security patches because of concerns over downtime. Alex noted the contrast of this data with other trends he’s seeing around moving databases behind firewalls because relying on privileged access hasn’t been sufficient to protect the database. Varun added that he’s also seeing theft of data reads as one of the big sources of today’s hacker attacks and noted that most database analytics tools do not log read access, only writes. The ScaleArc analytics engine logs every query, he contrasted, instead of just sampling data, and it logs both reads and writes.
Check out more resources about the survey results here, including the webinar recording, the survey report, an infographic of the key findings, and the full data set of responses.
comments powered by Disqus