According to Internet and cloud intelligence company Thousand Eyes' principal solutions analyst, Mike Hicks, instead outages for Australian ISPs showed peaks toward the end of February before dropping through March and April and rising to February levels toward the end of May - and returning to typical levels through June and July.
“The characteristics of the outages would indicate that these outages were pre-emptive traffic engineering exercises in expectation of the pending lockdown orders,” says Hicks.
Hicks comments follow ThousandEyes publication of its inaugural 2020 Internet Performance Report, a “first-of-its-kind” study of the availability and performance of Internet-related networks, including those of Internet Service Providers (ISPs), public cloud, Content Delivery Network (CDN) and Domain Name System (DNS) providers.
“As in the rest of the world, Cloud Service Provider (CSP) outages observed in ANZ were far less prevalent than outages in the ISP networks, but the pattern for CSP outages in ANZ when compared to ISP outages was slightly different with the peak in outages coinciding with the lock down orders for March, with a 400% increase in CSP outages when compared to February,” Hicks says.
“Mirroring the outage pattern for ISPs, the numbers dropped through to late May, with a small peak seen in early June, and returning to normal levels in July.
“The difference in outage pattern, between ISPs and CSPs in Australia, is likely attributed to the fact that many cloud providers also operate bespoke, software-defined networks purpose fit for their needs, allowing them to innovate and address issues faster, and implement changes as demand increases.”
According to Hicks, the timing of an outage is one of the most important factors in determining its impact on users.
“Outages in ANZ were more indiscriminate than experienced in other regions, with fewer outages taking place within traditional non-peak (e.g. early morning) hours, and a higher concentration taking pace throughout business hours and early evening.
The average duration and frequency of the outages, which were potentially disruptive to users, was low in comparison to other regions and so any impact would have been low,” Hicks added.
Key APAC findings of the ThousandEyes report are:
- Internet Service Providers - Weekly network outage patterns in Asia-Pacific share some similarities with North America, with sharp increases and decreases throughout the examined period, but higher than normal concentrations in March and parts of April.
- Outage Impact on Infrastructure Availability Varied - Normalisation based on infrastructure coverage and impact also contextualises outage numbers. Asia-Pacific is also shown to have suffered more availability damage than North America and EMEA in the first half of 2020. Despite higher numbers of disruptions observed in North America, ISP networks in that region are not “less available” than those of Asia-Pacific and EMEA.
Disruptions Due to Increased Traffic Engineering
- Network outage duration followed a slightly different regional pattern, where outages in North America and Asia-Pacific outages were, on average, longer in April, while in EMEA outages were longest in February and May.
- Long lasting outages that involve more than a handful of network interfaces may be the result of planned downtime or the unintended result of network changes.
- Network operators assess service requirements based on known or predicted traffic patterns. In the case of the “known” this is typically based on the physical location of the user population, pattern of usage, along with the type of services accessed. The “predicted” is typically based on foreseen events, such as seasonal holidays, major sporting competitions, streaming game releases, or shopping rituals. Given the unprecedented scenario network operators faced in the wake of the COVID-19 pandemic, any assertion of known and predicted traffic usage was lost. This meant that any provisioning or changes required were reactive, potentially increasing the likelihood of them disrupting service availability.
ThousandEyes says that user impact varies by region, and outages in Asia-Pacific were more indiscriminate from a time of day standpoint - and fewer outages took place within traditional non-peak usage (early morning) hours, with higher concentrations taking place during business and early evening hours.
“Given time of day occurrence, outages in Asia-Pacific, though on average smaller and shorter in duration than other regions, appear to be more disruptive to local users. However, these disruptions would be brief and likely impact fewer users,” ThousandEyes says, noting that
- Looking at the number of outages occurring within traditional business hours (Monday to Friday, 9AM to 6PM local time) in relation to outages outside of that window (i.e. weekends and non-business weekday hours), far fewer impact typical business operations than the total number observed for each region.
- While the Asia-Pacific region had more monthly variation in user impact between January and July, 2020, only approximately 29% impacted business users overall, largely due to a higher volume of outages taking place on weekends versus EMEA. Given that EMEA outages involved more infrastructure and were longer lasting, when they did take place during business hours, their effect would have been more profound than those in Asia-Pacific.
Noting that “not all cloud regions are equally stable”, ThousandEyes says that despite a smaller, more homogeneous set of operators, cloud provider network resiliency varied considerably depending on region, with more outages in EMEA and Asia-Pacific than North America.
“The distribution of these outages throughout the first half of 2020 fluctuated independently of one another and don’t appear to be as heavily influenced by traffic and usage increases reported post lockdowns in most regions,” ThousandEyes said.
"The Internet is inherently unpredictable and outages are inevitable even under normal conditions. However, with the overnight transition to a remote workforce, remote schooling, and remote entertainment that many countries experienced in March, we saw outages spike to unprecedented levels -- especially among Internet Service Providers who seem to have been more vulnerable to disruptions than cloud providers," said Angelique Medina, research author and director of product marketing at ThousandEyes.
"With the Internet Performance Report, businesses can benchmark Internet performance pre and post COVID-19 and plan for a more resilient IT environment as they continue to build out infrastructures that can manage the external dependencies on cloud and Internet networks that employee and consumer experiences now rely on."
Key findings from the ThousandEyes 2020 Internet Performance Report: COVID-19 Impact Edition, include:
- Global Internet disruptions saw an unprecedented rise, increasing 63% in March over January, and remained elevated through the first half of 2020 compared to pre-pandemic levels. In June, 44% more disruptions were recorded compared to January.
- ISPs in North America and APAC experienced the largest spikes in March at 65% (North America) and 99% (APAC) respectively versus January, and have since returned to levels typical of those regions. In EMEA, however, outages continue to increase month over month with 45% more disruptions in June versus January.
- ISPs were hit the hardest, while cloud provider networks demonstrated greater overall stability. Between January and July, cloud providers experienced ~400 outages globally versus more than ~4500 in ISP networks. Relative to total outages, more than 80% occurred within ISP networks and less than 10% within cloud provider networks.
- Though the total number of outages increased across all regions, impact on Internet users varied. Following pre-pandemic patterns, a larger proportion of disruptions in EMEA tend to occur during peak business hours as compared to North America, where a majority of large outages typically take place outside of traditional business hours and therefore may not have a meaningful impact on Internet users.
- Overall, the Internet held up. Despite unprecedented conditions and an increase in network disruptions, Internet-related infrastructures have held up well, suggesting overall healthy capacity, scalability, and operator agility needed to adjust to unforeseen demands. Negative performance indicators, such as traffic delay, loss, and jitter generally remained within tolerable ranges, showing no evidence of systemic network duress.
- Increased network disruptions due to operator adjustments. Many of the network disruptions observed post-February appeared to be related to network operators making more changes to their networks to compensate for changing traffic conditions.