Cloud upgrades no big deal if deployment is done right

2 Comments

Credit: CarbonNYC [in SF!]

When Verizon said it might shutter its new Verizon Cloud for 48 hours this weekend for a major upgrade, users were shocked. Some promised to live-tweet the event which starts Saturday at 1 p.m. EST. Verizon’s subsequent qualifications — that 48 hours is the worst-case scenario, that customers received ample notice, that this work would make future upgrades less traumatic — eased the situation a bit.

Cloud vendors take these occasions — whether the issue is a planned-in-advance upgrade or more hastily put together patch  — to educate users about best deployment practices. They tick off a range of tips — that workloads should be deployed across availability zones and regions, for example. And third-party tools and management vendors likewise parlay these events to promote the use of their own products.

To mitigate downtime, plan, plan, plan

Cedexis, which offers a cross-cloud load balancing service, sent out an email Thursday with its own list of best practices. From the email:

Whatever your opinion on Verizon Cloud and the way they are rolling this upgrade out, what is true is that there is no reason for this type of system maintenance to impact the correctly configured enterprise. It is time for architects and designers to realize that cloud outages are a fact of life — just like Data Center outages.

In cloud, as in past IT deployment models,  disaster recovery “relies on the use of geographically diverse deployment of applications. Why would anyone adopting Cloud think single-homing an application is a reasonable practice?,” according to Cedexis.

Deploy across zones, regions and vendors

Cedexis goes further than the multi-availability zones mantra to say enterprises should use multiple cloud vendors as well. That’s something you probably won’t hear from [company]Amazon[/company] or [company]Microsoft[/company].

Vendor specific outages — whether the vendor is Amazon Web Services, Microsoft, Rackspace, IBM or Verizon — are “more common than ‘acts of God,'” according to Cedexis, so selecting multiple vendors with (of course) global load balancing protects the user from these events. [company]RightScale[/company] is another vendor who provides these cross-cloud management and deployment capabilities.

So go easy on [company]Verizon[/company], said Carl Brooks, an analyst with 451 Research. In a comment on Gigaom’s earlier story, Brooks wrote that the hubbub around Verizon’s planned upgrade was way overblown.

“I’d bet real money they could do most of this upgrade without kicking users off … [company]Microsoft[/company] has planned monthly outages to patch and tells you if you’re in the target zone ahead of time. This isn’t so different, just a bit more ham handed. You know what AWS says to its [most of its] users before it upgrades and reboots a bunch of its platform? Nothing.” Brooks later amended his statement via email to say that most customers never notice routine AWS maintenance.

Update: A Verizon customer pointed out that at least one Verizon Cloud (compute and storage) shut down in September did, in deed, last two days. It kicked off at 11 a.m. CT on September 11 and services were not back on till 11:38 p.m. two days later, according to this Verizon support post.

Note: This story was updated at 1:43 p.m. PST with information on Verizon’s September 11 shut down and againa t 7:19 p.m. PST January 9 with Carl Brooks’ amended statement.

 

2 Comments

Keith Craig

Uptime is the cloud user’s gold standard, valued more than processing power, memory allocation, I/O bandwidth and anything else that affects cloud performance.

Many cloud hosts provide solid but not constant uptime (check CloudHarmony.com for its sharp benchmark testing); however, no cloud host can ensure constant uptime. Too many variables can throw a wrench into the works. Something’s going to happen someday.

Consequently, the downtime-response plan recommended by Cedexis is spot-on. Always have at least two contingency plans ready to bootstrap in case of sudden – or planned – downtime. Plan 1 might deploy your servers across cloud-hosts and zones and incorporate load balancing to distribute traffic evenly. Plan 2 – call it the “peace of mind” contingency – is to enlist a cloud-host’s managed services team. That way, if downtime occurs, the cloud-host’s managed services team can address the situation immediately, giving the user lots less to worry about.

And always, always, always, always, always perform backups.

John Lyons

A 48 hour outage is absolutely unacceptable in any environment, Cloud or not. Also, from our experience, the final comment regarding AWS upgrades is completely untrue. In the past 18 months we’ve never once had an AWS “upgrade” which we weren’t informed about well in advance.

Comments are closed.