Hosting uptime claims verifications?

July 14, 2004 5:26 PM

Recommendations, please, for verification of uptime claims. The company in question runs a large server farm that hosts sites it creates for a niche industry. Its customer base is comprised of technically clueless unsophisticated folks, so questioning their references is useless. Are there other ways to verify a claim that a certain domain or IP hasn't had a single service interruption in several years? Or at least verify whether they're delivering 99.9% uptime?
posted by nakedcodemonkey to computers & internet (15 comments total)

Netcraft monitors uptime historically. Enter in your hostname, like Dreamhost.com, to see their uptime.
posted by waxpancake at 6:05 PM on July 14, 2004

No one really delivers outage-free service. Try asking the company to describe their last major outage.

Around six months ago, there was a brief power outage in California that managed to take out the fiber cross-connect switch in Los Angeles's One Wilshire complex. Although power came back online quickly, the fiber switch did not. It took a long time for companies that relied on connectivity via One Wilshire to recover. Since One Wilshire is the major meeting point for SoCal connectivity, the resulting telcom outage hit most of California for the better part of a day.

Similarly, the big east coast outage last year must have had far-reaching consequences.

A compitent colocation provider will be happy to talk about the utility failures they have endured, and what steps they have taken to mitigate future outages. Anyone who claims outages never happen either hasn't been in the industry long or is willfully dishonest.
posted by Kwantsar at 6:22 PM on July 14, 2004

No one really delivers outage-free service.

Yep. Which is why I'm trying to get independent verification of a real number.

Try asking the company to describe their last major outage.

...except they say there's been none to describe.

From their perspective, uptime is a measure of the server's stability, not the site's response. The claim being made is that the multiple redundant servers have stayed up continuously for X number of years without so much as a short glitch. All that's required for this project is at least 99.9%, but an eyebrow-raiser like this sets off alarms.

(Thanks for the thought, waxpancake, but unfortunately Netcraft lacks any uptime record for this company. It's a niche player, not a major mainstream provider like dreamhost or pair.)

Any other options?
posted by nakedcodemonkey at 6:57 PM on July 14, 2004

It's easy to tell who is offering 99.9% availability: they're ridiculously expensive beyond your wildest dreams.

The company I work for spends millions of dollars every year on keeping its web site available. We achieve about 98%.
posted by majick at 7:01 PM on July 14, 2004

%99.9 is not good. 8760 hours in a year 1/1000 = 8.76 hours downtime a year. It's "good enough".

As for the claim of servers having multi-year uptimes thats very believable Ive seen production servers with uptimes of 5 years and more. A good system admin can do that. Doesnt mean the rest of the network or applications were running solid though. It's just used car salesman talk.. it's like saying the pistons in your car will last for 500,000 miles.. probably true.
posted by stbalbach at 7:16 PM on July 14, 2004

Are there other ways to verify a claim that a certain domain or IP hasn't had a single service interruption in several years? Or at least verify whether they're delivering 99.9% uptime?

You shold take a second to note how far apart these two things really are.

99.9% uptime means that it's okay for the server to be down about an hour and forty minutes every week. This "uptime" percentage is a bullshit figure bandied about for marketing purposes. What you really want to know is how sophisticated the host's platform is, how redundant. The length of the average outage, time since the last one. Those will give you a better sense of what it will really be like to depend on them.

I had horrible downtime issues with an old host. Multiple lengths of 45 minute outages all the time. When I called them on it, they said they were still well within their 99.9%, and they were right!
posted by scarabic at 7:18 PM on July 14, 2004

99.9% uptime means 0.001*7*24 = 0.168 hours downtime per week, or about 10 minutes. scarabic, you were lied to.
posted by gleuschk at 7:38 PM on July 14, 2004

I've seen it too (100%) - if the site is architected correctly, then it can be done.

Another way to keep them honest is through SLAs and your own monitoring of their website.

gleuschk - check your numbers - 99.9 means .01*7*24=1.68 hours, or about 1 hour 40 minutes. Scarabic is right.
posted by drobot at 7:51 PM on July 14, 2004

1-.9 = .1
1-.99 = .01
1-.999 = .001

right? the birthday cocktails haven't destroyed my arithmetic completely, have they? no, google backs me up.
posted by gleuschk at 7:58 PM on July 14, 2004

I think some people are talking about different things here - are you looking for a *hosting* company? or are you just checking up on some websites uptime? It sounds like you've identified a company with a web app that you want to evaluate. Right?

I would look at where their webfarm is located - if they are with a good colo (Savvis, AT&T, etc.) with redundant power, connectivity, etc., and their apps are on a stable platform, and they have redundant servers with redundant storage (disk arrays), processors, power supplies, and they have load balanced or otherwise redundant databases backed up at timely intervals, then it's not that far fetched to assume that they are telling you the truth that they haven't had a major outage.

gleuschk - duh, you are right. Sorry.
posted by drobot at 8:04 PM on July 14, 2004

Another way to attack this would be to look at their architecture and identify for yourself the points of failure and ask them what they would do if that piece failed. Ask to see their Disastor Recovery Plan and Scalability Plan. Find out their backup schedule and ask to see documentation. In the absence of real uptime statistics, at least you can evaluate their expertise with keeping their operation running. Even if its true that they've had 100% uptime, you might not be so confident in them if its only because of dumb luck.
posted by drobot at 8:09 PM on July 14, 2004

Sorry, one more suggestion - ask them to see documentation indicating that they've been up 100% - what do they use for monitoring and alerting, and what do they measure to determine that they're 'up'.
posted by drobot at 8:10 PM on July 14, 2004

The claim being made is that the multiple redundant servers have stayed up continuously for X number of years without so much as a short glitch.

If they have multiple, geographically- and topologically-dispersed servers, this is reasonable. It doesn't mean that any given service hosted on their servers will have 100% availability to customers, though, since failover isn't perfect. Some services (e.g. DNS, NTP) can fail over more seamlessly than others (e.g. web, shell hosting).
posted by hattifattener at 9:53 PM on July 14, 2004

drobot, great suggestion.

To those who offered to explain the math and clarify other misc points: yes I do know what 99.9% translates to, of course that's a generously low standard, and no this isn't a shopping trip. The host comes pre-selected, not by me. Ditto for the uptime minimum. A number of the company's other claims, not related to the servers, were demonstrably over the top though harmless enough to let pass. However, drastically overstating the server's stability would NOT be harmless, thus the extra caution.

Thanks for the help, everyone!
posted by nakedcodemonkey at 11:17 PM on July 14, 2004

100 - 99.9 = .1

But as a percentage, I guess that's .1/100, or .001

Never let an English major do the math! :\
posted by scarabic at 2:19 AM on July 15, 2004