Network Uptime?

[James]

Is it really possibly to guarantee five nines uptime? I was visiting a prospective client of ours site and reading a few of their promotional pages when I came across this statement:

"Company X guarantees that the network will be available 99.999% of the time in a given month (no more than 24 seconds downtime per month), excluding scheduled maintenance. Company X will refund the customer 5% of the monthly fee per additional 30 minutes downtime (up to 100% of customer's monthly fee). Network uptime includes functioning of all network infrastructure including routers, switches and cabling. Network downtime exists when a particular customer is unable to transmit and receive data and Company X records such failure in the Company X trouble ticket system. Network downtime is measured from the time the trouble ticket is opened by a customer to the time the server is once again able to transmit and receive data."

Is this a brave statement to offer, or has someone already worked out that if they fail to make the 5 nines the financial downside is worth the PR risk?

[Jaiem]

IMO it's probably more marketing hype than reality. 99.9% is standard and pretty much accepted as the best that's possible.

Now, that they gave an actual time (24 sec) is more interesting.

[James]

24 secs of down time per month! I wish ;)

[JTY]

I bet their service comes at a premium though.

[tabernack]

You can guarantee any % you want for uptime, as long as you have a plan to compensate your clients if you fail to provide this.

For example, verio's datacenters provide a %100 uptime guarantee. They will obviously one day experience some sort of problem that will force them to compensate their clients.

[Jaiem]

Except for a few celebrated cases most hosts have a good uptime record.

Keep in mind the technology isn't proven. Eventually even Verio will have some problem too.

[akashik]

I remeber getting some junk mail a while ago was explaining how I could get 100% uptime. Naturally I had to buy their services *smirk*, but it involved having multiple mirrors of everything (and I mean EVERYTHING) set up around the world at different NOC's on different providers. They themselves had a redirection type thing set up that if it detected a server down somewhere would automatically reroute to a mirror, or another one again if that was down too.

There own system was set up the same way to prevent their own system breaking down... I'm sure you're beginning to imagine the hoop jumping involved in this and the extra costs.

In general 99% is pretty easy to maintain if you've done the right research. 99.9% is harder but not impossible. Guaranteed 100% is next to impossible without some interesting solutions.

Everything in that first paragraph would have apparently booted our uptime slightly more than 1% on what it is now.. Worth the money? Very doubtful...

Greg Moore

[Phoenix]

Is it really possibly to guarantee five nines uptime? I was visiting a prospective client of ours site and reading a few of their promotional pages when I came across this statement:

"Company X guarantees that the network will be available 99.999% of the time in a given month (no more than 24 seconds downtime per month), excluding scheduled maintenance.


It is possible-just not easy or cheap-but that company has given themselves a big loophole- excluding scheduled maintenance. And if a company is unscrupulous enough, any and all outages can be explained away as scheduled maintenance, regardless of what the real cause of them was. It also depends on how they are calculating their uptimes. Daily uptimes can be averaged, data can be massaged, etc.

We do HA hosting, and have for years (which is a bit different and requires adherence to certain industry standards uptime and SLA's) our network uptime was at 100% and our overall uptime was 99.999% for a year) for a full year-we did a press release and mailing about it a year ago. But, an HA SLA doesn't allow for any exclusions and our monitoring system records all outages, so we took a big hit in our percentages when we upgraded our routers a few weeks ago.

It took our ***ulative monthly network downtime to 99.89%, and our ***ulative monthly overall reliability went down to 99.98% The only thing that saved us from complete disaster is that it was done in short bursts at ungodly hours of the night to minimize the affect on the customers. So we are only hanging on to our Class IV HA status with the skin of our teeth for the next few weeks, but now that the router upgrade is done, we don't have any network-wide upgrades to do for a while.

So, to answer your question, it is possible to guarantee 5 Nines, but it is not likely that what this company is providing is true Class V High Availability Hosting backed by an SLA without loopholes, unless Company X is Conxion, and it's their FailSafe hosting-which is what all HA hosts aspire to. And if so, the hosting fees will be very steep.

(NOTE: apparently the censor is active on this UBB installation-what it is asterisking is the word c-u-m-u-l-a-t-i-v-e).

[James]

Originally posted by Phoenix


So, to answer your question, it is possible to guarantee 5 Nines, but it is not likely that what this company is providing is true Class V High Availability Hosting backed by an SLA without loopholes, unless Company X is Conxion, and it's their FailSafe hosting-which is what all HA hosts aspire to. And if so, the hosting fees will be very steep.


COMPANY X MIN SPECS AND PRICE
750 Mhz Processor
256MB RAM
20GB EIDE Drive
10GB/Month Burstable Bandwidth

24x7 Support
Windows 2000 Preinstalled

Monthly $409.33

[Phoenix]

That's pretty cheap money for 5 nines network uptime-I'd sure like to see their network architecture and see just how it is that they can guarantee 5 nines network uptime plus give you a dedicated server for that low a cost.

Beyond just the technical issues and the pricing, their wording of the guarantee is tricky. They tell you what 5 nines is defined as (24 sec of downtime per month). Then they tell you that after a half hour of additional downtime, (which comes out to ~99.95% if that's all the downtime for that month) they will refund 5%. To get a full refund of your monthly fees, the routers/switches/cabling need to be entirely non-functional for 10 straight hours (which is below 99% if that's it for the month). And the clock doesn't start ticking until after you open a trouble ticket.

So, what they are really guaranteeing is far less than the 5 nines they are using to entice you. They are guaranteeing ~99.95% And there's a big difference between that and 99.999%.

I'd be interested to see if they provide live uptime stats or if your uptime stats are based on when you find out the site is down and call the ticket in.

[James]

Nice one Phoenix!

Your right I've been back to their site and when you look at the small print they're actually only guaranteeing 99%.

Nice bit of marketing though ;)

BTW: Since I own the domain www.fournines.com I'm pretty pleased :)

[akashik]

Originally posted by James Cross


Nice bit of marketing though ;)



Was just thinking to myself - maybe this will be the new 'unlimited' blah sites use now the U word is falling out of favour...

Greg

[James]

i know that every magazine I've picked up lately has an ad for Microsoft which has 99.999 in big letters all across it.......


So you might be right

[Phoenix]

Originally posted by James Cross
I've been back to their site and when you look at the small print they're actually only guaranteeing 99%.

Nice bit of marketing though ;)



It is indeed excellent marketing. For in marketing (customer) perception = (market) reality, and they got you to perceive that they were going to give you guaranteed 5 Nines service, when they were only guaranteeing 2 Nines. And what they are charging is a bit much for 2 Nines.

Had you not questioned it and shared it with the group for picking apart, they might have made the sale, and somewhere along the way you'd have found out the truth about their guarantee the hard way.

We went from using the HA Classes to using Nines on our site to leverage the marketing money that Microsoft is spending to promote their server uptimes (does anyone out there actually believe MSFT's marketing?), but I personally don't like it, and after a few more months of tracking, we may be making our guarantee a flat Class IV or 100% refund.

Ours is currently tiered based on monthly c-u-m-ulative (#@!! censor) stats and requires only that a customer call in a trouble ticket for that outage, and then email us to request a credit. We are seeing how it goes, the SLA is only a month old, as we do get a lot of mistaken outage calls when a customer can't access their site because they or their customer's third party Internet access isn't working, problems on their internal network or computer, pilot error, or once-my right hand to God on this one-because they couldn't print their email from Outlook (and restarting the printer fixed their problem-say it with me: PEBKAC!). Their assumption is always that our web or mail server's down, and that's usually not the case. Hopefully, having our live uptime monitoring made public will help minimize false credit requests for an inability to access a site that had nothing to do with our network or our servers being down.

[thewitt]

Here's an interesting uptime guarantee:

The lead in page says:

With 30 days unconditional money back and guaranteed 99.9% uptime, you can rest assured that the quality of all our products is nothing short of astounding.

The Contract, which I did not have to acknowledge or sign to actually purchase my service from them, then adds the following for the uptime guarantee:

99% Uptime Guarantee: The uptime guarantee pertains to network uptime only. It does NOT apply to any of the services on the server, or the server itself.

My site experienced about a 20 hour downtime episode because the /var partition filled up on a shared server, on a weekend where there was no helpdesk coverage (their site claims 7x24) and no one answering the 800 number. mySQL driven sites were completely offline for the duration, of course this does not count as "network" uptime, so I was not covered.

In addition, this hosting company is in the VDI NOC, so they suffered some network downtime with the VDI problems of the past 4 weeks. I'm still waiting for my refund as part of the VDI network downtime issues, however I'm not holding my breath. I actually do not expect anything in the way of compensation - there is certianly something here that they will wiggle out of again. I have not received an answer to my request for guaranteed network uptime compensation. I suspect they will ask me for trace logs or something that covers the incident, it's start and end times, etc, in order for me to prove that there was a network problem that they are liable for.

Beware the guarantee - it's only worth as good as the integrity of the company behind it. If it's marketing hype, you'll likely never see it actually implemented.

-t

[Jaiem]

Sounds like they give with one hand and take away with the other.

[Phoenix]

Originally posted by thewitt
My site experienced about a 20 hour downtime episode because the /var partition filled up on a shared server, on a weekend where there was no helpdesk coverage (their site claims 7x24) and no one answering the 800 number. mySQL driven sites were completely offline for the duration, of course this does not count as "network" uptime, so I was not covered.


20 hours of downtime ????

Our network admin was ready to slit his wrists over a 15 minute outage on one of our shared servers last month. And he's still sulking over the 30 minutes or so the network was down when he upgraded the memory in the routers. He had a perfect record for a full year, and hates seeing a drop in his nines.

In defense of SLA's that only guarantee network availability for shared hosting environments, all it takes is one person with a runaway script or who isn't closing his SQL calls, to bring down the entire server, and then the hosting company has to pay out refunds to all the rest of the customers. We don't offer an SLA for shared hosting because of that fact, we'll do SLA's on a dedicated hosting environment, because if it's the customer's own code that causes the problem, he's only hurting himself, not everyone else who is sharing the sandbox with him.

Definitely, read the fine print before agreeing to anything, and analyze it carefully for loopholes, like the first guarantee that we picked apart. And if the company doesn't fulfill their network downtime compensation guarantee report them to the Better Business Bureau-they don't have to be a member for them to track reports of unethical business practices.

It doesn't matter who wrote the guarantee (although it should be the legal department and not marketing), if you guarantee something, you are legally obligated to fulfill the terms of that guarantee. It's the law.

[Locu]

As per industry standard, 99.9% is normal, the extra nine's are just hype, and as previously mentioned all about marketing. Funny thing is, in my marketing research, 99.9% or 99.999% brings more customers than 100% does. Even so, when looking at the SLA that tells you what to expect.

What we adopted was a very unique policy to ensure our customers absolute minimal downtime. We have in only one case had to actually apply the credit, but we offer a 5x credit for any downtime beyond the 99.9%.. it has proved to be one of the strongest SLA options I have ever seen.

[Phoenix]

Originally posted by Locu
As per industry standard, 99.9% is normal, the extra nine's are just hype.

Ours is an industry with no standards. For some providers 99% is normal, for others, the norms are different, but it is all on an individual basis as our industry consists of individuals setting their own standards.

The additional nines are possible, but they are not easily or cheaply obtained and they require a high level of control and adherence to rigid standards from the backbone network on down. Conxion is the company that leads the way when it comes to high-availability hosting.

I've noticed a recent increase in the use of the term High-Availability (HA) hosting by providers-including Interland, and a number of other providers. This too is being used as a tool for marketing hype, especially by Interland, who considers uptime of three nines to be High-Availability.


Funny thing is, in my marketing research, 99.9% or 99.999% brings more customers than 100% does.

That's because even among HA hosts, 100% is not even considered a theoretical possibility given current technology. Something new must be instroduced to render the classification system of nines and SLA's obsolete. What that something new is, remains to be seen.


What we adopted was a very unique policy to ensure our customers absolute minimal downtime. We have in only one case had to actually apply the credit, but we offer a 5x credit for any downtime beyond the 99.9%.. it has proved to be one of the strongest SLA options I have ever seen.

Is that a pro-rated guarantee? Do you refund 5 times the cost of the actual downtime itself (determined by dividing the monthly fee by the number of total minutes in a month), or 5 times the monthly service fee? Also, what factors go into determining your uptime. Is it network uptime? Server uptime? Averages? Monthly cumulatives? What exceptions are there such as scheduled downtimes.

[Locu]

Is that a pro-rated guarantee? Do you refund 5 times the cost of the actual downtime itself (determined by dividing the monthly fee by the number of total minutes in a month), or 5 times the monthly service fee? Also, what factors go into determining your uptime. Is it network uptime? Server uptime? Averages? Monthly cumulatives? What exceptions are there such as scheduled downtimes.

That would be 5 times the outage, not the monthly fee. Geesh.. giving away 5 months for a 10 minute outage I do believe could put nearly any company out of business and would be un-realistic. The way it works is as follows.

There are apx 43200 minutes in a month. 99.9% SLA gives apx 45 minutes downtime each month allowable. Which is actually quite a lot. But in the event a server is down for more than that, the amount over 45 minutes is credited back at 5x. I know you can do the math, but here it is..

Let's say a server goes down for an entire day (1440 minutes) (ouch!). Take 45 minutes from that (1395 minutes). Take 1395/43200. That leaves an outage for 3.23% of the month. Monthly amount times 3.23% (say monthly is $500) is: $16.15, times 5 is $80.75 credit due.

A day of downtime is un-excuseable neverless; however, it is just a point we give to our customers that say we are VERY serious about uptime! We may not have to pay the bandwidth for the day the server is down, but we will certainly be paying for it for the additional 4 days credit we give. The only exception to this would be an act of God (earthquake, war, flood, etc..) or if our uplinks were cause of the outage, but having multiple uplinks this is not likely to happen.

[Phoenix]

Originally posted by Locu
Let's say a server goes down for an entire day (1440 minutes) (ouch!). Take 45 minutes from that (1395 minutes). Take 1395/43200. That leaves an outage for 3.23% of the month. Monthly amount times 3.23% (say monthly is $500) is: $16.15, times 5 is $80.75 credit due.

So, according to your calculations, if the customer gets uptime of 96.77% (100% - 3.2%) they get 16% of their monthly fee back.

In order to get a full refund for the month, the outage has to be over 6 days (~80% uptime).

[Locu]

That sounds about right. Nobody should ever have this kind of downtime, but most companies will give a full month credit only if the customer is down for a full month. Most credits are given on a 1 to 1 outage. 1 day outage is 1 day credit.

Now I would hope there isn't any customers out there that would signup for ANY company which is gonna have the amount of time outages we are speaking of.. it's just offered to tell the customers we are SERIOUS about downtime, and we will pay if there does happen to be any.

[Phoenix]

That sounds about right. Nobody should ever have this kind of downtime, but most companies will give a full month credit only if the customer is down for a full month. Most credits are given on a 1 to 1 outage. 1 day outage is 1 day credit.

Now I would hope there isn't any customers out there that would signup for ANY company which is gonna have the amount of time outages we are speaking of.. it's just offered to tell the customers we are SERIOUS about downtime, and we will pay if there does happen to be any.

It's eerily similiar to the other SLA we were looking at up at the beginning of the thread, it purported to be 99.999%, but once we analyzed it thoroughly, their 100% guarantee was for ~99%.

Just as your 99.9% SLA gives a full refund at about ~80%.

[Locu]

Phoenix,

Your mis-reading something, or perhaps just intend to smash on a policy that exceeds yours? Perhaps the phrase "99999 [Five Nines]" rings a bell? A company which portrays to offer 99.999% uptime (5 minutes down/year), but when reading the SLA only offers 99.99%. On top of that

the service will be considered down if the web site is unreachable from all of !!insert company name!!'s major Internet Transit Providers (as of the time of this writing, this includes UUNet, Digex/Intermedia, Global NAPs, and Allegiance Telecom.) "

So, in essense, let's say somebody at your NOC trips over the master cable, brings all your circuits down except the unknown provider "allegiance telecom", which could be your own internal CLEC name.. so, basically, according to your SLA as long as one of your backbone providers can ping the service, it will not be considered down. That leaves a lot of room for margin.

Don't get me wrong, I am not trying to further fuel a p*ssing match among SLA's, because _every_ provider has a clause somewhere in there to indemnify themselves in the event of a mass-credit scenario; however, just don't bash on a policy because you are afraid it is better. You say 80%, whereas that number is clearly untrue. We clearly credit anything the surpasses 99.9%, and with our recent uptime, we could easily add another 9 on there and still not owe a single credit. As mentioned earlier, the 9's are a marketing angle, nothing more. Some hosts and back it up by showing solid uptime performance, some can not.

I think we got off the realm of the original discussion here anyway. Credits are typically never bound 100% into a SLA due to the liability there, credits are given on a per instance basis, but not ever absolutely required. Many webhosts will put in there (i know you know what i'm talking about), "a credit will only be given if you call and report the outage, and your reported time will be credited". In reality, during an outage, the phones are taken off the hook right and nobody can report it anyway, so therefore the company is not bound by SLA to give the credits. There is always a loophole.

I anxiously await to see a company which does not have 1 loophole in there to save their butts if they had a huge system crash.

[Locu]

ps.. "All your 9's are belong to us." ;)

[Phoenix]

To return this discussion to the topic at hand I've discovered a new guarantee that turns out to be no guarantee at all.

In keeping with our earlier method of anonymous analysis, I won't reveal the name of the host, but I did find this way of handling uptime guarantees to be unusual.

Their site prominently states their hosting comes with a 99.9% guarantee, but no where could I find the details of that guarantee. However, buried in their FAQ on their network status page, the following:



Q: What is your Up-Time Guarantee?

A: No provider can guarantee 100% uptime. Servers and systems must be brought down at times for routine maintenance and upgrades to ensure that your site will perform optimally. However, we strive to keep such service interuptions to a minimum, and, if possible, give you advance notice of scheduled maintenance routines. We use only top quality brand name servers, switches, routers, and other networking hardware. This too translates in greater reliability and uptime. As a result, within our own data center that is under our own control and management, we have been able to reach near 100% uptime. Of course, there will be times that you may not be able to reach your server, due to traffic conditions on the internet or problems occurring at our upstream providers facilities. Both of these conditions are entirely out of our control. However, if the problem occurs with our immediate upstream providers, we will do our utmost to have these providers correct the anomaly as soon as possible.

Did you see anywhere in there an answer to the question? Neither did I.



Q: How often will my server or the (edited).com network be down?

A: You should expect around 1/2 percent downtime.


in other words, 'there is no guarantee' and you should expect uptime of 'around' 99.5%.