What is so great about 99.9% uptime?

[Will]

It's always been there, but I guess I've never fully thought about it before. What is a "99.9% uptime guarantee?"

0.1% of a day is 86.4 seconds
0.1% of a week is 10.08 minutes
0.1% of an average month is 43.68 minutes
0.1% of the year is 8.76 hours

Just wondering how annoyed some clients would get if there site was down for an average of 86.4 seconds every day. (That could be a 10.8 minute downtime on monday and then a 100% uptime record for the rest of the week)

Or even worse, your yearly 8.76 hours could all happen on one day!

The other question is 99.9% of what?

Client: My site has been down for 2 days, that's not 99.9% uptime - I want a refund.
Host: It will be 99.9% if we don't have any more downtime for 47952 hours.
Client: But that's not 99.9% of the year!
Host: Who said anything about the year?

I wonder how many hosts have pulled this stunt. Maybe not padding their percentages out over infinity but stating that the percentage must be over, say, 10 years! Any host would be glad to give you a month / year's refund if you stayed with them for 10 years!

Which is another point - what, exactly, is the penalty? I've never checked. A refund for the hosting "provided" during the downtime? Refund for that week? For that month? Year? All time? Maybe more hosts need to draft up some big Terms and Conditions with these answers in black and white.

99.99% - now there's a good offer. But I'd have to make sure that the host specifies a time that those figures apply to!

 

[Matthew]

When you start adding 9's to the end of that it gets more expensive for the host each time. When I browse around hosts I see that 99.9% is promoted and assume its a monthly guarantee and if the site should be down for more then 43.68 minutes in a month then some form of compensation can be requested by the client to the host for lost time. However, when promoting a web hosting company I often see (and I would do the same) the uptime over several years.... like 99.96% for the last 4 years. I also think it is ok to do this because your track record is what ever % for those few years and the client wants to see numbers for a few years and not just last month.

 

[Will]

Hopefully, when I finally get round to doing some work on the HostReview system we will have a fair system to judge hosts on.

 

[monaghan]

Of course you can also interpret 99.9% over your whole estate, then the more servers you have, the longer downtime you can have on an individual server without impacting your overall figure.

Claims like 99.9% must always be backed up with a formal SLA that defines all the criteria for the figures. Of course once you get into this level, then you're way above the average budget host in terms of the cost to meet the SLA and anyone offering this should be looking at redundant hardware, auto failovers and so on, but then again if you need to be up for 99.9%, then your revenue model would allow for the investment in server technology to achieve this.

I still chuckle when I see hosting requests wanting 99.99% and 24/7 support for a few £/month.

 

[Will]

Of course you can also interpret 99.9% over your whole estate, then the more servers you have, the longer downtime you can have on an individual server without impacting your overall figure.

Hahaha - I never thought of that!

Customer: My site has been down 25% of the time this year! I want my money back!
Host: Ah, but we have 20 servers, 120 network cards, 120 network cables and 20 routers - all of which have been up 100% (How would the consumer know this is a lie?)

281 (pieces of network equipment) * 365 (days) = 102565
Your server was down: 91.25 days
91.25/102565 = 0.09 % (rounded up)

So you had more than 99.9% uptime!

 

[monaghan]

That's the good thing about SLA's, you can word them how you want and most customers won't check the fine print

None of our customers are daft enough to run a business needing high availability on a shared hosting server. Anyone needing high availability has their own servers with backups or whatever is needed for their account.

 

[Will]

That's the good thing about SLA's, you can word them how you want and most customers won't check the fine print

None of our customers are daft enough to run a business needing high availability on a shared hosting server. Anyone needing high availability has their own servers with backups or whatever is needed for their account.

I don't think anyone likes low availability, though. I had a host once who was down almost daily. And another who had their security compromised weekly. Even though my sites had literally no income gernating side to them, it was bad to have your sites down so much.

 

[monaghan]

Our hosting servers have high up time, an SLA should not be used as en excuse for bad service. 95% uptime averaged over a month should be sufficient for most average web sites.

What annoys me are those who want to pay peanuts and have 4 or 5 9's availability.

Anyone dependent on their site for their income shouldn't really be hosting on a cheap shared system and their accountant would be able to offset the costs of a decent setup against their tax liabilities so the cost is not really an issue.

 

[XNL0]

Hahaha - I never thought of that!

Customer: My site has been down 25% of the time this year! I want my money back!
Host: Ah, but we have 20 servers, 120 network cards, 120 network cables and 20 routers - all of which have been up 100% (How would the consumer know this is a lie?)

281 (pieces of network equipment) * 365 (days) = 102565
Your server was down: 91.25 days
91.25/102565 = 0.09 % (rounded up)

So you had more than 99.9% uptime!

That's the best joke I've ever heard!

The best uptime offered is five "nines"- 99,99%

 

[monaghan]

That's the best joke I've ever heard!

The best uptime offered is five "nines"- 99,99%

Shouldn't that be 99.999% if it's 5 nine's

You can get 100%, however this requires multi location DC's and lots of expensive equipment

 

[Matthew]

I expect SLA to be per server per month. We run a few hefty connections between here and the US with a 99.9% SLA if I remember correctly. I would just laugh if it went down and BT, Verizon or Easynet said... well our exchange in London was up and running, so that 8 hour downtime while we rebuilt the frame doesnt count. To me, SLA is the agreement on your service (not others or other servers in the building). If I run a website then I expect to be able to reach it. The other thing to watch out for with SLA is what you actually do get in return. Just because a 99.9% SLA is in place it doesnt mean you will get 99.9% uptime that month. It just means that if the service level drops below what was promised then some thing is expected in return such as compensation which is usually a few % of the account cost. In shared hosting, it isnt worth worrying about as I personally wouldnt be fussed in trying to reclaim a few dollars. If I had an ecommerce site and needed the income then I would look more seriously at the equipment running my site rather then thinking "I only transfer 100GB/Month... that $2.99 account is all I need". 100% though... I think I have seen all the majour sites down at 1 point so I think 100% is hard to sustain over a long time. If it's per year then it is possible... but just 1 slip up and thats the 100% gone

 

[Matthew]

My PC is driving me mad recently... that post above was sperated in to paragraphs. Sorry it's all mixed in to one.

 

[monaghan]

As you expressed above, you'd expect it to be per server per month, but how many people actually read the SLA?

For normal shared hosting, then in my personal opinion, if the site's down, it's down, it's hardly critical for most people's sites.

What does however surprise me is the way people scream all over forums when their whole business is down because their $1.99/month oversold webhost has an outage. I guess some people will never learn that if your business depends on a service you should expect to pay for the necessary tools to keep that service operational. If my whole family's next meal was dependent on my ecommerce site, I think I'd put it on something better than a cheap shared account and I'd read the SLA fully

I had an occasion a few years ago when in a previous position (not in a hosting environment) where I got a document to review and noticed that they were selling the customer a 15 minute maximum failover capability. The guy who'd authored it had not actually checked the SLA he had in place with our department as the servers in question were on a 4 hour SLA and my guys had a 30 min response to the Helpdesk call, he took great offense when I politely asked him how he intended to guarantee his 15 min SLA without increasing his budget significiantly.

 

[monaghan]

Your other point on 100%...

100% is possible, just not in most people's budget

4 or 5 9's is sufficient for most people as most customers would probably re-try a couple of times before moving on.

 

[Will]

I wonder how such things are tracked. Sure, network connectivity is usually a good indicator but there are so many things that can go wrong. DNS responses, the web serving software itself, SSL certificate services down, seperate or secondary nameserver machines down, email, ftp, control panel etc... The list is almost endless. Does the term "downtime" stretch across all of these services? Does it actually straddle them all? (eg: all services must be down before it's considered "downtime".)

I'm curious from a third-party perspective. I've never had an SLA or had to argue about downtime. I'm just curious about how challenging it could be to prove, even with an extensive document to use as a guide unless it covers all of these points.

 

[monaghan]

If you were doing a formal (and critical) SLA, then you'd most probably detail each element of the service and it's individual impact on your business and therefore it's SLA and penalty clause. You would often have differing availability requirements at different times of the day (or week/month), a good example of this might be a bank, loosing 1 ATM on a mid week afternoon would have less impact than loosing it on Pay Day.

I would guess that most people offering an SLA actually have no idea what an SLA actually is or even have a clue how they can implement it. Most commercial SLA's will generally have maintenance windows and things like that and your SLA time will be x% of the agreed availability period, not x% of actual wall clock time.

Seeing the "demand" for APC style remote reboot switches and so many comments on reboot times in other forums, I would guess that many "hosts" have to reboot their systems frequently, this could easily consume all their SLA down time just in reboots.

I would suggest most hosts understanding of an SLA is very much like the way everyone offers 24*7 support and GB's of space for £/month. Unless a host publishes the exact criteria of the SLA and how it's measured, then you have no chance of getting anywhere.

 

[Addmember]

Whose SLA would you call ideal?

 

[monaghan]

For an SLA to be meaningful, it would need to be drawn up against both side's expectations. If you're a customer and just looking for high up-time, then buy a quality service rather than a cheap one, then you'll more than likely not have to invoke an SLA penalty.

If 5 or 10 mins downtime is going to cause significant impact to the business, then I would suggest that the business could afford to buy a better service in the 1st place.

 

[shade]

There is a lot of talk about Uptime Guarantees and some hosts have that 100% Guarantee on their sites, but how are the things in reality? Is it possible to prove that this guarantee has been violated and get refund?

 

[Will]

If you see 24 hour, 7 day a week, 365 days a year uptime guarantees be careful in 2008 - there's 366 days! :laugh: