PDA

View Full Version : How many 9s are enough?


allan
04-20-03, 06:22 AM
I just finished reading a very interesting Cisco book that outlines what is involved in increasing availability within a network. Each 9 you add has an incremental cost, at some point adding the extra nine becomes too expensive to make it practical.

Networking tries to follow the standard set by phone companies: 99.999% availability (also known as 5 9s). Most of the time, we fall short of reaching that goal.

From your perspective how many nines are enough? Do you prefer to increase your level of availability through redundancy, or by buying better equipment?

interactive
04-20-03, 06:42 AM
I'm with the more uptime the better just like everyone else. Standing in a customer perspective, 6 9's would be awesome. Standing in a service provider's perspective, it would depend entirely on my network, and what I thought it could withstand. But probably either 4 or 5 9's. What book was it by the way sounds interesting?

allan
04-20-03, 06:48 AM
The book is "High Availability Network Fundamentals"

http://www.amazon.com/exec/obidos/ASIN/1587130173/

no1v2
04-20-03, 11:35 AM
It all depends on the application and the budget. If my personal site (if I had one) only had 99.9% availability, oh well. But if a business' revenue and reputation is based almost entirely on their website, downtime is simply not an option - whatever the cost. If the cost doesn't fit the budget then the business plan is fundementally flawed anyways.

Letting my mind wander about it right now it seems that the difference between, say, 99.95% availability and 99.999% availability is in a way usually an all-or-nothing upgrade. If a site needs that kind of reliability then skimping for less isn't an option, but if it doesn't need it then going part of the way, but, say, still having one or two single points of failure, would be a waste of money.

allan
04-20-03, 01:51 PM
Originally posted by no1v2:

It all depends on the application and the budget. If my personal site (if I had one) only had 99.9% availability, oh well. But if a business' revenue and reputation is based almost entirely on their website, downtime is simply not an option - whatever the cost. If the cost doesn't fit the budget then the business plan is fundementally flawed anyways.


Doesn't it depend on your target market though? Is it worth building 5 9s uptime for $19.95 a month clients? The vast majority of businesses don't need a website any more complicated than what $19.95 a month can buy -- but at that price (and realistically a $19.95 a month account is a $10 account on these boards). The website may be important to their business, but they may not be able to justify more than $20 a month.

From a hosting perspective 99.99% availability means 4.5 minutes of downtime in a 30 day month. 99.999% uptime is equal to 25 seconds of downtime in a 30 day month. However, it may cost an extra $100,000 to get those extra four minutes.

If your clients are spending $1500 a month (Rackspace's average per client billing), then you can easily justify the cost. For a company focusing on shared hosting it might be hard to justify the extra expenditure.

no1v2
04-20-03, 02:01 PM
I was thinking of this from the perspective of a non-hosting business or organization. Anyone who needs 99.999% availability isn't relying on a single host/data center, just like how they're not relying on a single ISP, a single switch, a single cable, or a single server.

hostpath.com
04-22-03, 11:59 AM
Do you mean ACTUAL or what the SLA says? Big difference.

allan
04-22-03, 12:28 PM
Originally posted by hostpath.com:

Do you mean ACTUAL or what the SLA says? Big difference.

Actual availability -- and I am of the opinion that you shouldn't offer an SLA that you think you will have trouble meeting :D.

hostpath.com
04-22-03, 01:00 PM
In that case, 3 nines is probably the most reasonable level you should expect to offer. 8 hours of downtime over the course of an entire year is darned good, depending on what you mean by "availability" (different people interpret that in different ways).

And in fact, SLA's really guarantee nothing -- except some sort of remedy when the delivered service falls below the expected level of service. I've heard many newbies (and some not-so-newbies) brag about SLA's with five nines and claim that their server "will NEVER go down". Nope, not even close.

kunal
04-22-03, 04:11 PM
Hey,
How do you guys define uptime? does downtime for network upgrades get included in the uptime calculations? I know some of the bigger hosts do not include network upgrades in there downtime. But what is the industry, if any exists?
Kunal

allan
04-22-03, 04:19 PM
This goes back to what hostpath was saying: If you are laying out an SLA, most hosts are going to exclude scheduled maintenance from the SLA -- so that time won't count against an SLA.

On the other hand, if you are tasked by your boss to create a web infrastructure with 100% availability then you are not going to exclude maintenance, because your boss will have different expectations.

racklmy.com
04-25-03, 06:50 PM
How was the book?

allan
04-25-03, 06:54 PM
Originally posted by racklmy.com:

How was the book?

It is a good theoretical overview, but it really doesn't delve into practical examples.

allan
04-25-03, 06:54 PM
Oh, and what's with the l, rackl :)?

DizixCom
04-25-03, 07:01 PM
I'm a big believer that guarantees create artificial confidence. I'm much more concerned with how my provider reacts to an issue over how well they market their supposed ability to keep their network operating.

I loved the spin Chris Farley put on this in Tommy Boy: Tommy: "Let's think about this for a sec, Ted, why do they put a guarantee on a box? Hmm, very interesting."
Ted: "I'm listening."
Tommy: "Here's how I see it. A guy puts a guarantee on the box 'cause he wants you to fell all warm and toasty inside."
Ted: "Yeah, makes a man feel good."
Tommy: "Course it does. Ya think if you leave that box under your pillow at night, the Guarentee Fairy might come by and leave a quarter."
Ted: "What's your point?"
Tommy: "The point is, how do you know the Guarantee Fairy isn't a crazy glue sniffer? "Build model airplanes" says the little fairy, but we're not buying it. Next thing you know, there's money missing off the dresser and your daughter's knocked up, I seen it a hundred times."
Ted: "But why do they put a guarantee on the box then?"
Tommy: "Because they know all they solda ya was a guaranteed piece of sh*t. That's all it is. Hey, if you want me to take a dump in a box and mark it guaranteed, I will. I got spare time. But for right now, for your sake, for your daughter's sake, ya might wanna think about buying a quality item from me."
Ted: "Hmm. Okay, I'll buy from you."
Tommy: "Well I...What?!"

kunal
04-26-03, 10:24 PM
Originally posted by allan:

This goes back to what hostpath was saying: If you are laying out an SLA, most hosts are going to exclude scheduled maintenance from the SLA -- so that time won't count against an SLA.

On the other hand, if you are tasked by your boss to create a web infrastructure with 100% availability then you are not going to exclude maintenance, because your boss will have different expectations.


Are there any consensus on which method hosts chose?

kunal

allan
04-27-03, 05:54 AM
Originally posted by kunal:

Are there any consensus on which method hosts chose?

kunal

All hosts that I know of choose the first -- some will actually create the infrastructure to back it up, but they also charge a boatload of money :).

Bogdan
04-29-03, 05:39 AM
LOL - For a second there I thought you were asking about guns.

----please don't mind me---:baghead:

kunal
04-29-03, 05:46 AM
Originally posted by Bogdan:

LOL - For a second there I thought you were asking about guns.

----please don't mind me---:baghead:

hehe... :D