server uptime

[koddos]

What is an average server uptime?

[wswd]

Any good company should be darn close to 100%.

[Seraphim]

Uptime as in the result of typing uptime in a ssh session?

It depends on how good you are at managing the server, and how stable the server and providers feeding it are. I've had uptimes in excess of 300 days of nonstop operation, and I've heard people talk about servers pushing 5 and 6 years since they were rebooted last.

Now if you mean uptime as in a site's availability, that introduces a whole new set of rules. Usually site availability is measured as a percentage, with most providers being able to maintain site availability in excess of 98%. A perfect 100% is actually not possible, even the biggest clustered servers you could possibly build could still go down due to a software defect or a network fault making them unreachable. But you have to be downright careless to drop below 95% availability.

Point of fact I usually change providers if their network is unable to maintain at least 98% availability, with a good goal to be in excess of 99.7% availability over the past 3 months.

[wswd]

99.7% equates to 5 minutes of downtime every single day, or 2.5 hours of downtime in a given month. 98% is 30 minutes of downtime per day. That's completely unacceptable, IMO. Even 99.9% is almost 45 minutes of downtime in a given month.

As a host, 99.95% is a good, realistic number to shoot for. That allows for about 22 minutes of downtime per month (or 4 1/2 hours over the year). That's more than enough for any reboots after updates (if you aren't using Ksplice), or any unexpected maintenance that might need to occur...hard drive replacement, RAM replacement, upgrades, etc. If you have some catastrophic network or server failure, the uptime might be lower than that, but how often do you really have those types of failures?

[Seraphim]

99.7% equates to 5 minutes of downtime every single day, or 2.5 hours of downtime in a given month. 98% is 30 minutes of downtime per day. That's completely unacceptable, IMO. Even 99.9% is almost 45 minutes of downtime in a given month.

As a host, 99.95% is a good, realistic number to shoot for. That allows for about 22 minutes of downtime per month (or 4 1/2 hours over the year). That's more than enough for any reboots after updates (if you aren't using Ksplice), or any unexpected maintenance that might need to occur...hard drive replacement, RAM replacement, upgrades, etc. If you have some catastrophic network or server failure, the uptime might be lower than that, but how often do you really have those types of failures?

One good screwup from anything is enough to throw you over that 2.5 hours in a month. Now if you get lucky you can manage your downtime so that those few hours also take place at like 4AM pacific time when the majority of the end users are asleep anyway. If you're not so lucky, it's an upstream network fault or attack condition at 7PM PST on a Saturday when tons of people will be online browsing, and everyone sees that you went down.

Uptime and availability alone are poor benchmarks, because you could have a 99.95% uptime and if that 22 minutes of downtime in a month just so happens to take place right smack in the middle of peak hours when all of your client sites are lively, you get as much flak from them if not more than if you take 5 minutes a night at 4AM when nobody is looking.

I'm starting to wonder if we should use a different benchmark instead- Observed Outage Per Server, similar to Tao's metric where only outages that are likely to have been seen by the majority of the end users would be counted. Using that basis, you should have an OOPS of no more than 1 per month.

[CS Squad]

I really wanted to know, do any of you using ksplice?
Coz I did not use it at all, and I'm not sure if it is as good as it advertise.
How does it able to patch the kernel and make the new kernel running while there is no reboot at all?

[[JSH]John]

We used to use Ksplice but half the time the updates required a reboot for new features.

It's great for patching security issues but if a new feature is included in a new kernel, it's of no use as you'll require a reboot to get into the new kernel.

[wswd]

^^^ What he said. It was great for routine patches, but anything more than that would require a reboot.

[Seraphim]

How does it able to patch the kernel and make the new kernel running while there is no reboot at all?

Probably similar to how some malware gets control of your system, or how a buffer overflow works. It simply redirects the execution such that causes it to go into a different operating loop, in this case enabling it to shift from one kernel to another while live. From the standpoint of actual operations it is relatively easy, but to actually make a working implementation inside something as complex as a kernel is probably why it is something that only Linux even offers.

I wondered about that too with ksplice, if the reduced upgrade outages were actually worth the costs. Probably more true at larger scales than individual machines.

[CS Squad]

If thats the case, I think it would be best to bring the server down for a few minutes, instead of depending on ksplice...

[storminternet]

100% uptime is practically impossible because server may go down because of hardware, network issues or require reboot for the emergency update like kernel upgrade, windows update etc.. 99.95 % is recommended uptime but at least 99.8% can be achievable.

[Tyler]

100% uptime is practically impossible because server may go down because of hardware, network issues or require reboot for the emergency update like kernel upgrade, windows update etc.. 99.95 % is recommended uptime but at least 99.8% can be achievable.

I've had servers with a 99.9% uptime, 99.95% uptime and 100% uptime in a given month. It's possible, but it's not a guarantee you should shove on your main website or sla since obviously anything can happen.

[harrychristopher]

As per my knowledge average 99.5%, bcz even a single day by any cause it may be down due to hardware, network issues.

[Schmarvin]

I usually aim at 99.8% uptime. Especially while trying to maintain up-to-date software.

[Santrex_]

99.99% uptime is quite good.