Re: Enquiry regarding Linux in Mission Critical situation

From: lasernet (lasernet@mail.ny.optonline.net)
Date: 06/29/02 

From: "lasernet" <lasernet@mail.ny.optonline.net>
Date: Sat, 29 Jun 2002 08:44:03 -0400

"Klaas Vaak" <klaus@vaak.net> wrote in message
news:3D1D43FB.D8B9B3B3@vaak.net...
> lasernet wrote:

> > Generally the reliability figures for a single fault tolerant (RAID)
Solaris
> > or Intel box is around 99.95% uptime (4.38 hours down time per year).
And
> > that's just the hardware alone, does not include OS, applications,
network
> > infrastructure, etc. After all, what the mission critical application is
> > good for, if it is not accessible. The availability of any chained
system is
> > the product of the availability of individual critical components,
assuming
> > components fail independently of one another. Even if each element of a
15
> > component critical path is 99.999%, the resulting overall system
> > availability is (0.99999)15, or 99.985%. For the entire system the
weakest
> > link sets the upper boundary on overall availability. Drop in a single
> > switch (99.9% reliable) in to this chain, the previous figure drops to
> > 99.89%. That's just how the reliability of the system is calculated when
all
> > of the critical components are accounted for. The system might actually
have
> > better availability than calculated, however, that's not the figure what
> > most mission critical systems based upon..........
>
> I believe the figures were 25 hours for Unix/Linux boxen, 70 hours for
> NT and 5 minutes for mainframes.

That depends, should one spend as much as the mainframe cost then your
numbers are not correct. Otherwise comparing mainframe's uptime to
distributed servers is meaningless, they aren't the same systems.
There's a noticeable difference in the maturity of these systems also, the
older platform has a better availability record since the OS code is more
mature.
Even within the same platform there are differences, quote from:

http://www.aberdeen.com/2001/research/01022781.asp

"Data offered by interviewees at production sites Aberdeen tracked during
the past year shows that Windows 2000 on Intel servers are running at an
average 99.964% uptime rate. Noteworthy, here, is that these reliability
statistics represent stand-alone server uptime, but yield uptime metrics
that were once associated only with NT cluster environments. In fact, under
the Windows 2000 Datacenter Server Program, some suppliers are offering
guarantees of up to four-nines uptime - i.e., system availability for all
but a few minutes per year. From Aberdeen's perspective, this level of
availability is outstanding compared with that of its Microsoft OS
predecessor, NT."

To achieve this level of availability, it took the NT platform about ten
years. Presumably the .NET platform will have even higher availability in
couple of years. Although this progress was relatively speedy, it's nothing
like what Linux did. The free OS cut this time in half.......

Relevant Pages

  • Re: [Total OT] Trying to improve some numbers ...s
    ... rebooting a server once a month to upgrade it, even if its down for 5min, is about 99.989% uptime, which is a good number, but the OS is still up to date ... ... The 'metric' one should be looking at is how *much* the server is up, ... "Uptime since first start-up ever", ... availability in percent. ...
    (freebsd-questions)
  • Re: [Total OT] Trying to improve some numbers ...s
    ... rebooting a server once a month to upgrade it, even if its down for 5min, is about 99.989% uptime, which is a good number, but the OS is still up to date ... ... The 'metric' one should be looking at is how *much* the server is up, ... "Uptime since first start-up ever", ... And here "availability" is defined as ...
    (freebsd-questions)
  • Re: C portability is a myth
    ... Of course it was the C standard itself which was ... I did not even think about portability. ... platform, all compliant with the exact same version of the ANSI C ... its a limiting factor on availability. ...
    (comp.lang.c)
  • Re: Thou shalt have no other gods before the ANSI C standard
    ... >>It depends on whether integrity is more important than availability. ... >>kind of tradeoff right requires knowing our priorities, ... INTERnet facing servers are separate from ...
    (sci.crypt)
  • Re: Small to medium web farm configuration question
    ... My client wants the availability of the ... We've used that configuration for their external ... to the two-server web farm for availability more than for capacity, ... Bottom line, they want the high availability of two web servers, but don't ...
    (microsoft.public.sharepoint.portalserver)
Quantcast