Orchid XP v8 <voi### [at] devnull> wrote:
> >> Now that's reliability engineering for ya.
> >
> > Indeed. That's another important differentiation.
>
> Even today, I see people selling "servers" which are really like
> "desktops". (1.2 GHz Intel Celeron with 512 MB RAM? I don't think so...)
>
> I think the "real" difference between a desktop and a server is probably
> fault-tolerance. When I first joined the 2-bit company I'm with now, we
> had a server with a 500 MHz AMD Thunderbird CPU and 128 MB RAM. Now we
> also had 4 GHz Intel Pentium IV workstations with 256 MB RAM in the lab,
> which makes a bit of a mockary of the "server" tag. But the lab PCs
> didn't have multiple Ultra320 SCSI HDs, hardware RAID controllers,
> redundant PSUs or 30,000 cooling fans. But the "servers" did. ;-)
>
> (Seriously, it didn't have 30,000 cooling fans, but it did have *a lot*
> Probably way, way more than necessary, IMHO...)
>
> The HP ProLiant I was briefly in charge of was really nice, actually. It
> had a little diagram on the front showing the system board, and a little
> red LED for every component on it for which a failure sensor exists.
>
> It has ECC RAM, and yet it has multiple RAM banks, and it can compare
> them and tell you if one RAM bank is faulty. (I gather this works in up
> to a 4-way configuration, for 4x RAM redundancy, in case the ECC doesn't
> catch it. Oh, and the RAM is hot-swap. HOT-SWAP!)
>
> It also has more fans than it is sane for any one device to have...
> although... it is quite a small form-factor, so maybe it does need it,
> actually.
>
> Also has hot-swap HD bays with indicator lights. The RAID software even
> has a function to make the lights change colour so you can yank the
> correct unit. (Nice!)
>
> And as if all that wasn't enough, there's a second computer inside it
> which you can use remotely to manage the server. Do stuff like see the
> video output, control the keyboard and mouse, and even make the main
> computer think there's a CD in the drive when really it's an ISO image
> you're serving from your remote control PC. So you literally turn the
> server on and off, fiddle with the BIOS and install the OS, all without
> ever physically being in the same country.
>
> Some day I may own a desktop computer which makes this server's dual
> quad-core Xeons seem puny and pathetic. But it won't have reliability
> and management features like a "server" does.
>
> --
> http://blog.orphi.me.uk/
> http://www.zazzle.com/MathematicalOrchid*

Reliability and redundancy is one of the strong points of IBMs last few
generations of mainframes.  If any page of memory fails, the error can be
corrected, if it is a hard error, that page of memory will be automatically
taken offline. If a processor fails, there are spare processors that can take
over and continue the operation if one fails.  Multiple paths to all i/o devices
where possible.  Of course redundant power supplies. Much of the machines
circuitry has redundant components. When a component fails, the machine phones
home and reports the failure.  Many time operations is completely unaware of any
problem until an IBM CE shows up at the door with a replacement part in hand.
And many components can be replaced on the fly, with out any downtime. The
latest machines are not totally fault tolerant, but each new generation is
better than the last.  I think the latest z/9 and z/10 models have a MTBF of
over 30 years.

Isaac

Post a reply to this message