(OT)Re: Cottage industry suggestion.. [long]

"jamie powell" <jamie_p84@xxxxxxxxxx> wrote in message
news:gsl65h$5vr$1@xxxxxxxxxxx

"Mikeapollo" <usenet@.removethisbit.mikeapollo.net> wrote in message
news:BuadnQOfdv0JaXDUnZ2dnUVZ8sSdnZ2d@xxxxxxxxxxxxxxxxx

Depends on the load.. Here that doesn't work so well as the CPU's are
extremely modest and the reason there are 3 boxes is as one gets
"saturated"
another physical processor can be added.

Are you operating them in a distributed-load configuration, managed by the
virtualisation software? Or just 3 independent boxes?

Distributed load - usually 2 to 3 "virtual" machines on each of the 2 boxes,
1 standalone (no virtualisation)

2 of the boxes now run at pretty
much 80-90% duty cycle during most of the day and if a processor had too

What do you run that generates 80-90% for most of the day?

Various odds n' sods - main tasks include but aren't limited to Oracle
(Siebel CC+OS8i) and Windows Media Encoder 10 (running on the fly) takes a
bit a good whack on Windows Server 2003. On the same machine under that lot
in a separate "container" is 2 mailservers running under SuSE 9.2 (I know I
know.. OLD) and a Win2K container running hMailserver. Antispam stuff of
course on both SuSE and Win2K hence that is a *very* busy little box.

Second processor is quieter and runs a couple of VNC based systems, all the
web gunk in Windows and linux flavours (Apache, MySQL, Squid, etc etc) in
separate containers, FTP/SFTP, software MPEG encoder for the "house
distribution system" (basically MythTV with Freeview running onboard and
analogue capture card hooked up to the sky box), An MS-exchange server
(training installation) with a broken "push" email service (grr) and the
admin suite (UPS monitors, sync, network time services, etc etc)

Third processor is a hybrid running as a NAS/lazy desktop/VNC client. It's
the lowest power (in terms of consumption) device for the job and there is
no real limit to the amount of cards and drives that I can daisychain into i
t (well, OK there is - 256 devices but I'm never gonna hit that!) It's also
runs KDE so makes a nice little "desktop" for browsing the net when I can't
be arsed to boot something else up. This holds the OS images that the other
machines boot into as well. With the exception of my desktop PC it's the
only box that contains hard discs, currently 15 of them of various sizes and
access times. Smallest being 2.1Gb (ancient fujitsu drive), largest being
250Gb. All IDE drives and arranged into 2 banks. One bank of 4 as RAID 0
(500Gb+500Gb mirrored) and the rest of the drives as a large RAID1 to hold a
"grandfather" backup image of the first bank - just in case.

The rig is all set up in such a way that if any single physical unit dies it
can be changed pretty much instantly - the "working" drives on the
fileserver (4 x 250gb) are easily transplanted into any of the other "spare"
machines caddies should the fileserver fall over. The "grandfather" backup
image however can only be read by the hybrid and due to RAID0 has no
protection against a single drive failing - however the grandfather backup
image should *never* be needed but given I have to look after the data of
others for the moment, it's a nice third layer of safety if BOTH sets of
drives in the first RAID1 bank fell over (and I have seen that happen with a
couple of IBM drives!)

much "idle" time then it's tasks are moved to another box anyway (I hate
idle CPU's as they cost money to run).

Manually move them then, I presume...

Yes. If a physical PC has more than 15hrs idle time in a 24hr window
consecutively over a number of days then it's "redundant". It's hard disc
image gets scooped up and then the whole system gets virtualised onto an
existing box.

Likewise, if a physical PC starts creaking for any reason, one of the
virtual containers can be shed onto another available physical device. This
rarely happens but can happen for "special" tasks or indeed if the Siebel
stuff is asked to do something daft :)

Running one massive machine with a set of multi-core processors and
buckets
of RAM would mean plenty of free cycles but again would lead to a
machine
that would be out of the budget for my hobby and also would probably
consume
more power than the existing setup. (My desktop machine at peak consumes
almost as much power as all the machines left on 24/7 but of course is
more
powerful than any of the "worker" boxes).

Yep, my desktop PC was using 160w even with idle CPU, so I switched to an
eeebox (intel nano CPU), which uses 13.5w with wifi disabled.
The eeebox handles most of my day-to-day computing tasks perfectly well,
and
it's whisper quiet too.

Yes - Noise for me is the killer - I don't mind power consumption too much
but like to keep it low where I can. Unfortunately, I think of the x86
platform as a hugely inefficient beasts running little more than a very
souped up Z80 with a few custom extensions that have grown with it over the
years.

AMD have the right idea - start changing the architecture step by step and
moving over to full RISC and only providing interpreted (emulated) x86
support ;)

But where servers are concerned, even a basic celeron box with no graphics
hardware installed is likely to consume at least 40w, so it soon adds up
and
becomes less efficient than virtualisation on a single fast box.

Again, I disagree slightly - that is a matter of scale really. I have a
Compaq box sat here right now plonking away at 37.5% of it's rated speed
with 512Mb RAM, sound, graphics but no hard disc (network boot) - it's
sitting there drawing 0.1004A from the UPS - Voltage is currently 237.5v so
I make that about 24w?

Admitted, not a typical scenario, however by rejecting many components of a
system and using high efficiency power supplies and clock throttling it's
very easy to reduce the overall consumption of 3 machines to less than one
fully spec'd one - but of course it's a very non-typical setup and one that
wouldn't suit everyone and definitely not for a real-world commercial
environment but pretty good for hostings your old bosses stuff, your
employers "technical demos" and your mates business ventures until they get
off the ground and end up on a rack in telehouse :)

On the setup here the biggest bottleneck with virtualisation is disc and
NFS
access. Mail, MySQL and Oracle when performing complex queries can tie
up
the discs quite severely which then limits what can be done while the
other
requests are queued.

You need faster disks and ideally a RAID array, with redundancy incase one
of the drives fails.

The whole network runs on two rather large RAID arrays.
Hades = 4 x 250Gb in Raid 1,
Hera = 4 x 120Gb + 2 x 80Gb + various tiddly discs in RAID 0 (used to backup
Hades as a grandfather file)

The machines here are not quite home test environments, I do run a few
services - 3 commercial (that pay the upkeep and the lekkie bills -
sometimes) and a few charitable sites and services and of course my own
"junk" and "tests" however, it's not flawless and *is* just as a hobby.
If
it was a serious venture I'd be using Telecity over in Hulme again!

I used to do commercial hosting, but this meant I couldn't play around
with
my kit as much as I wanted, and it's not as if I needed the money, so I
stopped.

Well, I don't see why it stops you playing with your kit as you want it
unless you mean it takes over hardware that you'd rather use for other
reasons?

My venture into serious commercial hosting was dire and a *huge* mistake - I
had the technical skills and knowledge but not the management skills or the
"selling" skills to market what I had to offer - and although I thought I
did have a USP for my enterprise I found there was a hell of a lot of
competition in Manchester and at that time didn't think of the broader
picture of trying out of area..

I'm older and wiser now and prefer being an employee (even if it seems I
just do whatever the hell I want - according to my manager! *LOL*)

As a final point, all this really *is* just a hobby and depending on my
financial status in 8 months time (read as "if I've not been made bankrupt
by this time next year" but it's sadly looking likely) then the plan is to
move everything onto one box anyway - especially given the number of things
I'm running both experimentally and for people who've asked nicely and
crossed my palms with chocolate silver will also decrease over the next 12
months to pretty much nothing anyway as they either outgrow what I can offer
and move to proper facilities or fold... In which case we'll be down to
running an old 486 DX2 running just my website and mailserver and back to an
easy life of a single IP address ;)


.

Relevant Pages

  • Re: Reducing energy consumption through software
    ... The highest power consumption ... Having another core running a memory ... typical 3.5" 7200rpm drives). ...
    (comp.arch)
  • Re: Reducing energy consumption through software
    ... The highest power consumption ... Having another core running a memory ... typical 3.5" 7200rpm drives). ...
    (comp.arch)
  • Re: self built NAS "how-to" anywhere?
    ... Emphasis would be on low energy consumption and noise level. ... have a free slot between the drives. ... continually playing music etc from it and the actual power use will depend on the nature of the burst use etc. ... just because they are nice and quiet and dont ...
    (comp.sys.ibm.pc.hardware.storage)
  • Re: self built NAS "how-to" anywhere?
    ... and the actual power use will depend on the nature of the burst use etc. ... effect unless the drives are allowed to power down when not being used. ... Judging by your last answer, I would only get near the TeraStation consumption values if I allowed the drives to power down when idle, right? ...
    (comp.sys.ibm.pc.hardware.storage)
  • Re: Require competent person to fix RiscPC
    ... It's because it's a full-scteen app with no readily apparent ... When it runs on RM machines, ... Potentially bad for disc drives. ... > The belief that the HD will be 'buggered' by a power off is a fallacy. ...
    (comp.sys.acorn.misc)