Data Center Wiring Standards

Mark Foster blakjak at blakjak.net
Sat Sep 9 01:23:59 UTC 2006


> My thoughts go like this:  We put a patch panel in each rack.  Each of
> these patch panels is permanently (more or less) wired to a patch panel in
> our main patch cabinet.  So, essentially what you've got is a main patch
> cabinet with a patch panel that corresponds to a patch panel in each other
> cabinet.  Making connection is cinchy and only requires 3-6 foot
> off-the-shelf cables.
>
> Does that sound more correct?
>
> I talked to someone else in the office here, and they believe that they've
> seen it done with a switch in each cabinet, although they couldn't
> remember is there was a patch panel as well.  If you're running 802.1q
> trunks between a bunch of switches (no patch-panels needed), I can see
> that working too, I suppose.
>
> Any standards?  Best practices?  Suggestions?  Resources, in the form of
> books, web pages, RFCs, or white papers?

Theres a series of ISO Standard for data cabling but nothing is yet set in 
stone around datacentres.  I think the issue of Standards in datacentres 
was touched on here some time back?

Ok, a quick google later,

TIA-942 Telecommunications Infrastructure Standards for Data Centres 
covers off a lot of the details. Its pretty new and I don't know if its 
fully ratified yet?

I quote...

--8<--
Based on existing cabling standards, TIA-942 covers cabling distances, 
pathways and labeling requirements, but also touches upon site selection, 
demarcation points, building security and electrical considerations. As 
the first standard to specifically address data centres, TIA-942 is a 
valuable tool for the proper design, installation and management of data 
centre cabling.

The standard provides specifications for pathways, spaces and cabling 
media, recognizing copper cabling, multi-mode and single-mode fiber, and 
75-ohm coaxial cable. However, much of TIA-942 deals with facility 
specifications. For each space within a data centre, the standard defines 
equipment planning and placement based on a hierarchical star topology for 
backbone and horizontal cabling. The standard also includes specifications 
for arranging equipment and racks in an alternating pattern to create 
ìhotî and ìcoldî aisles, which helps airflow and cooling efficiency.

To assist in the design of a new data centre and to evaluate the 
reliability of an existing data centre, TIA-942 incorporates a tier 
classification, with each tier outlining guidelines for equipment, power, 
cooling and redundant components. These guide-lines are then tied to 
expectations for the data centre to maintain service without interruption.

--8<--

The source url for the above was 
http://www.networkcablingmag.ca/index.php?option=com_content&task=view&id=432&Itemid=2. 
You may like to see if you can track down a copy of the referenced 
standard.


From my personal POV -

You have a couple of options depending on your switching infrastructure 
and required cabling density - and bandwidth requirements.  One way would 
be to have a decent switch at the top of each cabinet along with a Fibre 
tie to your core patch / switching cabinet.  All devices in that rack feed 
into the local switch, which could be VLAN'd as required to cater for ILO 
or any other IP management requirements.  Uplink would be a trunk of 
1000SX, 1000LX, MultiLink Trunk combinations of same, or perhaps even 
10Gig Fibre.

The other option would be to preconfigure each rack with a coupla 
rackunits of fixed copper or fibre ties to a core cabinet and just patch 
things around as you need to.  Useful if you are in a situation where 
bringing as much as possible direct into your core switch is appropriate, 
and cheaper from a network hardware pov - if not from a structure cabling 
pov.

Good luck. I know what a prick it is to inhereit someone elses shoddy 
cable work - I find myself accumulating lots of after-hours overtime, 
involving essentially ripping out everything and putting it all back 
_tidily_ - and hoping that I don't overlook some un-documented 
'feature'...

Mark.



More information about the NANOG mailing list