On 19/12/12 14:16, Edward Ned Harvey (lopser) wrote:
From: [email protected] [mailto:[email protected]]
On Behalf Of Jonathan
For many years we have run stratum 2 NTP servers on some of our mail
gateways. The primary purpose is to provide an NTP service to assorted
RFC1918 networks. We are planning to virtualise those mail gateways,
and so we need to find a new home for NTP. I am aware that we could
purchase dedicated GPS-based stratum 0/1 NTP servers, but I don't want
the hassle of getting a good GPS signal in to the heart of our data
centres and our time sync requirements aren't too tight. Before
purchasing three dedicated 1U servers which I plan to site at three
different (geographically close) campuses, I thought I'd ask if anyone
has any cheap/neat solutions which can support multiple NICs/VLANs and
have a generally drift-free clock. Bonus points for kit available off
the shelf in the UK. Apologies if my terminology offends any NTP gurus;
hopefully my requirements are clear.
Do you have special requirements? Even for providing NTP to assorted RFC1918
networks, most people are fine with running a normal NTP server without special
hardware, on normal servers. I know I always configure a virtualization host
(vmware, xen, virtualbox, whatever) to sync with something upstream, and then
all the VM guests sync with the host. It just so happens the Windows Servers
are vm guests, and the Windows Servers provide NTP to the windows clients and
the LAN. While I certainly acknowledge a lot of degrees of indirection in this
configuration, everything works to the degree that's acceptable for most people.
Not too long ago, I started deploying openindiana / virtualbox and
decomissioning vmware esx. I haven't bothered, but if I want to eliminate some
of the indirection, I'd run the NTP server directly on openindiana. Bypass
windows server and the virtualbox guest addititions that are necessary for
keeping the windows server accurate.
Depending on the hypervisor you're choosing ... Even if you have special needs
... You might be able to continue providing NTP from the hypervisor just as you
previously provided from the OS that you're now going to virtualize. For that
matter ... Depending on your time source hardware and your hypervisor, you
might be able to USB or PCI pass-thru the hardware time source to a guest OS,
and continue exactly as you were before.
I apologise for a slightly belated reply. No special requirements, and for many
years NTP has run happily on our mail gateways. We only have a handful of
non-virtualised systems left. We're running on VMware ESXi, with EMC VPLEX
across two geographically dispersed data centres. I doubt we could run NTPd on
the ESXi servers, but even if we could, we would not want to expose them to our
local DMZ subnets. To pick up on points made earlier, I do not want to have to
run external antennas for GSM or GPS, and GSM reception in our data centres and
comms rooms is poor (and one is two floors below ground level). I will look at
mini-ITX systems and see how they compare price-wise against entry level 1U
servers from Dell/HP/...
Thanks!
Jonathan.
_______________________________________________
Tech mailing list
[email protected]
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
http://lopsa.org/
