Frequently Asked Questions

Physical / Network / Other requirements for boxes
Out-of-band access / Power Cycling / Physical Access
Domain Registration / DNS servers / resolvers
/ SLAs / Other Colo
IP address assignment / Portability / Bandwidth
Things to know / Mailing list / Boasts

Q: What are the physical requirements for boxes hosted?

1U is 1.75 inches or 44.5mm. Boxes must be rackmountable in such a way that they do not impinge on space either above or below their allocated rack units, and rackmounts are to hold the server without its weight being supported by boxes below (within reason). This generally means slide rails mounted front and back. The rack kit must be installable without access to the sides of the rack, or space above and below the allocated rack units. Therefore rack kits which need rear rackears attached to the case before insertion into the rack or via the side of the rack are not acceptable.

There is a 600x600 and a 600x900 rack, boxes which can fit in either are preferred. i.e. a maximum of 53cm between the front rackears and the back of the box. So if you're buying a generic PC case, please buy one which will fit into either (eg get an ACM105 not an ACM103). An incentive to do this might be that the 600mm deep rack is in a cooler temperature room.

Q: What network card should I use?

You should have a good quality ethernet adapter, something like the Intel 10/100/1000 ethernet cards seem to work well. Two of the switches are Cisco 3500XLs, with the third being a Cisco 2960G-48 (the others will be upgraded in due course).

Remember to choose a good quality network card, linux seems to like DEC Tulip based cards, and BSDs work well with Intel EtherExpress Pro. Avoid where possible RealTek cards, these aren't designed for server environments. Cheap ethernet cards seem to be the LAN equivilent of a WinModem, rather then offloading much of the processing to the ethernet controller, you'll suffer much higher CPU usage, much less buffer on the ethernet card, etc.

Q: What other requirements are there for the boxes?

The box must power on and boot without a keyboard or mouse automatically when power is restored after a power failure. We've had two incidents of power failure in Telehouse in the last year, and obviously want to remove hands-on work after the event as much as possible.

Any LEDs on the server MUST be connected up where possible. As a general rule at least one steady light (ie power) and one blinking light (ie hdd or network) is a good idea. Boxes with all the lights disconnected may be prone to having the on switch pushed multiple times should there ever be a power outage.

Boxes should be labelled with at least the primary hostname, owner name, contact number and email address.

 

Q: How can I remotely adminster my box (out-of-band access)?

If its a UNIX box, it *REALLY SHOULD* be running a serial console on its serial port, this will be connected up to the terminal server so you can access it via ssh. Serial settings should be 9600/8/N/1. We can't currently send break via ssh over the terminal servers, so if you can bind 'break' to some other character string that might be useful for you.

The racks are being fitted with 48 switched power outlets (APC masterswitch) each, and power cycles can be requested by emailing support.

For Windows boxes (yes, there are one or two in the racks), you may wish to run a serial console as this does give some limited access to the command shell in the event of you losing network access.

Lights-Out-Management cards with ethernet ports (such as the Compaq Remote Insight card) can be accomodated, extra ethernet ports are available for GBP5/month..

For any 'schedulable maintenance', we'd like you to arrange to go in with James, Nick, Jon or Ryan present.

For 'box down' situations, we're usually happy to authorise you to go in alone.

 

We're now a Nominet member (tag: JUMP), so can register domains under .uk, click on the link at the top left.

Probably quite a few. But you're probably not making the distinction between two types (or functions) of DNS server: Resolvers (which you set in your /etc/resolv.conf) that recursivly resolve your queries, or the other type, authoritive servers, which 'host' domains or zones.

Most people run their own local resolvers, you dont have to run BIND to do this, there are smaller more secure software available (eg dnscache). Ask on the list and you may find people willing to let you use their dns cache.

Q: I've only got one server, will you host my DNS for me?

We can do, although updates are currently manual, so whereas you get the benefits of having your zones hosted on four geographically and topologically diverse servers, you lose the ability to update instantly.

 

Nope we dont have one. Impressive as 'five-nines' or even '100%' uptime guarentees are, when something beyond our control happens, theres just nothing we can do about it. We aren't priced to be able to afford SLAs, and consider that most SLAs are just 'gambling' - no one provider can actually expect 100% uptime in all circumstances, they just prefer to quote it and give service credits or refunds.

By not offering an SLA, we're just moving the 'risk assesment' on to you, it doesnt necessarily mean any worse uptime. Take a look at the various hosts in the rack on the availability charts on webperf.

The colocation facility is Telehouse Docklands, North building, perhaps the single most important buiding for the UK Internet: being the original home of, and containing the busiest LINX switches. More on the UPS, Fire Suppressant, and Security at Telehouse can be read on the Telehouse Website.

Not at all, we have a very cautious procurement policy, which often involves ebay for cisco hardware :) And a good knowledge of the industry means we know where to get the best rates for quality connectivity.

You could try hosting with Konnex (http://www.knx.net.uk/), or one of our upstreams, Intensive (http://www.intensive.net/)

Sure, theres a few small colos that we peer with, you could try:

Bogons (http://www.bogons.net/) Mostly based in Redbus Sov, they usually only accept sun netra servers.

 

As many as you can genuinely justify under the RIPE rules for allocations.

If you need three or less, this is easy and you can be given IPs on the shared colo LAN, if more, then we can help you fill in a ripe-219 (although we'd like you to go read that and the supporting notes - ripe-220, and attempt to fill it in the online form yourself first and email the results to us). Once you've done that pass it on to us and we can assign some space - or forward it on to ripe for approval if its outside our assignment window.

No. They are PA space IPs (Provider Aggregratable). If you already have your own PI allocation (Provider Independant), we can announce those for you.

If you're taking bandwidth based on GB/month, such as the usual colo offering, then its metered as the larger of the in or out GBytes in a calendar month. If you're taking a larger bandwidth chunk, measured in Mbps/month then its the larger of the 95%ile measurement of your in or out octets for that month.

Explaination of ninety-fifth percentile (95%ile) metering: Every five minutes, you take the value of the in and out octet (byte) counters for the customer port, you then divide the octet counter by the number of seconds since the last reading, multiply by eight, and end up with the average bits per second since the last reading. At the end of the month, you take all the InOctet bits per second readings, sort them in numerical order, kill off the top 5% of readings, and the highest reading thats left is your 95%ile of your input readings.. Do the same for the OutOctet counters, and you've got the 95%ile of your output readings.. Take the highest of the two values, and thats the 'industry standard' way of calculating bandwidth usage.

The easy way of thinking about it: Ignore the highest 36 hours usage in a month, and the highest of whats left, rounded up to the nearest Mbps is taken as your Mbps usage.

We have BGP sessions for transit with Level3 (AS3356), one of the Tier-1's, via Packetexchange, and from KPN (AS286), as well as multiple partial transit suppliers and peering.

Theres no reason why not, as long as ordering and payment for any circuits / cabling is arranged by yourself, and as long as you can provide ethernet termination at the racks in telehouse, or appropriate router cards to terminate the circuit.

 

Q: This all sounds too easy, am I missing something?

Probably. Theres a lot of maintainance required when you put your own server on the Internet. You become responsible for ensuring its security, that its maintained up to date for all the relevant OS and application patches. New worms and exploits appear on the Internet semi-regularly, so you have to keep on top of them before your box is comprimised. Should your box be comprimised, then the switchport will be shut down to prevent further activity from the affected system, and common sense will insist that you do a clean rebuild, as its never possible to be 100% certain that you've cleaned an infected system.

In general things that cause the most trouble are open mail relays (usually people check and secure these on first installing, but fail to check when reconfiguring mail servers), linked IRC servers (lots of kiddies like to DDoS these), ftp servers with public writable incoming directories (you'll quickly find yourself hosting all sorts of large files), or selling / casually giving out shell accounts (you're responsible for your users traffic).

When you bring your server along to go in the rack, it should be all ready; have all the LEDs connected, OS and applications installed and locked down to be secure for connecting to the internet, have IP configured with gateway, and probably most importantly, have serial console (9600/8/N/1) running on a serial port. Theres usually no screens & keyboards around, so serial from a laptop is the best way to reconfigure or debug should this be necessary, once installed, the serial can be connected to a terminal server which you can access remotly by SSH.

And the last thing to know about is payment terms: Payment is quarterly in advance, perferably by standing order, into the Jump account. Invoices will be raised 30 days before the quarter billed for, on 30 day terms. Account details will be on the invoice. If you were wondering, Jump Networks Ltd is a registered company in the UK, and is VAT registered.

Theres a 'jump-announce' mailing list, which the owners / admins of all the boxes are subscribed to on joining the colo, and a 'jump-discuss' list, which is opt-in, and can be used to collaborate with each other, and receive announcements. It also provides a helpful peer-to-peer means for people to report and debug problems.

Theres only very limited space left in the racks right now (7/Mar/2007), and once this is full there probably wont be any more, right now there is a total of 8U unspoken for.

In future, we'll continue to be able to offer new customers services which don't take up rack space, such as providing bandwidth.

Something along the lines of

This server is housed at Telehouse Docklands, a top UK data centre, with 100Mbit connectivity from Jump Networks, a Cisco based network with direct connectivity to over 40 peers at LoNAP, and one hop from access to hundreds of LINX members.

Care should be taken not to suggest that your server itself is peering with/at LoNAP, or that Jump is a LINX member - it isn't - we're one AS hop away.