Seeking Reviews of Colocation Facilities?

h_box asks: "My small company is looking for a colocation facility to host a few Apache servers and Oracle database boxes. Is there a definitive review site on the net for this type of facility? Really what would be helpful would be a review site that comprehensively reviewed candidates physical facilities, their quality of support, their network architecture, and maybe even tracked network downtime. For those of us in Southern California, it would be additionally helpful to make sure the site will stay functional throughout rolling blackouts. Don't forget financial status! We want the servers nearby, for convenience, although we are well aware that there are excellent and cheap colo facilities in the midwest."

Comments

this?

[VA Software]

http://www.colosource.com/colo2002.asp

How to review a colo site.

[billn]

There's a handful of important things to consider when assessing a colocation facility. Major questions to ask the sales droids:

Connectivity.
How many providers bring connectivity into the facility?
- Companies like Exodus have no small amount of their own backbone, but other companies, like Equinix, bring in large pipes from multiple providers, making it much akin to having colo right on a peering point. Having direct access to multiple providers gives you room to negotiate for the levels of service you need, and can provide.
How is connectivity delivered?
- Basically, how many fiber entrances into the facility? Will a single fiber cut or repeater failure blind you to the rest of the network?
Can you get out-of-band circuits or facilities?
- Out-of-band could mean a T1 into the back end of your colo cage, or something as simple as a dial-in to a console server.
How does the company handle peering, routing, and traffic engineering?
- How is their reputation, in a nutshell? Check around, see what other networking companies think of them. Now more than ever, intelligent routing companies are getting into the mix and really leveraging multiple peer environments (Equinix is one colo space that is doing this).

Two: Power and Environment
What kind of backup power is available?
- It's common to have a online UPS through which all power flows. In the event of a brownout or power failure, your hardware should never know. On the outside of that UPS should be two power feeds: City power, and generator backup.
How often is preventative maintenance performed on backup equipment?
- Sounds extreme, but it's important. Companies like GlobalCenter performed weekly/biweekly tests of all generator facilities.
In the event of an extended outage, what is the colo company's fuel situation for diesel backups?
- Don't laugh. Some generators can power facilities for a couple hours or more. For a widespread failure (Like the PG&E debacle in the Bay area a few years back), is there a process in place to make sure a fuel truck will arrive on time?
Average maintained temperature within the facility?
- Some devices just don't like room temperatures. If you're running big servers, ask your engineers what your environment requirements are. The differences across platforms are significant.

Security:
What's the reputation of the site security team?
- Big dirty question. I've done a lot of work in a lot of facilities, and it will shock and astound you how far you can go with a little knowledge and a forked tongue. Companies won't be quick to fess up to famous failures, but people who work there might. Make the sales droid bring an engineer, and shush the droid when you ask this question.
Access limitations?
- When can you get in? Who can you get in? Who controls site access? What happens when your liaison departs the company (yours or theirs)? What happens when someone tries to get in without proper credentials?
- True story. I didn't have access to a facility, and had to have my boss call in to approve it. Approval was done over the phone. Anyone else see where the grease can go here?
How is equipment access and movement controlled?
- Signed in and out with notation of serial numbers OR MAC ADDRESSES, hopefully. Who did it, when they did it, where it came from or where they put it. Log of event emailed to liaison at the time, would be nice, but probably too much to ask for. Cook your SLA well.

Support:
Can you get remote hands in the facility?
- It happens. You reboot a machine, it doesn't come back up. You're in Turkey, the box is in LA. Can you get a console to it to identify the error in the short term? These are often rolled in to sales agreements or SLAs.
How long does it take to get support?
- Do you call a call center or a tech center? How long does it take for them to find your information? How much longer to get someone on the problem? Time means money. Your money.
How does their escalation process work?
- Escalation is a big magic word. I've personally been involved in shaking a union president out of bed to get more techs on a sonet ring turn up, after we ran out of escalation levels with the telco. Know what happens when things are broken and people aren't getting things done. Mention rebates to the sales droid and watch them squirm.
Do they do their own post mortem work?
- It's the day after a night full of fast hacks and dirty reassembly in the wake of a power failure or peering outage, or anything else that's clearly the provider's fault. Who calls first thing the next morning, you or them?

I'm not going to get into things like uptime gaurantees and network security, because those are joint responsibility areas. If you're relying on the provider to account for all your uptime, you're already missing the point of it. Hope this helps. If you're doing something big and scary and have other questions, I may consult. Email me.