I'm looking at colo options in the Seattle area. Immediate need is space for a 1U or 2U server, then migrating to a 1/4 or 1/2 rack late this year or early next year. Absolutely no idea what bandwidth I'll require, though it's for a Web 2.0 app that needs to be prepared for the possibility of large traffic spikes.
Prices seem to start around $40-75 per U for per unit rack space, and $60-125 per Mbps (95th percentile). I've also been quoted $100-120 per TB (in + out).
I've been told that 1Mbps billed 95th percentile is roughly equivalent to 190GB transfer for a typical web server, or about 56% traffic efficiency. If correct, the per TB pricing looks quite a bit cheaper and less risk of paying for extended traffic spikes.
However, since 95th percentile seems to be the standard for colo bandwidth, I'm wondering whether there may be reasons I'd want to avoid per TB. Only issue that comes to mind is oversubscription. Thoughts?
The Per TB would probably work better for you if your expecting traffic spikes.
95th percentile tends to help out when you have consistent bandwidth of both inbound and outbound
Incase you didn't know 95th percentile means takes how much your using every 5 minutes, calculates it together, and the top 5% of spikes gets knocked off. So if you spike for 10% of the time you'll be charged for it.
1 mbps actually works out to around 320gb when averaged out all month. The main advantage you are going to have with the per gig method is to be able to count down your remaining gigabytes and know exactly what to expect. This would open the possibility to negotiating additional transfer with your provider prior to exceeding your original commitment which would save you some cash in case of a Slashdot attack.
At the same time, if you commit to 1 mbps with, for example, a full two day spike to, say, 10 mbps, your average will still end up being under 2 mbps for the month, so that could be more cost effective:
Per Gig Method
320 gb per month commitment divided by 31 days = 10.3 gb per day
Two Heavy Days at 103.2 gigs (10 megabits per second): 103.2 gb + 103.2 gb = 206.4
Add remaining 29 days at 10.3 gb per day to two heavy days: 298.7 + 206.4 = 505.1 gigs
Overage is 185.10 gigs at (say) $0.50/ea = $92.55 in addition to commited rate
95th Mbps Method
1 mbps per month commitment, same 10.3 gb per day
505.1 gigs total averages to 1.58 mbps (505.1 / 320)
Average Increases to 1.5 megs at (say) $59/meg, total cost $88.50
Many providers will cut you breaks on bandwidth as well once you get near the 10 meg mark, so don't think you will be stuck with these rates for the life of your application.
Last edited by TonyAdtaq; 10-19-2006 at 01:05 AM.
Anthony M. Faoro II - tmf [at] adtaq.com - (425) 444-8787 ex 7000 Seattle Colocation @ Adtaq Internet | Seattle, WA | Be Happy! AIM TonyAdtaq | GTalk tmf [at] adtaq.com
In your example, I'd actually be charged for a 10Mbps connection. Two full days at 10Mbps would mean 576 readings at 10Mbps, only 446 of which would be dropped as the top 5% of the 8,928 total readings for the month. So the next highest reading would still be 10Mbps, total cost $590. That's precisely what I'd like to avoid!
(Actually, that's not technically correct either, since 95th percentile is calculated based on the higher of in or out, whereas per GB is combined in and out.)
The 1Mbps == ~190GB approximation is based on comparing 95th percentile and total bandwidth numbers from MRTG for a couple sites. I'd be very curious to hear if other sysadmins are seeing similar ratios. If anyone is able to share their numbers, please let me know!
Well it depends what your hosting, if your hosting something that has the same usage inbound as outbound you'll see much higher GB usage but your 95th percentile won't be affected, if your primarily outbound then 95th percentile usually doesn't work as well and you'll see more like 190GB per 1mbps like you said.