Hi,

We were contacted by a company with a BIG project.

They want a dedicated server who is able to send 200'000 Gb a month. The [ecommerce] website itself is not too big, but the customer want to lauch a TV-ads compain, so he expect millions visitors a month.

I don't think that such server exists anywhere and ready to go. I am considering a solution with few load balanced Xeon servers and 1 Gb unmetred connexion.

This website is still in projet but if we provide a good technical/financial solution, we have all the chance to win the deal.

Do you have any suggestion?

If you are an experienced provider for such projects, don't hesitate to MP.

Your best bet is to contact a few data centres and ask for help. They will tell you the best way to go about it and what connections you'll need.

However, i question whether 200,000GB is really what they need. Instead of just offering them what they have asked for do a little bit more. Research the company, find out what they will be hosting. If you can find out how much bandwidth they truly need, then you can offer them possibly a dramatically cheaper price. Im sure the company would be very greatful if you help them cut costs.

I strongly recommend that you do an impact estimation chart. That will greatly show the best solutions and the best path.
Like i said, 200k GB seems quite a lot even if they have millions of visitors. If its a huge download database then I can maybe understand, but just webpages with a few 1mb-5mb downloads, then 200k GB seems quite a bit much.

Well http://www.fdcservers.net/dedicated.html offers a 1 Gb server for $999/month, but that server would never be able to serve that much IMO, the specs are too low. Also, it's Cogent bandwidth and as such you might never be able to push through anything near the 32 TB no matter what server.

At any rate, I don't understand the 200 TB figure... I mean, let's say there are a million visitors and each uses 1 MB bandwidth, that's still only 1 TB, something you get with most dedicateds between 100 and 300/month anyway. And 1 MB is already rather high, if they visit e.g. a couple of pages each with no excessive media, their actual bandwidth usage should be a quarter or less of that meg.

OTOH millions of visitors at an ecommerce site would most probably break any single server, especially that it'll be in peaks and not evenly distributed. I'd suggest a load balancing environment with a couple of dual Xeons or dual Opterons, that should take care of it. On the quick, here is a link http://www.ev1servers.net/english/high_avail_clusters.asp but it's not exactly what I had in mind, the specs are too low (but it should give you an idea where the prices start).

Thank you stealthdevil. I also believe that 200'000 Gb is too much. I contacted the company manager with few questions to try to figure how much bandwith will be really used.

They want to start with a 100 Mbps for the first month then increase to 1 Gbps in matter of months. I don't know how they established this schedule, but I am waiting for more infos from them.

Going with a 1000-2000 GB dedicated server in a good datacenter is IMO better for them than a 100 mbit line, the first giving higher burst rates in peak hours.

For this you need lotsa load balanced servers in a big data center. 200K gigs is a boatload of bandwidth and only the best of data centers is gonna be able to pull this off reliably and no single server is gonna do it...

Originally posted by Dixiesys
For this you need lotsa load balanced servers in a big data center. 200K gigs is a boatload of bandwidth and only the best of data centers is gonna be able to pull this off reliably and no single server is gonna do it...
That's exactly it.

Also, for something like this, just contacting the big web hosting companies may not be the best idea. You may want to contact Hypmedia.com and other similar companies. Hypmedia has been around for 10 years, which is an eternity in this business!

I have no affiliation with Hypmedia, but I can reccomend them! :)

You're definitely looking at a clustered solution here, if only for reliability (under certain circumstances, I have seen single servers push several hundred megabits consistently). As for where you're hosting, you may want to consider if there's a targetted geographic area so that you can narrow your search down.

Sam

Let me just chime in here a moment.

When a customer comes to you with an impossibly big amount of bandwidth (for them at least) take it with a grain of salt and start smaller.

I ran into this a few months back. ;)

Look at ev1 unmetered

For 200.000 GB you'd probably need 20 or 30 clustered dual Xeons or something. But I think 200.000 GB is more than most DCs can handle so I'd recommend talking to some DCs before.

However, I think that company overestimates the ROI they will get from a TV campaign... They probably calculated that all 6 Billion people will look at the site at the same time LOL !!

Or if you calculate with only a million or two but think that 100-200 MB is realistic for every visitor, they should rather think about how to optimize the layout and content. ;)

Thank you for your help and valuable advices :)

I am going towards a reasonnable solution for this customer. 200000 Gb is too much and some countries never used such bandwith during all their history.

here my solution:

1 server with:

2 AMD Opteron 246
4 GB ECC RAM
2 SATA HD with hardware RAID-0
100 Mbps dedicated.
Full management + service monitoring every 4 mins.

The server will be hosted is the US as all our machines.

for the price we are not far from $3000 a month.

I'm not sure whether you wrote RAID 0 as a typo or you really meant that - in the latter case, don't do that, it's prone to data failure doubly as much as a single HDD. What you should do as a MINIMUM for such an ecommerce server is a SATA RAID 1 with two HDDs and a separate HDD on another channel for daily (hourly?) backups. If you are willing to spend 3000 bucks anyway you might want to get a SCSI RAID instead, with a hot spare drive included plus a separate HDD or array for backups.

At any rate, it's a bit surprising you opted for a single server, I mean if it's really tens of thousands of visitors a day and many of them want to shop, then even a couple of hours downtime could mean much more money in lost sales than a second 3000 bucks/month server (unless they are selling toothpick tenpacks for $0.99, that is ;)).

I don't think a single 100Mbit server is unreasonable. If they truely do end up pushing the server to it's limits then you can look at moving the database onto another server, if that still isn't enough you can round robin the DNS using 2 http servers, or more if needed.

Thank you for your valuable suggestions.

The website by itself is small. No more than 100 Mb. There will be an automated backup to one of our hosting severs. Every hour, the website will copied. So, if anything fail, we can restaure it fast.

Yes I wrote RAID-0 because as I have a backup, I prefer to optimize the server for maximum speed. The RAID-0 doubles the failure risk. In other hand, when a server with a RAID-0 sends 1000 Gb of data, that means 500 Gb trough each HD. So the failure risk is divided by 2.

With RAID-1, each disk supports all the load.

If the customer delivers more budget, I will add:

- 2 or 3 SCSI disks (36 Gb per unit)
- a double Xeon server as a backup server

Don't let yourself be misled through the word "RAID" in RAID 0, it is not really a RAID level because it has zero fault tolerance and no recovery mechanisms. If you want speed use 4 HDDs and run a RAID 10, it's the fastest RAID level and at the same time adds some data protection.

BTW FYI, RAID 0 is not that fast, especially not for what you want to use it. RAID 0 is best at writing huge amounts of data (like capturing a video), but in some cases it's actually slower than either RAID 1 or a single HDD. Actually, if it's about reading data from the HDD, RAID 1 is the fastest of all because every byte is contained on every HDD (you can make RAID 1 with more than two HDDs in case you didn't know it) and so the one which has the read head closest the the required sector can deliver the data. That's one of the reasons BTW for RAID 10, 0 sucks in read and 1 sucks in write.

If you have the money and want speed, what you shouldn't do is RAID 5, it's slower than any of the other three mentioned above, the main reason for RAID 5 is the cost effectiveness.

You need to seriously consider clustering the solution if it really will have 200,000GB of traffic each month. Setup a couple dozen Pentium 4 servers and balance the load using a couple CoyotePoint Equalizers. Also, be sure to cluster your database servers, so your dozens of frontend web servers can always get to the data.

Additionally, I would suggest setting up several clustered web servers just for serving the images and other statuc content. This will reduce the load on the web servers that are processing the dynamic content.

Anyway, I wish you the best of luck. It sounds like an exciting project.

Like others have said, talk to datacenters directly. They'll be able to give you realistic quotes--and quickly.

It sounds like some miscommunication to me :P

Also, when you get extreme requests, don't take it too seriously!

How did this go?

Tip: Get a good lawyer to make a good contract. Before you invest thousands of dollers and the customer wants something else

200,000GB Transfer is about 350-400 megs per sec?

That's like 3 OC3's or a little less than an OC12 (600ish) which is a whole lot of bandwidth!

I've heard of a single server serving around 70-80 megs per sec, but that's just something I've read that "can" be done. No single normal server would run this setup.

What's an OC12 run?

edelweisshosting,

I would be interested to find out how the deal went, if you do not mind sharing with us.

Originally posted by RambOrc
[B]Well http://www.fdcservers.net/dedicated.html offers a 1 Gb server for $999/month, but that server would never be able to serve that much IMO, the specs are too low. Also, it's Cogent bandwidth and as such you might never be able to push through anything near the 32 TB no matter what server.
"Cogent's bandwidth" is quite good. FDC Servers, on the other hand, is heavily over-subscribed. What happens when you have 3 servers on one gige uplink, each contending for 1Gbps of bandwidth? No one gets close to what they are paying for.

If you are thinking that you will get a dedicated 1Gb for $999/month than than you are crazy. Lets see that is under $1.00 a meg.

Of course for the price it is burstable/shared bandwidth and a good deal for many.

Originally posted by debrown3rd
200,000GB Transfer is about 350-400 megs per sec? 200.000 GB (195 TB) monthly transfer is 632 Mbit/s average, which logically in peak hours may need even more than 1 Gbit/s. However, it doesn't really make sense how would a website use that much bandwidth, unless it will do audio or video/audio streaming which would make it more logical.

As most people previously mentioned, I recommend edelweisshosting to start with something smaller and see how it will go.

How many people do we have on this earth? 5 billion? transferring 400mbs per second and if each page is 10 kbs means you need 40,000 requests a second. 2.4 million per minute and 144 million per day and 4.3 billion a month!!

Quite unthinkable for one single web site which is just starting off and doing TV advertsiements.... Does the company have any techs that could actually give them a clearer view of the situation?

like what a few of the other posters have mentioned. Get a signed contract first and some deposit because hardware is not a problem if you have the budget and enough expertise to do load balancing and clustering. I wonder if the company is really serious of what they are saying....

I don't know about "you" but for you to come to a webhosting forum and say a customer neededs to push 200TB of band.

Say's the site is on ecommerce (probaly HIGH usage of mysql).

To push 200TB you need at least 20 big servers (when I mean big I mean 4x processors at least 4/8 GB ram with RAID 10 SCSI drives) 15.000rpm

each server can be on a burstable 100 Mbit but the hole thing have to be clusterer on 20 server each on it's 100 MBits burstable conection with probally a fast intranet betwen than.

For best two nec cards on each server.

I'm thinking about 15 servers for DB processing and 5 for web.

With that you cold get something betwen 25/70 TB of band.

You are talking big big numbers....

I'm most certain that you need the DC up2date with this project and a personal project quote.


From your knowlage from your post I recommend you to REFUSE this jobs as you don't have/know our have at an ideia on what to due and there is a high chance of a errors on your side to provide what was requested.


Also you would need to test (Beenchmark) stress each machine and than the inteire "camp" to see if it works and how it works.


Any way good luck :)