High bandwidth servers?

[gulliver]

Currently looking at high bandwidth servers, I'm pleasantly surprised by the 100tb.com offer of servers at $200-400 monthly, on a 1gbit dedicated port with 100tb throughput, plus another $399 for unmetered.

If it's genuinely as advertised, it seems too cheap to be true.

I've no experience of them, and wonder why they seem able to offer such a spec so cheaply.

[jerom]

Currently looking at high bandwidth servers, I'm pleasantly surprised by the 100tb.com offer of servers at $200-400 monthly, on a 1gbit dedicated port with 100tb throughput, plus another $399 for unmetered.

If it's genuinely as advertised, it seems too cheap to be true.

I've no experience of them, and wonder why they seem able to offer such a spec so cheaply.

It is part of UK2 group. I think you could trust them, they must be having there own DC as well. read through reviews in forums and related websites before purchase.

[EastCoast]

UK2 don't have a great reputation. I've personally had serious issues with them in the past, I don't rate either their technical competence or their service. I recall there was a cdn that used their servers, that got 100's of servers pulled off line because they were consistently using high (but within their limits) bandwidth.

[PromptSpace]

Do you have any preferred location?

[dklynn]

Check WebHostingBuzz's dedicated servers as there are a lot of options including the Tb connection and they have data centers in the US and UK.

Regards,

DK

[gulliver]

Thanks to you all.

On the individual points...

Jerom - I've read good and bad reviews of them - much as I'd expect about anything.

EastCoast - if the stuff works most of the time I don't mind poor service. But your 'within their limits' remark troubles me... at their prices I'd expect overselling and small print, but they're clear on 'dedicated, unmetered, unthrottled port' - and that's potentially a lot of data (432tb?) and at a price other providers don't seem to be anywhere near.

PromptSpace - for location, probably US - because it's nearer for most of the likely clients and probably cheaper than UK/Europe.

dklynn - I looked at WebHostingBuzz and their service/pricing is similar to others, with much less bandwidth than 100tb.com - and with their upgrade at $20 per tb it soon gets uncompetitive when compared to 100tb.com. But, every other provider I've seen seems far pricer too.

So I'm back to where I came in - are they simply misleading, and not expecting many to reach such high throughput and therefore likely to pull anyone who gets high throughput (which, with streaming audio will be probable)? Or are they genuinely differentiating themselves and taking slim margins? Of course, some are happily selling (and buying) domains at $35 compared to the more-usual $8-12.

In a less-critical situation i'd simply try them and see. But that's not so here... as I'm doing viability work for a future project, one which has nice finances at 100tb.com rates but which could quickly become horrible if bandwidth costs increased significantly - and, if my numbers are right, in the case of WebHostingBuzz (who're simlar to others and cheaper than many), the additional 322tb @$20 makes a huge difference. (And yeah, I know it's daft to base viability on availability from a single source.)

Hhhmmm...

[mdrussell]

Thanks to you all.

On the individual points...

Jerom - I've read good and bad reviews of them - much as I'd expect about anything.

EastCoast - if the stuff works most of the time I don't mind poor service. But your 'within their limits' remark troubles me... at their prices I'd expect overselling and small print, but they're clear on 'dedicated, unmetered, unthrottled port' - and that's potentially a lot of data (432tb?) and at a price other providers don't seem to be anywhere near.

PromptSpace - for location, probably US - because it's nearer for most of the likely clients and probably cheaper than UK/Europe.

dklynn - I looked at WebHostingBuzz and their service/pricing is similar to others, with much less bandwidth than 100tb.com - and with their upgrade at $20 per tb it soon gets uncompetitive when compared to 100tb.com. But, every other provider I've seen seems far pricer too.

So I'm back to where I came in - are they simply misleading, and not expecting many to reach such high throughput and therefore likely to pull anyone who gets high throughput (which, with streaming audio will be probable)? Or are they genuinely differentiating themselves and taking slim margins? Of course, some are happily selling (and buying) domains at $35 compared to the more-usual $8-12.

In a less-critical situation i'd simply try them and see. But that's not so here... as I'm doing viability work for a future project, one which has nice finances at 100tb.com rates but which could quickly become horrible if bandwidth costs increased significantly - and, if my numbers are right, in the case of WebHostingBuzz (who're simlar to others and cheaper than many), the additional 322tb @$20 makes a huge difference. (And yeah, I know it's daft to base viability on availability from a single source.)

Hhhmmm...

I'm happy to provide some insight from a service provider perspective if it helps you decide. But first I'd ask - do you really need all of that bandwidth? If so, what will you be using it for? It's a lot of bandwidth for one server to push and there aren't many scenarios in which you'll have the physical server resources to do so. A download and maybe a streaming site are probably the only examples that will be able to push anywhere near that bandwidth without running out of CPU/Memory.

For a download or a streaming site, you probably want to look at a CDN solution anyway. There's options like CloudFlare, NetDNA and others that let you pay per GB or buy bandwidth in bulk that may reduce your costs and will definitely increase customer performance and server capacity.

Our customers aren't the typical 100TB type customers but I'd estimate only 10-20% use more than a few thousand gigs of bandwidth per month.

Matt

[mdrussell]

I think I read your post a little quickly as re-reading it, I do note that you mention streaming audio.

So I'm back to where I came in - are they simply misleading, and not expecting many to reach such high throughput and therefore likely to pull anyone who gets high throughput (which, with streaming audio will be probable)? Or are they genuinely differentiating themselves and taking slim margins? Of course, some are happily selling (and buying) domains at $35 compared to the more-usual $8-12.

Using the US as an example, bandwidth costs from major Tier1 providers will range from around $0.75 per Mbps (Cogent) to $5.00 per Mbps (someone like Internal). The theoretical throughput of 1Mbps fully sustained is 324GB per calendar month. You can see from the maths that the law of averages does kick in and some overselling will exist, even when using lower-quality, cheaper bandwidth providers.

This isn't necessarily a bad thing; overselling is commonly accepted in the shared hosting market and governed well, it doesn't pose a problem. And overselling bandwidth capacity is common place for your home/business broadband connection, just as it is your mobile/cell phone connection.

[gulliver]

Thanks, Matt.

Answering your points...

1 Yeah, I really do need all that bandwidth (as you note in your second response, it's for streaming audio). And I probably need more, as should become clear toward the end of this reply.

2 With Shoutcast and associated software, I've been told that a decent-spec machine on a 1 gigabit port can handle the projected load and push the theoretical max throughput of 432gb. (You mentioned 324, so maybe my numbers are out or you typo-ed.

But a one-third difference won't be critical to the viability - but seemingly-common rates of 8-12 cents (or higher), and even your own much lower rate, per gb would.)

3 I'm a marketer and hence don't have the tech knowledge, but would the extra flexibility performance of cloud and CDN services provide an appropriate advantage?

With streaming audio rather than multiple requests for other and smaller files, the server should be able to handle the load (and splitting it between two or even three would likely remove any potential problem). I don't need burstable, and the ability to serve from a few hundred/thousand miles nearer wouldn't greatly improve user experience.

And the pricing would make the project unviable. (Hell, with the required bandwidth it quickly gets that way even on regular dedicated packages.)

4 I mentioned 'probably need more' (bandwidth)... if my numbers are right, that 1 gigabit port can serve 7500 simultaneous users of a 128k stream.

Now whilst that size of audience is uncommon for Internet radio, it's still very small compared to the potential. I can wide-awake-not-dreaming project short-term (maybe an hour a day) peaks of ten-times that... so my requirement is relatively high.

At 100tb-com rates, I can afford ($7k monthly) to have 10 servers sharing the load, and sit there idling much of the time (although there's a not-yet calculated on-demand requirement too for other audio streaming/download).

I don't mind overselling, and accept it where expected - but that won't be with this project.

The promise of 'dedicated, unmetered, unthrottled port' at the stated pricing attracts me - but can it be sustained?

[PromptSpace]

Well, have you considered co-locating your server and see if the DC can pull up a special deal with you. Like stated, all hosts will limit your port speed to something like 10mbps when you cross a certain limit on the BW. Co-locating might be a bit huge investment but it will be an one time investment.

[mdrussell]

Well, have you considered co-locating your server and see if the DC can pull up a special deal with you. Like stated, all hosts will limit your port speed to something like 10mbps when you cross a certain limit on the BW. Co-locating might be a bit huge investment but it will be an one time investment.

Colo will cost a lot more, and 10Mbps is a long way off the 1Gbps that the OP is after.

[gulliver]

Well, have you considered co-locating your server and see if the DC can pull up a special deal with you. Like stated, all hosts will limit your port speed to something like 10mbps when you cross a certain limit on the BW. Co-locating might be a bit huge investment but it will be an one time investment.

I don't mind colo where appropriate, but that sholdn't be necessary for what's essentially a very simple requirement. As Matt suggests 'Colo will cost a lot more, and 10Mbps is a long way off the 1Gbps that the OP is after.'

And, with the 'dedicated, unmetered, unthrottled port' there shouldn't (legally) be any restriction on bandwidth.

Buying high throughput on an unthrottled connection genuinely isn't a black art - my qualms are simply a) why is so cheap? and b) will (and can) they deliver what's promised?

I'll just have to try it and find out.

[TimIgoe]

its cheap, cos its oversold on the premise that 99% of users just won't use the high amounts of bandwidth they have actually "paid" for. They should be able to deliver to a point, but as more users start to get towards the limits or at peak points of the day you may notice a decrease in throughput.

[gulliver]

Thanks.

[gulliver]

I need more info...

Matt mentioned costs of 'bandwidth costs from major Tier1 providers will range from around $0.75 per Mbps (Cogent) to $5.00 per Mbps (someone like Internal)'.

I don't know enough to understand this, and hope someone can elaborate.

To me, Mbps is a speed. I understand that a 1gbit port can deliver a max of 324gb per month. And I understand a price per gb/tb of data transfer.

But I don't understand how the Mbps relates to gb/tb.

And I've read that buying bandwidth on a per Mbps model rather than a per gb basis will be cheaper, but can't correlate the two.

[mdrussell]

1000Mbps = 1Gbps

[gulliver]

Thanks Matt.

Sorry, I'd poorly worded that - and I'm not sure this is any better.

What I meant is that I don't understand how the port speed relates to the total transferred - unless that figure of $0.75 per Mbps multiplied by 1000 is the price for an unthrottled port capable of transferring that 324gb?

[mdrussell]

1Mbps fully sustained throughput a month is around 324GB

So 10Mbps is 3240GB, 100Mbps is 32400GB and 1000Mbps is 324,000GB/Month.

Note that this is a theoretical maximum. You will start to see some loss/performance degradation above 80% usage on nic port / switch port so the actual throughput is going to be lower.

[gulliver]

Matt, I'm sorry that I'm still not clear...

So is the price for an unthrottled 1 gbit port $0.75 multiplied by 1000 = $750?

[mdrussell]

Gulliver,

Cost and price are two different things. Yes, that would be an accurate cost from a carrier perspective. There are several things that can bring the price down:

- Contention ratio - Assume 40 x 1Gbit servers in a cabinet. That cabinet won't have 40Gbit of connectivity back to the core and the core is unlikely to have 40Gbit of connectivity upstream of that. A more likely scenario is 10Gbit, or maybe less, shared between the 40-odd servers.

- Traffic shaping - some types of traffic may be filtered/QoS'd and given access to less bandwidth

- Peering - networks peer with each other, sometimes free, sometimes paid. Generally peering is seen as a way of offloading some bandwidth for less.


Matt

[gulliver]

Thanks, Matt.

I'm now understanding more - but not sure this is getting easier. ;-)

Clearly, I have more to consider - ahead of which, is it right for me to assume that 'guaranteed connectivity' (that's probably not the right term) is unlikley, and that finding a provider who'll guarantee 'yes, we can get all of that 320-odd gb out into the world in timely fashion' is impractical/impossible and expensive?

[mdrussell]

Thanks, Matt.

I'm now understanding more - but not sure this is getting easier. ;-)

Clearly, I have more to consider - ahead of which, is it right for me to assume that 'guaranteed connectivity' (that's probably not the right term) is unlikley, and that finding a provider who'll guarantee 'yes, we can get all of that 320-odd gb out into the world in timely fashion' is impractical/impossible and expensive?

I'd honestly look for a specialist streaming host, or a CDN that specializes in streaming radio so you know listeners don't get cut off.

I think it will end up being more expensive than 100TB. But why not offset the costs with some advertising? Doesn't have to be too intrusive but may justify the higher expenses. If you find a good CDN that can support all those listeners, you will only need 1 or 2 servers.

Matt

[gulliver]

Thanks. Good points. Appreciated.

[damoncloudflare]

"For a download or a streaming site, you probably want to look at a CDN solution anyway. There's options like CloudFlare, NetDNA and others that let you pay per GB or buy bandwidth in bulk that may reduce your costs and will definitely increase customer performance and server capacity."

We really won't help with downloads or streaming per se (recommend moving large downloads or streaming to subdomains that we don't touch in your CloudFlare DNS settings) because we only cache static content files by default.

Note:You can set more advanced caching rules with PageRules. But I would still recommend keeping streaming content running through a subdomain we don't proxy or cache...

[logic_earth]

Akamai? CDN used by a lot of major companies, I know Microsoft used them to stream their tech shows live, like TechEd and Build.