An unmentioned benefit of a dedicated server is that you aren't forced to use Linux. Many support FreeBSD, and a few will allow you to run anything you can install; OpenBSD, NetBSD, x86 Solaris, or even Windows.
I agree with you, but to be fair, NetBSD and OpenSolaris run quite well as a paravirtualized xen guests. (irritatingly, most providers don't support them. Even I don't support OpenSolaris, due to bootloader weirdness I haven't quite figured out how to safely integrate into my system.)
There are also obvious security benefits to running on unshared physical hardware, as Theo likes to point out.
AFAIK you can run FreeBSD and of course Windows on KVM
The only requirement is your CPU's full virtualization support. A lot of CPU recently are capaple of hardware virtualization, except low end processors.
you can under HVM mode in xen, too. Full virtualization, though, in it's current state, is inferior to paravirtualization. Significantly so. This may change as hardware advances, but right now, you are much better off going with something that can run paravirtualized.
Full virtualization, though, in it's current state, is inferior to paravirtualization
Do you have benchmarks to back it up ? Is performance the reason to choose Xen over KVM and other virtualization technologies for your hosting company ?
reliability even more than performance. KVM was not even close to production ready in '05. HVM mode under xen was also crap. I expect 3 year uptime; running some 'qemu with hardware support' (which really is what both xen hvm and quemu are) just could not deliver that at the time.
Also emulated devices leave a much larger attack surface from the domU -> dom0 host.
I am sure that KVM has improved a whole lot in the last few years, but most of the people using it seem to be using it for 'primarily use box as hardware, spin up a virtual to test something' (this, I think, is why the kernel guys like it so much... that is what they need, and KVM is much better than Xen if that is your use case.) I really need a 'dedicated virtualization server' - the hardware only exists to service virtuals.
If I had some time, I'd do some benchmarks; I'd be interested to see what the backends the kernelnewbies bench uses... for example using tap: or phy: is sometimes an order of magnitude less bad than using file:// (of course, sometimes you get caching going in the dom0, and file:/ looks faster, but it's still a really bad idea.)
I did start with FreeBSD jails, but switched to xen after I found just what a bad idea sharing pagecache is.
I can personally attest to the fact that ServerPronto, at least, lets you run OpenBSD. (Of course, you'll have to upgrade from their very old version yourself, and I presume support won't know much about it. But it works, about as well as can be expected for the price, and most virtualization doesn't.)
EDIT: it turns out that the OVH deal isn't so wonderful. That's if you commit to a term. The month to month price is more expensive than advertised.
First, Slicehost hasn't changed their prices since they opened up shop over three years ago. Hardware costs have come down a lot since then so their margins are growing fat.
Second, that provider that you mentioned happens to be the best deal on the list.
Third, if you read their information on it, you see things like "Unusable due to very high and unpredictable storage access latency." Specifically stating that latency "sits at around 90ms." and that peak speed is 8MB/sec. That's one of the OVH offerings. The one you specifically mention seems sketchy: they will only sell to you if you have a mailing address within their area. So, no US sales at all and if you live outside of Ireland, you have to use their site for your country. Odd at best.
Fourth, Slicehost has RAID to protect from hardware failure and these servers usually just have one drive which means a hardware failure means restoring from a backup with downtime.
Fifth, OpenVZ isn't the same as a Xen host and, apparently, can easily be oversold ala shared hosting. Many of the good deals there are OpenVZ VPSs.
Most of the others aren't such amazing deals. A 768MB server with a Celeron for $40 isn't such a bargain today. I mean, clearly it's a good deal, but not amazing when compared to what's out there in the VPS space. And they do link to lowendbox.com which has VPSs cheaper than Slicehost/Linode. And just like this site, those VPS providers aren't the cream of the crop like Slicehost and Linode.
--
Really, there are VPS providers that are nice and cheap now-a-days. I'm currently using ChunkHost which has a free beta program for a 512MB VPS which costs $19 (about half the price of Slicehost and $37 for 1GB and $66 for 2GB). I'm sure there are more that I don't have experience with. Slicehost is now the Rackspace of VPSs (literally) and charges prices accordingly. It isn't a terrible deal.
It'll be hard for low-end dedicated servers to compete with VPSs because of the cost of rackspace and power. I mean, if you have a 5 year old Pentium 4 box that you want to repurpose as a server (and a lot of low-end dedicated hosts used to do this with desktop hardware), you're talking about using 150 watts or more and 4U of rackspace for one customer. Compare that to getting a SuperMicro twin 1U and putting 2 quad core 32GB servers in 1U and yes, probably using more power, but not 64x more power - probably in the range of 5x more power - and you can see how it just can't compete.
I mean, now you have at least 10 servers in 1U using half the power and 1/40th the space. At best, you now have 64 servers each using less than a tenth the power and 1/256th the space.
Really, the site even says it: "So why on earth would a person want such a crappy dedicated server? In most cases you don't - go get a Virtual Private Server instead." There are a few reasons to get a dedicated box which they note, but not a ton.
--
On the high-end dedicated side, dedicated boxes still look pretty good because VPS providers want high margins. SoftLayer will sell me a box loaded with RAM and bandwidth for a lot cheaper than most VPS providers will. But that's on the high end. And there's little reason for VPS providers to want larger clients. Why would you give a volume discount? If 16 people are willing to pay $40/mo for 512MB, why sell 8GB for less than $320/mo? Clearly there are some billing efficiencies and support efficiencies to justify a slightly lower price, but RAM is RAM and doesn't cost less if you slice it up less.
So, that's the real position for dedicated boxes in this economy: when you're going to be using the majority of a VPS box. Otherwise, the power and space efficiencies of VPSs usually win out.
I love how technology costs always come down over time. Server4You has $39 USD actual dedicated with .75 RAM - not bad. Although, I always shy away from AMD processors, maybe because of the horribly slow experience I had on a friend's Gateway computer back in the day. Is this just me?
I've looked at their prices, and I wouldn't call them insanely cheap. I'm with Hetzner (both rented and an empty rack with own hardware) and the prices are about the same.
I find this a bit confusing. I run a hosting company myself (unbelievably shameless plug: http://asmallorange.com ) and the overhead in maintaining smaller and underpowered servers is relatively massive. On top of the financial aspect, there's also the matter of reliability and performance. Why run your stuff on inadequate, antiquated hardware?
If you're looking at this end of the spectrum, just go with a VPS/Cloud system. You'll be paying about the same amount, maybe less. And you get drastically improved reliability if your provider is using failover and isn't using crap hardware.
Basically, would you rather take a business class seat on a jet or charter a 50 year old plane being piloted by a drunk hobo to fly somewhere?
eh, sounds like this guy is selling atom boards, which are relatively modern. Now, they don't use ECC ram (and I fail to understand why he doesn't throw in 2GiB ram rather than one; if he put int 2GiB, he'd be competitive with my 2GiB VPSs.) and they are slow, cheap hardware, but speaking as someone who sells VPSs for a living, for some things, having a box to yourself is much better than a vps, even if the cost per performance is higher.
Now, my experience with the atoms (and why I'm not selling 'em) is that they use more power than you expect; 50w each or so, down from 75w each for a core2duo with 8GiB ram. So I'm guessing this person is also in a region where power is cheaper than it is for me. (I am in the SF bay area, and thus I pay enough for power that margins on a power-hungry atom system would be rather small.)
You can get the operations overhead per machine down with a good pxeboot/automated provisioning system. some larger places I've worked, provisioning 5,000 machines was a relaxed afternoon of work. Yeah, physical servers are a bit more work than virtuals, but they don't have to be that much more work, if you have a good system.
That said, getting a VPS on modern dual-socket 8 or 12 core boxes with ecc ram usually gives you more bang for your buck, if the security tradeoffs are worth it.
soon, I hope. all the parts I don't buy locally (I buy hard drives locally due to a history of packing problems; newegg, specifically, tends to damage oem drives in shipping, and it's hard to find retail-packaged "enterprise" drives) are in the garage, and I've got one side of the 2 in 1u up and passing memtest. the other side I'm still dorking with.
Right, which is why I wouldn't use them for production, other than firewall. I've just oredered a http://www.supermicro.com/products/system/1U/5015/SYS-5015A-... kit for 199 EUR sans VAT, add about 35 EUR for 2 GByte Kingston ValueRAM, a port riser for 10 EUR and a dual-port Intel NIC (e.g. E1G42ETBLK) for about 100 EUR, a little flash DoM with pfSense and you've got a nice firewall.
It's a question of price and low power. I'm currently running two mini-ITX C3 systems in parallel, which don't have ECC either.
I'm hoping most of the traffic will just remain on the Intel NIC (I'm running a transparent bridge), which is server-grade, so it will itself be not a source of corruption. The rest will be caught by checksums, hopefully.
> Why run your stuff on inadequate, antiquated hardware?
Sometimes the problem isn't about performance, rather it's about fault-tolerance and redundancy. Granted, if all of these servers are located in the same data center, redundancy takes a hit. But all else being equal, having one's software running on multiple machines reduces down-time.
Actually, that's the perfect case for a VPS. Most modern virtualization systems will automatically failover your system to another host node if there is a hardware failure. It's all done without downtime or any changes on your end.
While this is possible, it requires shared storage, which is a single point of failure, and can be quite expensive. None of the VPS providers I know of support live migration. I know I don't.
Also, live migration doesn't help you of the hardware dies. (I mean, if the host is using shared storage, it is easy to bring you up on a new host, but you still crashed. you can run a dedicated off shared storage, if that is all you want.) You can only 'live migrate' off a 'live' host. Now, if you wanna get really expensive, vmware (and xen, if you want to do some patching) has the capability to essentially 'live migrate every Xms' - the idea being that they never bring up the second host unless the primary host fails; but it's ready, and only Xms behind. (they do some things I don't understand to insure that I/O to the shared block device is in sync) This is very expensive, though, because not only do you need to pay for 2x the hardware, you need a vmware license. (or to run a non-standard xen)
A dedicated server with mirrored drives and ecc ram is likely going to be more reliable than a VPS.
> would you rather take a business class seat on a jet or charter a 50 year old plane
You have relatively smaller seats and have to share with other passengers if you take the jet, whereas you can take the whole crappy 50 year old plane to yourself -- it's dedicated!
Well. I'll still take the former, but apparently some people prefer the later.
> If you're looking at this end of the spectrum, just go with a VPS/Cloud system. You'll be paying about the same amount, maybe less. And you get drastically improved reliability if your provider is using failover and isn't using crap hardware.
Of course you roll your own, e.g with Linux VServer and HA (heartbeat/drbd). If Atoms were really low-power (they aren't, 8 W is noise in comparison to the chipset), offerend ECC support and decent mothboards with Intel NICs and IPMI onboard then they would be interesting hosts for virtual server guests.
As is, they're reasonable as firewalls (which are typically redundant) and SOHO servers. I hope ARM will deliver something competitive here.
It doesn't really make sense to me to pay more than $50 a month for something like this. For under $120 you can get an AMD Athlon X2 5000+ (2.2 GHz dual core) and a solid matching motherboard on Newegg. Throw in some old hard drives RAIDed together (I personally had about 2 TB laying around that I RAID 1+0'd together) and you have a server that will fill all of your needs WAY into the future. Once your service starts catching on and you need better than 95% uptime, convert your residential cable Internet service (what, like $40/mo or so?) into a $100 (?) business line (better upload speed and reliability), and you are good to go.
With hardware and bandwidth being so cheap these days, why are we still so hung up on VPS solutions? Personally-hosted solutions are much more manageable and the costs are quite similar.
I understand buying your own machine, but you still need to co-locate. $40/mo cable internet connections aren't suitable for hosting are they? I server up port 80 and various other traffic from mine, but it's minuscule, and I think if you tried to do much of it, your ISP would shut you down.
It's not really HTTP that is going to be a problem - as long as you are willing to live with the latency and low (upstream) bandwidth, that is - it's SMTP. Many mail servers will drop anything from a residential line without even telling you, and with (some) good reason(s).
Bandwidth is only cheap if you commit to buy large amount of traffic, like a 10Gb uplink. It's still expensive from single server level.
Take a look at EC2, i think income from bandwidth charge is the main source of profit. The base price of an instance only covers electricity, maintaince and initial hardware cost IMO.
Is $3.99 per mbps with a minimum commit of 10Mbps expensive? A 10Mbps would give you approximately 1250GB of traffic in a real-world scenario. See: http://uberbandwidth.com/pricing.asp
Depend on servers usage. $399 per month for a 100mpbs line is not expensive if you have for example a streaming server. But my sites consum about 1TB a month in total, so why pay for more when my need is only 5Mpbs ? (10Mbps gives you 3TB up and 3TB down if it's full duplex)
As technology advances bandwidth price will definitely go down. My point is that as of now bandwidth cost still represents a significant share of a server's total cost. 100Mpbs becomes cheap when you have a bunch of servers, like 15+. That's why low end dedicated servers and VPSs still have a market, hosting companies can buy wholesale bandwidth/hardware for cheap and then resell, though the margin is not that good.
43 comments
[ 3.7 ms ] story [ 91.8 ms ] threadThere are also obvious security benefits to running on unshared physical hardware, as Theo likes to point out.
The only requirement is your CPU's full virtualization support. A lot of CPU recently are capaple of hardware virtualization, except low end processors.
Do you have benchmarks to back it up ? Is performance the reason to choose Xen over KVM and other virtualization technologies for your hosting company ?
Some benchmarks after quick google:
http://virt.kernelnewbies.org/XenVsKVM
Also emulated devices leave a much larger attack surface from the domU -> dom0 host.
I am sure that KVM has improved a whole lot in the last few years, but most of the people using it seem to be using it for 'primarily use box as hardware, spin up a virtual to test something' (this, I think, is why the kernel guys like it so much... that is what they need, and KVM is much better than Xen if that is your use case.) I really need a 'dedicated virtualization server' - the hardware only exists to service virtuals.
If I had some time, I'd do some benchmarks; I'd be interested to see what the backends the kernelnewbies bench uses... for example using tap: or phy: is sometimes an order of magnitude less bad than using file:// (of course, sometimes you get caching going in the dom0, and file:/ looks faster, but it's still a really bad idea.)
I did start with FreeBSD jails, but switched to xen after I found just what a bad idea sharing pagecache is.
Eg picking one at random - celeron processor, 2GB ram, 500GB disk - 25 euro a month. Or with the same provider goto 4GB Ram and Quad Core for 40 euro.
First, Slicehost hasn't changed their prices since they opened up shop over three years ago. Hardware costs have come down a lot since then so their margins are growing fat.
Second, that provider that you mentioned happens to be the best deal on the list.
Third, if you read their information on it, you see things like "Unusable due to very high and unpredictable storage access latency." Specifically stating that latency "sits at around 90ms." and that peak speed is 8MB/sec. That's one of the OVH offerings. The one you specifically mention seems sketchy: they will only sell to you if you have a mailing address within their area. So, no US sales at all and if you live outside of Ireland, you have to use their site for your country. Odd at best.
Fourth, Slicehost has RAID to protect from hardware failure and these servers usually just have one drive which means a hardware failure means restoring from a backup with downtime.
Fifth, OpenVZ isn't the same as a Xen host and, apparently, can easily be oversold ala shared hosting. Many of the good deals there are OpenVZ VPSs.
Most of the others aren't such amazing deals. A 768MB server with a Celeron for $40 isn't such a bargain today. I mean, clearly it's a good deal, but not amazing when compared to what's out there in the VPS space. And they do link to lowendbox.com which has VPSs cheaper than Slicehost/Linode. And just like this site, those VPS providers aren't the cream of the crop like Slicehost and Linode.
--
Really, there are VPS providers that are nice and cheap now-a-days. I'm currently using ChunkHost which has a free beta program for a 512MB VPS which costs $19 (about half the price of Slicehost and $37 for 1GB and $66 for 2GB). I'm sure there are more that I don't have experience with. Slicehost is now the Rackspace of VPSs (literally) and charges prices accordingly. It isn't a terrible deal.
It'll be hard for low-end dedicated servers to compete with VPSs because of the cost of rackspace and power. I mean, if you have a 5 year old Pentium 4 box that you want to repurpose as a server (and a lot of low-end dedicated hosts used to do this with desktop hardware), you're talking about using 150 watts or more and 4U of rackspace for one customer. Compare that to getting a SuperMicro twin 1U and putting 2 quad core 32GB servers in 1U and yes, probably using more power, but not 64x more power - probably in the range of 5x more power - and you can see how it just can't compete.
I mean, now you have at least 10 servers in 1U using half the power and 1/40th the space. At best, you now have 64 servers each using less than a tenth the power and 1/256th the space.
Really, the site even says it: "So why on earth would a person want such a crappy dedicated server? In most cases you don't - go get a Virtual Private Server instead." There are a few reasons to get a dedicated box which they note, but not a ton.
--
On the high-end dedicated side, dedicated boxes still look pretty good because VPS providers want high margins. SoftLayer will sell me a box loaded with RAM and bandwidth for a lot cheaper than most VPS providers will. But that's on the high end. And there's little reason for VPS providers to want larger clients. Why would you give a volume discount? If 16 people are willing to pay $40/mo for 512MB, why sell 8GB for less than $320/mo? Clearly there are some billing efficiencies and support efficiencies to justify a slightly lower price, but RAM is RAM and doesn't cost less if you slice it up less.
So, that's the real position for dedicated boxes in this economy: when you're going to be using the majority of a VPS box. Otherwise, the power and space efficiencies of VPSs usually win out.
Looks slightly better value than Linode - a bit more generous on the RAM and disk side.
http://www.hetzner.de/en/hosting/produktmatrix/rootserver-pr...
It's really cheap (but only on the price! :-)
And the advertised prices have 19% sales tax included, so it's even cheaper for companies :-)
They used to have a killer of a deal (empty 19" rack for 99 EUR/month, inclusive 19% VAT) they unfortunately no longer offer.
They offer some insanely cheap servers, which makes me pretty skeptical. Does anyone know anything about them?
(I'm not affiliated with joes DC, but thinking about switching from slicehost)
I've looked at their prices, and I wouldn't call them insanely cheap. I'm with Hetzner (both rented and an empty rack with own hardware) and the prices are about the same.
If you're looking at this end of the spectrum, just go with a VPS/Cloud system. You'll be paying about the same amount, maybe less. And you get drastically improved reliability if your provider is using failover and isn't using crap hardware.
Basically, would you rather take a business class seat on a jet or charter a 50 year old plane being piloted by a drunk hobo to fly somewhere?
Now, my experience with the atoms (and why I'm not selling 'em) is that they use more power than you expect; 50w each or so, down from 75w each for a core2duo with 8GiB ram. So I'm guessing this person is also in a region where power is cheaper than it is for me. (I am in the SF bay area, and thus I pay enough for power that margins on a power-hungry atom system would be rather small.)
You can get the operations overhead per machine down with a good pxeboot/automated provisioning system. some larger places I've worked, provisioning 5,000 machines was a relaxed afternoon of work. Yeah, physical servers are a bit more work than virtuals, but they don't have to be that much more work, if you have a good system.
That said, getting a VPS on modern dual-socket 8 or 12 core boxes with ecc ram usually gives you more bang for your buck, if the security tradeoffs are worth it.
Right, which is why I wouldn't use them for production, other than firewall. I've just oredered a http://www.supermicro.com/products/system/1U/5015/SYS-5015A-... kit for 199 EUR sans VAT, add about 35 EUR for 2 GByte Kingston ValueRAM, a port riser for 10 EUR and a dual-port Intel NIC (e.g. E1G42ETBLK) for about 100 EUR, a little flash DoM with pfSense and you've got a nice firewall.
I'm hoping most of the traffic will just remain on the Intel NIC (I'm running a transparent bridge), which is server-grade, so it will itself be not a source of corruption. The rest will be caught by checksums, hopefully.
Sometimes the problem isn't about performance, rather it's about fault-tolerance and redundancy. Granted, if all of these servers are located in the same data center, redundancy takes a hit. But all else being equal, having one's software running on multiple machines reduces down-time.
Also, live migration doesn't help you of the hardware dies. (I mean, if the host is using shared storage, it is easy to bring you up on a new host, but you still crashed. you can run a dedicated off shared storage, if that is all you want.) You can only 'live migrate' off a 'live' host. Now, if you wanna get really expensive, vmware (and xen, if you want to do some patching) has the capability to essentially 'live migrate every Xms' - the idea being that they never bring up the second host unless the primary host fails; but it's ready, and only Xms behind. (they do some things I don't understand to insure that I/O to the shared block device is in sync) This is very expensive, though, because not only do you need to pay for 2x the hardware, you need a vmware license. (or to run a non-standard xen)
A dedicated server with mirrored drives and ecc ram is likely going to be more reliable than a VPS.
You have relatively smaller seats and have to share with other passengers if you take the jet, whereas you can take the whole crappy 50 year old plane to yourself -- it's dedicated!
Well. I'll still take the former, but apparently some people prefer the later.
Of course you roll your own, e.g with Linux VServer and HA (heartbeat/drbd). If Atoms were really low-power (they aren't, 8 W is noise in comparison to the chipset), offerend ECC support and decent mothboards with Intel NICs and IPMI onboard then they would be interesting hosts for virtual server guests.
As is, they're reasonable as firewalls (which are typically redundant) and SOHO servers. I hope ARM will deliver something competitive here.
With hardware and bandwidth being so cheap these days, why are we still so hung up on VPS solutions? Personally-hosted solutions are much more manageable and the costs are quite similar.
Take a look at EC2, i think income from bandwidth charge is the main source of profit. The base price of an instance only covers electricity, maintaince and initial hardware cost IMO.
Is $3.99 per mbps with a minimum commit of 10Mbps expensive? A 10Mbps would give you approximately 1250GB of traffic in a real-world scenario. See: http://uberbandwidth.com/pricing.asp
Yes. I currently pay about 12 EUR for a TByte of traffic, 100 MBit/s dedicated router (i.e. not shared switch) port.
Depend on servers usage. $399 per month for a 100mpbs line is not expensive if you have for example a streaming server. But my sites consum about 1TB a month in total, so why pay for more when my need is only 5Mpbs ? (10Mbps gives you 3TB up and 3TB down if it's full duplex)
As technology advances bandwidth price will definitely go down. My point is that as of now bandwidth cost still represents a significant share of a server's total cost. 100Mpbs becomes cheap when you have a bunch of servers, like 15+. That's why low end dedicated servers and VPSs still have a market, hosting companies can buy wholesale bandwidth/hardware for cheap and then resell, though the margin is not that good.