Linode seems to occupy a good middle-ground. You get much more for your money than Slicehost and it's a fairly large, well-established host with multiple locations. A $20 Linode for instance gives you 360mb of of RAM, 16gb storage and 200gb transfer. They also use more powerful CPUs than the Athlon's listed for Slicehost/prgmr - my Linode is on a quad-core Xeon L5420 (2.5ghz, 6mb cache).
Just guessing, but I'd expect disk access is a bigger bottleneck on highly loaded VPS servers than the CPU, so the worst and average case disk performance are probably what would concern me the most. I think prgmr is a fantastic deal - I certainly haven't seen anybody else offering so much RAM at that price point. I'd be a customer if they were located on the east coast.
Finally, for those who don't know, prgmr is owned and run by lsc who's a member here:
I agree about the disk access. Disk is by far the weakest part of my system. I do provide you mirrored disk, putting me ahead of the low-end ec2 stuff, but you are sharing that mirror with a lot of people.
Fortunately, more ram can usually make up for slow disk.
I am working on a zfs based network storage system that just might solve the problem and then some, but it's not anywhere near ready yet.
The cool thing about linode is that they hand out free upgrades often, usually ram and storage. I think the storage was 12gb before march for a 20 dollar linode and it got upped and if you look at their blog you'll see a large history of this usually about every six months.
Slicehost does however have very nice and useful articles and it is nice to support them if you use those.
Keeping an OpenVZ VPS running smoothly requires a lot more SysAdmin attention than keeping a Xen VPS running smoothly. The Johncompanies people, last time I tried them, were excellent.
OpenVZ is a thinner virtualization layer. everyone shares the same kernel. It's resource separation isn't nearly as strong. On the other hand, it is also more efficient. you can reasonably expect to get more total performance out of a OpenVZ system than the same system running xen, but the problem is with OpenVZ, if one user is hogging resources, that user is much more likely to effect other customers than they would be under Xen, with it's non-shared ram and stronger CPU scheduling.
Also, thanks in general for your HN participation and frank and open discussion of your hosting business. I find it really interesting and also helpful the way you classify different types of hosting needs and potential good hosting fits, ( that are not always your own company ;D ). Very cool!
I'm reasonably familiar with OpenVZ and other "container" approaches (zones, jails).
It sounds like there's a missing abstraction layer, or maybe just fewer administrative tools for these container approaches than the built-in-at-time-of-creation attributes of a true virtualized guest.
I do feel like slicehost "must" be overpriced these days, just based on the fact that their prices haven't decreased that I can remember. (At least ~2 years, maybe more?)
Not sure if prgmr is where I'd end up, but can't be all bad when "figlet prgmr.com" was all the time they spent on logo design. ;)
In that time, Linode have consistently increased their offerings at the same price-point and they deploy newer hardware. It seems a little odd to me that Slicehost seem to get so much more love in the hacker community and Linode remains the underdog. Perhaps Slicehost tends to attract more passionate users who market it more effectively to others? If Linode would get their backup service out of beta and clone the Slicehost API I would see little reason to prefer Slicehost.
I think Linode really needs to make their website more appealing. I know it shouldn't really matter, since as a customer I would use their slices not their web design skills, but it does. Their site has been getting better, but it still seems like they don't have someone there that gets how to tune a website to increase conversion.
I'm a long-time Linode customer, and I recently had to make the choice between Linode and Slicehost. I was extremely unimpressed by Slicehost's offerings compared to Linode - with Linode you get so much more bang for the buck, and they even upgraded harddisk space 3 weeks after I signed up for a new server. I can absolutely not understand why Slicehost gets more buzz than Linode.
Their beautiful copy/paste ready articles and fantastic support in their Campfire chat room is what sold me when I started with them. Recently, I moved over to Mosso.
I've moved to their sister product/company Mosso by Rackspace. It's about half the price & billed per hour... but bandwidth is not included. Bandwidth is billed based on usage ... and for low-traffic/new apps, that's a better deal.
> Much of what’s automated over at Slicehost (like provisioning a new instance) is handled manually at Prgmr.
(Which is also true for Linode). That might potentially be a big difference - I found it very quick to bring up new Linodes or Slices, and it's nice to know that that's a quick and easy possibility if you start generating more traffic.
Also... Calexico? Seems an odd place to be running a hosting business. It appears their actual servers are in San Jose... so where did he get Calexico from?
you have a good point about provisioning. my automation needs a lot of work, and that's a big weakness for me right now. I'm working on it, but I'm not anywhere near done. If you are built in to 'the cloud' and provision more nodes in the day than at night, yeah, my competitors are probably better for you. But if you leave your server on all the time, then I think I do offer some significant value.
I am much smaller than linode or slicehost... It's two SysAdmins, with some support from others. I have been around for a while, though. I've been selling VPSs with xen since 2005.
our old ___location is in he.net Fremont, our new ___location is in svitx in san jose. I'm looking at a rack in market post tower, but I'm told they've had power problems lately, while svtix has been pretty solid.
It is inaccurate. The servers are located at the svtix data center [1] in San Jose, CA [2]. That seems like a perfectly sensible place to operate a business to me.
this is correct for all new customers. I have a really old ___location at herculies data in sacramento, and another old ___location at he.net in Fremont.
I am planning on spreading even more. I'd like a rack in the UK or eastern europe. I do plan on getting rid of some of my current locations once I move the customers, though. but it would be nice to offer some low-latency options to our friends on the other side of the globe.
My next rack will probably be around here, though. Some place that I can get a cheap dark-fiber connection to from svtix, or maybe even within svtix, but I'd like to offer my customers the option of buying VPSs in different locations.
I am a slicehost customer, and a programmer without much sysadmin experience. To me, the documentation and support community at slicehost is invaluable - something this article touches on but doesn't put as much importance into as raw performance. Of course, depending on your level of Linux admin experience, saving $12/mo. may be worth it to you.
right now, my system is not designed to be easy for people who don't have *NIX experience. (I do have some neat features for those who do, like pvgrub and console access)
I'm working on it, but yeah, if you need a easy control panel to quickly re-image your system or things like that, I don't yet have those things, and yeah, for some people, getting those things is unquestionably worth the price difference between my competitors and I.
But for people who do have more UNIX/Linux experience, I think my system has some advantages that I will be careful to keep as I try to make easier interfaces. I give you a rescue disk that lets you boot your vps even if you break your boot config. I give you a ssh public-key based out of band console and rebooter. I let you run whatever kernel you want (I have a few customers running NetBSD, and I have some experimental OpenSolaris domains)
really? i think with today's hardware/bandwidth prices, it's perfectly reasonable to setup a fairly busy site and not worry about revenue at all. If it only sets you back a few dollars a month, it might make sense for you to build something popular for the experience and the social capital.
It might only be $12 but it's also a 150% increase. I think when you're comparing apples and apples (as in two products that have roughly the same features-- not a Rolls Royce and moped even though they share a basic function) there are going to be very few situations where a 150% increase is justified.
I'm on Slicehost, with about a $100 a month bill. Given my technology needs, that is overpaying -- both for the number of servers I bought (I could consolidate to save) and the price per server.
And I just couldn't care less. That's like one night's worth of sales. I've (hopefully) got better things to do with my time than spending it preparing for a migration and inevitable breakage in the interests of saving myself an hour's salary a month.
A) Slicehost isn't an overnite operation. Not saying I don't wish Prgmr all the luck in the world, but still ...
B) vCPUs don't matter as much as the guest/host contention ratio does. If I gave you 4 VCPU's and stuck you on a host with 128 other VMs .... you performance is going to suck no matter what. The fewer the number of guests, the better the performance of the guests.
Now only if Slicehost would stop sending me email I've got credit with them even if I have used 'em for over a year ...
heh. I've been selling Xen VPSs since 2005, and I sold FreeBSD jail based VPSs for a while before that. you can call me small, understaffed and cheap, but I've been in the market for longer than many of my competitors.
Heh. Nice. I remember when ec2 came into public beta I thought it was going to wipe the floor with us VPS providers. Their prices looked good at that point in time.
contention is... more complex than that. are you dedicating a core to the dom0? (it helps network performance rather a lot) the fewer VCPUs you have per DomU, the more DomUs can run at once. (on an 8 core box, with one cpu pinned to the dom0, I can run 7 1vcpu DomUs and one dom0 (for network/disk io) at one time. Alternately, I can run 1 4vcpu DomU and one Dom0)
On the other hand, if nobody else is trying to do anything, the box with 4 VCPUs has 4 times the resources that the box with 1VCPUs has.
VPSFarm, unlimited bandwidth and awesome staff. I called them from Cambodia and told them I wasn't be able to pay this month since Paypal locked me out of my account (I logged into paypal from 3 different countries in less than 48 hours, and they didn't like it.) I didn't get around to paying them for another month and a half. My service was never interrupted.
I love a business with a human face :-)
[Currently on linode, for the easier resource upgrades since I'm getting steady increase in users. Also amazing service.]
FTA: "Community ... If you want to use Prgmr you are pretty much on your own."
That alone is reason enough for me to stay with Slicehost.
This article is weighted towards performance aspects. When judging any service provider, I'd hope people here are wise enough to consider other factors as well.
I've heard great things about Linode and may try them if I ever get a chance–but between Slicehost and Prgmr, nothing I see would make me consider switching (I have 9 slices with SH).
I may be off base, but even though Slicehost is more expensive, that doesn't mean they aren't a better value.
1. Slicehost's support is great. Quick responses to tickets and very helpful people in the chatroom
2. Slicehost's articles–while free to anyone–make me want to support them for the willingness to contribute knowledge.
3. Their DNS tool is very handy
4. I'm sure the prgmr.com team is very competent and I hope they do well. But their site does little to instill confidence in me.
I agree that for some people, linode and slicehost are a better value. (btw, I've heard nothing but good things about either one, I mean, aside from the pricing. Both, as far as I can tell, provide excellent service, and both are easier to use for non-sysadmins than I am.)
My core audience is SysAdmins who need off-site servers and hobbyists, and I think I serve those markets better than my competitors. I am trying to build systems that will allow me to better serve people who are not as familiar with Linux (or bsd)
The website still needs a lot of work. But that is consistent with my goal of serving existing customers before new customers, if there is not enough time to do both.
As a longtime slicehost customer, thank god for competition. Slice has been slightly overcharging for memory (my current bottleneck). I also don't use any of slicehost's (admittedly great) tools.
I'll probably think about reevaluating my server costs (about 300 mo) if slicehost doesn't boost their memory offering in the next few months. Especially using chef for scripted
infrastructure, all I'd have to do is change my ip list and rerun.
Thanks, I was keeping an eye out for a way to cut costs for some non-mission-critical stuff - this looks very interesting (that's a BIG price difference for the larger plans). Will give it a try.
Yeah. Ram, disk and CPU are cheap. Now, dealing with abuse, etc... is expensive, but most of the per-customer costs are the same no matter how much ram they order, so I think it makes sense to drop the cost per gigabyte as you order more ram.
the formula I use to calculate the 1 month pricing is this: $1 for every 64M ram plus $4 per account. I think it works out reasonable; I was figuring $512/month for each 32GB / 8 core hardware, which is maybe a little high, but it is within reason.
> The cutover from Cogent to Rippleweb was messy, and resulted in nearly a full day's down time This was completely my fault, (That is, Luke Crawford's) as I set the cut off date for the old service to early. Rippleweb moved quickly to help me recover; the rest of the down time was due to a faulty cross-connect. All customers get one month of credit for this, as this has been a rough month (and we are not entirely out of the woods yet.) On the upside, we have avoided renumbering. Neal has survived the baptism by fire, and we can expect prgmr.com services to improve in reliability now that I have help again.
Take the servers at hetzner.de, they don't charge much more than a VPS but you et much more power. The only real advantage of a VPS when you grow is when they offer you tools to automatically deploy new instances (like amazon ec2). Otherwise you should be switching to dedicated servers.
If you need more than 4gb ram, buying your own server is usually a better deal than anything else. parts for a core2quad with 8gb unbuffered ecc and 2 hard drives usually set me back around $700-$800. 1u co-lo runs from $50-$100 a month.
so why do they charge $199 per month for core2duo with 2 gb ram and 250 gb sata? i really dont get it. server prices are this low but cost for dedicated servers are very high.
lsc, can you say a bit more about your # of VPSs per box, average cpu usage by user and disk contention. in your system if there is available extra cpu and one user wants that he cant, can he? your prices seem to be too good to be true comparing to others. is there any drawback?
EC2 kicked my ass when they started their public beta. that's right, something like $70 a month for 1.7GiB was cheaper than I was. I upgraded to new hardware and lowered prices.
Under Xen if there is available CPU, any user can burst to it. I use the credit scheduler, which means you spend scheduler credits to get that CPU. You gain scheduler credits by not using as much CPU. Your weight determines how fast your credits accumulate, so a ___domain with twice as much weight, assuming contention, gets twice as much cpu as another ___domain. I weight my domains based on the amount of ram they have, so a 64Mb ___domain has 1/4th the weight of a 256MiB ___domain.
The ceiling on how much CPU you can use when nobody else wants it is set by the number of vcpus you have. I give you one, and in this example, slicehost gives you 4. this is in an effort to make the jokers who run mprime all day (I'm looking at you) from dragging down the service for everyone else. And they don't. I see some users who regularly use as much cpu as they are allowed, and it doesn't cause problems for my more restraned users.
my standard system has 32GiB ram and 2 quad-core 1.9Ghz opterons. (I have one with 64GiB and 2 2.1Ghz shanghais, but most of that ram is being used as cache for my experimental ZFS ___domain, so it shouldn't have more contention than the 32GiB servers)
As for the number of domains, depends on how many users sign up for what plan. 62 domains are on the system that the user who ran this test is on.
as for dedicated servers, the big question in my mind is "why don't they give you more ram?" there's no reason to run a single socket with less than 8GiB and a dual socket with less than 32GiB (well, not no reason, but in the general case. Ram speeds up disk, usually, which is horribly slow, even if you buy expensive disk.)
I've found an interesting model for renting servers; I charge a kinda hefty setup fee (like $500 for a core2quad with 2 disk and 8GiB ram. within a couple hundred of the parts cost) but then charge $70/month ongoing. (for one of my 8 core 32GB ram systems, with some disks thrown in, it was $1200 up front and $175/month)
It does lock the customer in, which is not providing value, but on the other hand, if they stay for a while, it's much cheaper than market rates, and I'm ok with it because the customer has covered most of my capital outlay.
I think the thing is that most of us small operators are leery of capital outlays that take many months to repay, and I think the large operators actually pay more for the hardware.
No, I'm serious. every large company I've worked at bought pre-built servers from HP or Dell or whatever, and they always ended up paying a large premium over what I would pay buying parts (when they are on sale) for servers that did the same thing. I think part of it was that salesmen are professionals, and part of it was that in a large corporation, the guy making the buy decision, well, it's not his money. He's more concerned with making a decision that looks right than a decision that is right.
On the other hand after making an especially bad mistake I once said "Hey, that was my Porsche."
If you're spending less than $150-200 it seems to be hard to beat a VPS. Certainly there are cheap dedicated servers to be had, but I don't believe you'll get such a high quality network and standard of support as with Linode/Slicehost. $150-200 seems enough to get a decent dedicated server with a respected host such as Softlayer, or 10tb.com who resell Softlayer servers at crazy prices. That said, I have heard a number of people recommend Hetzner.
I can only recommend hetzner. We got > 10 servers there and are very happy. In June, they will have their new datacenter ready with new server offers. If all goes well, I will order another > 30 servers for our data processing.
I don't think it's a fair comparison. Slicehost grew very quickly then got acquired. Prgmr seems to be the smaller guy growing slowly, getting rack space http://wiki.xen.prgmr.com/xenophilia/2009/03/colocation-in-s... and trying to grow to a full rack. The automation is a huge plus especially if you are in the hosting business. The pricing model and the added features are what sets the competition apart from EC2. I think many of these VPS providers have to catch up and come up with some of their own "Why choose me" features, services if competing with EC2 is their focus.
I think comparing me to slicehost or linode is completely fair... they are my direct competition. I think ec2 is in a different market. there is some overlap, but really if you want to leave your instance on all the time, a vps such as I sell (or linode or slicehost) is vastly superior to the instance with unmirrored disk ec2 provides you.
the opposite is true as well; if you need instant provisioning... if you need 50 servers during the day and 20 at night, I simply can't do that for you. (I hope to be able to at some point, but I can't yet.) Ec2 has been doing that for some time, and I think slicehost and linode both have similar functionality.
It's very nice to see one nice VPS host that supports Paypal and appreciated by HN. Slicehost and Linode don't support Paypal.
Coming from a geographical region where credit cards are rarely used, this matters too me.
@Isc: you should probably have an FAQ page about the payment methods and other common questions. People usually don't check the signup page to know the payment methods.
yeah. I need to do that. The billing system badly needs work. We improved the provisioning system a bit yesterday, but we still have a backlog from this attention. Fortunately, I've got plenty of capacity, so the only problem is the setup.
This depends on the availability of a suitable API. Unfortunately, providers haven't decided on a common one yet. Sun were working on one (http://www.tbray.org/ongoing/When/200x/2009/03/16/Sun-Cloud), although perhaps other providers will end up just cloning Amazon or Slicehost's.
Check out Cloudkick (A YC company) for migration between Amazon and Slicehost.
Just guessing, but I'd expect disk access is a bigger bottleneck on highly loaded VPS servers than the CPU, so the worst and average case disk performance are probably what would concern me the most. I think prgmr is a fantastic deal - I certainly haven't seen anybody else offering so much RAM at that price point. I'd be a customer if they were located on the east coast.
Finally, for those who don't know, prgmr is owned and run by lsc who's a member here:
http://news.ycombinator.com/user?id=lsc