Hacker News Reader

Comment by barredo | Hacker News Reader

I have a question about VPS, hope someone could help and not be much offtopic:

I have a dedicated server (in Softlayer) with Intel(R) Xeon(R) CPU X3220 @ 2.40GHz, 4GB RAM, 250GB (which I only use 5 or 6, so it's not an issue) and 2000Gb transfer. Price is 189$ a month.

Is it worth to switch to Linode's 4GB ram plan? I was thinking in getting a new server in softlayer with 12-16GB ram.

tl;dr: is it worth switching from a dedicated to a vps with similar specs?

View on HN

We've done some performance and price comparisons of different providers including Linode and Slicehost (a.k.a. RackspaceCloud). In that price range there are a lot of options you could consider with cloud providers. Although cloud is generally (but not always) multi-tenant, many provide dedicated CPU cores so CPU performance is very predictable. One advantage to cloud also, is that oftentimes the OS runs of an external SAN so if the physical host machine does fail, it can be immediately migrated to another host with very little downtime. There are other benefits like backup images and instant upgrading.

http://blog.cloudharmony.com/2010/05/what-is-ecu-cpu-benchma... http://blog.cloudharmony.com/2010/06/disk-io-benchmarking-in...

As someone who sells VPSs for a living, well, I'd want a pretty significant discount to take a VPS over a dedicated server of the same specs. The big problem with VPSs is I/O. If you have a 1TB disk and you split it so that two servers are using it, each server, sure, gets 500gb, but each server gets /way less/ than 1/2 the I/O performance. When you have more than one server hitting the same disk, all your sequential transfers become random transfers, and random transfers, especially on the SATA that nearly all VPS providers use, suck.

That depends on what % of your cpu you're using, and how sensitive you are to disk bandwidth. Look at what fraction of the host machine you get ("How many Linodes share a host", http://www.linode.com/faq.cfm ), do you use more than that fraction of the resources that would be shared (disk I/O, CPU time)? How much is having a nice web interface for rebooting and such worth?

> Linodes of the same plan are grouped together onto a host. We adjust the number of slots on each host according to its resources and hardware specification. On average, a Linode 512 host has 40 Linodes on it. A Linode 768 host has on average 30. Linode 1024 host: 20 Linodes; Linode 1536 host: 15; Linode 2048 host: 10; Linode 4096: 5.

Does this mean that a 512 host has 40*512MB = 20GB of physical RAM? If it is less than that, then where does the 512 figure come from?

I believe they have 24GB of physical RAM (with the remainder going towards dom0 and the hypervisor).

I find that... unusual. the hypervisor eats 64MiB, and few people use more than 1GiB in the Dom0.

I believe the current Xen hypervisor uses about 128MB now, but yeah — I sincerely doubt it's out of strict necessity, but out of wanting to ensure not to overload each machine with too many nodes. They wouldn't bump down to 22GB because I assume their mobos are dual-channeled.

Or, they're fibbing a bit by using that "on average" language, and they actually have more like 42 or 44 nodes, but wanted the nice round numbers for the marketing copy.

Edit: Oh, I didn't check the username. I imagine you'd know better than I.

Depends on how process hungry your server is, if you constantly maximizing quad core xeon server, then its not a direct comparison in processing. But if you are using processing speed of a dual core, then I think you should move.

They might give you four "virtual cores" but its not a direct comparison to a xeon server. In everything else, memory space and transfer, you are better off moving.

at $30 per month you'll get $360/year from switching, is that more or less the cost of the time to switch?

a part from this, linode service and facilities (such as instant backup and restore) are great, don't know about softlayer's

if you're concerned with CPU performance at all, then stick with dedicated. VPS is performant, but i think you can yield a decrease of 10-30ms in pageload times from dedicated.

On a well run VPS, CPU is rarely the problem. Often it's better than a dedicated, because it's a much bigger box, and you can often burst to use more cores. (of course, if everyone else is using a lot of cpu, then CPU can start to suck, too... but my experience has been that CPU is usually not a problem.)

It's disk I/O where VPSs start to really suck.

Reply on news.ycombinator.com