Linode vs Hostgator VPS

What CPUs were the machines using? Even with this info it's hard to say for sure because you don't know how loaded each VPS host is - someone could have been executing something a bit CPU intensive on the Linode box you are on while there aren't many VPSes on the HostGator one.

There are a lot of factors that determine VPS performance so just be mindful of that.

You can always look at /proc/cpuinfo to see what kind of CPU your Linode has (note that the host has dual quad-core CPUs, but your node only has access to 4 cores), and that's probably possible at HostGator.

The linode can see 4 Xeon cores at 2.13Ghz, and the HG can see one core at 2.27Ghz. Linode runs Xen, and Hostgator runs Virtuozzo.

As far as I am aware, Xen isolates your VM from cpu usage of other VMs that may be running on the same physical box, which is something that doesn't happen with Virtuozzo.

Maybe what's really happening is that the linode instance has a given fixed share of cpu or I/O. In contrast, at hostgator this share is flexible. So if the physical box is idle, I might get more cpu or I/O throughtput. The drawback is that with my high cpu usage, I might interfere with other VMs in the same box.

I'm just gessing here, I'm not exactly a guru in virtualization :)

On a Linode:

$ wget http://www.kernel.org/pub/linux/kernel/v3.0/linux-3.0.4.tar.bz2
$ tar xf /linux-3.0.4.tar.bz2
$ cd linux-3.0.4/
$ make defconfig
$ time make -j5
real    3m25.995s
user    11m29.025s
sys     1m42.290s

So, 3.5 minute kernel build.

That's not really an apples-to-apples comparison. You're comparing a $40/mth HostGator package to a Linode package that goes for as low as $25.50/mth…

@Guspaz:

That's not really an apples-to-apples comparison. You're comparing a $40/mth HostGator package to a Linode package that goes for as low as $25.50/mth…
Yes, and we still crush them.

-Chris

@jiffier:

The linode can see 4 Xeon cores at 2.13Ghz, and the HG can see one core at 2.27Ghz.
If I remember correctly, both gzip and gcc are single-threaded. In that case, one fast core might appear to give you better performance than four slightly slower cores. But you'd only be using 25% of your Linode's potential, whereas the HG box is already maxed out.

@hybinet:

@jiffier:

The linode can see 4 Xeon cores at 2.13Ghz, and the HG can see one core at 2.27Ghz.
If I remember correctly, both gzip and gcc are single-threaded. In that case, one fast core might appear to give you better performance than four slightly slower cores. But you'd only be using 25% of your Linode's potential, whereas the HG box is already maxed out.
gcc is, but caker used -j5 which tells make to use up to 5 concurrent threads. Obviously this is bound by dependencies (there may not always be 5 different files to compile if a dependency is causing a bottleneck), but it's generally an improvement.

For parallel compression, there are already drop-in replacements. To replace gzip, there is pigz. To replace bzip2, there is pbzip2 (a Canadian inwention). I believe pigz is parameter-compatible, while pbzip2 seems to merely implement most zip2 parameters but not quite all of them. I think both of them are in the ubuntu repositories at this point.

caker, you nailed it! As you guys point out, the linode really shines in handling concurrency:

Linode:

# time make -j5
real    37m54.608s
user    100m25.946s
sys    17m40.254s

Hostgator:

# time make -j5
real    76m18.904s
user    64m29.440s
sys    11m39.400s

So this numbers are a WTF.. Linode is cheaper and can handle load much better than Hostgator, probably because the kernel can see 4 cpus? I don't care If HG is faster, concurrency is what matters if you're running a webserver, DB or something.

I'm impressed now. Well, I think I know what to do with the Hostgator VM :)

Two questions:

a) The better the linode (512, 768, 1024), the more processing power? Is that true? Because my linode didn't compile the kernel in just 3min…

b) About virtuozzo vs Xen.. is it true that Xen (as oposed to Virtuozzo) isolates the load of one VM for the neighbours in the same physical box? I mean, when I'm compiling the kernel, my neighbours don't see any impact in their system, right?

Cheers! I've definitely become a Linode fan now! :D

To answer some of your questions:

a) If the box is idle, all levels have access to the same amount of CPU power. If the box is under contention, the larger plans have a higher guaranteed share since there are less VMs on each host.

b) Xen isolates a lot more than Virtuozzo (which some call a "glorified chroot jail"), and it does have better resource isolation, but there are some things you can't isolate, such as disk IO. I believe Xen does have IO priorities, but until everything moves to SSD (which is still cost-prohibitive for linode), IO scheduling is limited in effectiveness. But in terms of the CPU load you cause from compiling the kernel, it won't impact the other linodes on the host, no, not unless there is significant contention on the box, and even then, everybody gets a guaranteed share.

@jiffier:

a) The better the linode (512, 768, 1024), the more processing power? Is that true? Because my linode didn't compile the kernel in just 3min…

Probably your specific config. Note that caker's test used defconfig. I just ran the exact same commands as caker on my 512 and got:

real 4m35.553s

user 15m30.394s

sys 2m2.385s

So, about a minute slower than caker's, but 33 minutes faster than yours.