Experiences with Joyent Accelerator?

Hi.

Anyone have experience to share about deploying production environment
with Joyent? TextDrive had a lot of instability problems I recollect
(possibly only with shared plans?).

Thanks for any input.

Morten

I’ve been with TextDrive two years, and I don’t recommend the shared
hosting. Stability has been much better over the last year, but
overall it’s been very difficult to get responses from support. My
feeling is Joyent/TextDrive’s heart is in the right place, but it’s
really all about the cool stuff, and not the fundamentals of quality
hosting.

All that said, the Accelerator is a one-of-a-kind setup in the
industry, and I am dying to try it because I really do believe it is
better than anything else out there. As far as VPS technology goes,
kernel-level support really makes the difference, and Solaris is by
far the most advanced in this area. Accelerator is leveraging
technology from the latest Solaris builds for a VPS that is both
burstable, scalable and more stable overall. No one else has anything
like it. I’d say it’s a pretty good bet for hosting your Rails app.
However, I would stick to just the web hosting and keep my email
somewhere else. Although my email account there has been pretty
stable, it’s high latency (email coming in is delayed by the spam
filtering from 2-20 mins), and when there IS a problem (such as being
flagged as spammy) they just don’t answer tickets in a timely manner
at all.

I’m glad to hear it because TextDrive really needs a kick in their
ass. It’s not enough for a hosting company just to have really cool
tech. They’ve got to get the fundamentals down, which has proven very
difficult for them in the first 3 years.

We talked a bit about it at RailsConf, but we’ve got a Solaris zones
based product being released in 2-3 months. Just as there are many
companies offering Xen based solutions, it will be nice to have a
choice when looking at solaris.

Thanks

Jesse P.
Blue Box Group, LLC

p. +1.800.613.4305 x801
e. [email protected]

On May 23, 8:40 am, dasil003 [email protected] wrote:

All that said, the Accelerator is a one-of-a-kind setup in the
industry, and I am dying to try it because I really do believe it is
better than anything else out there. As far as VPS technology goes,
kernel-level support really makes the difference, and Solaris is by
far the most advanced in this area. Accelerator is leveraging
technology from the latest Solaris builds for a VPS that is both
burstable, scalable and more stable overall. No one else has anything
like it.

I’m interested to hear what you perceive as the benefits of Joyent’s
offering -vs- a Xen based solution?

I’m obviously biased, but I’m serious about my interest. :slight_smile:


– Tom M., CTO
– Engine Y., Ruby on Rails Hosting
– Support, Scalability, Reliability
– (866) 518-YARD (9273)

I had to get a clients app up on a new Joyent Accelerator in one day.
I was hoping to have the latest and greatest software installed
already and was somewhat disappointed. I was expecting to have the
same setup Paul I. got in his blog post (http://www.oobaloo.co.uk/
articles/2007/4/23/a-little-capistrano-recipe-for-joyent-
accelerator). I did not have time to mess with upgrading to the
latest Apache or Ruby versions. I just upgraded to the latest gems I
needed for Rails and used Capistrano for deployment. The site is not
on a shared plan (http://www.heroesorganization.org).

Jeff

We’ve also had some reservations about the performance of Xen under
high server load. HP recently conducted a performance evaluation of
Xen compared to OpenVZ (aka Virtuozzo) and while Solaris obviously
isn’t in this report, the performance degradation that Xen faces upon
heavy server load really concerns me [1]. Then there are the Xen
bugs (as well as Virtuozzo bugs) that lead to random reboots of
servers. That kind of problem, even in a load balanced server
environment, really bother me. The burstability, easy duplication,
and other features in Solaris really make it shine for web applications.

[1] http://www.hpl.hp.com/techreports/2007/HPL-2007-59.pdf

Jesse P.
Blue Box Group, LLC

p. +1.800.613.4305 x801
e. [email protected]

On May 28, 12:06 am, “[email protected][email protected]
wrote:

On May 23, 8:40 am, dasil003 [email protected] wrote:
I’m interested to hear what you perceive as the benefits of Joyent’s
offering -vs- a Xen based solution?

Although I’m not the OP, I’ll take a stab at why Solaris’ VPS solution
interests me over Xen at the moment.

The company I work for builds semi-custom web app tools which we then
host for the customer, so we use at least one server for each customer/
product. So far, we’ve been using Xen-based virtual servers, and this
setup has been working great so far. Until last week…

One of our production servers stopped responding. And by server, I
mean the physical box – every single VM on it was down. It turns out
we were hit by a bug in Xen[1] that’s been open since 2005 and seems
to affect the very core of the architecture. The layman’s description
of the problem is that the kernel in one or more of the virtual
servers develops an issue with (IIRC) its interrupt timing and starts
spending all of its available CPU cycles logging a message that “time
went backwards”. This somehow “infects” all of the other virtual
servers on the same box. The servers don’t actually go /down/, they
just run out of processing power to handle any other task within a
reasonable amount of time. We have, thus far, been unable to determine
what specific conditions cause this behavior to start happening; so
even though it hasn’t occurred on our other (physical) production
servers yet, we have no reason to believe that it won’t.

Now, I don’t want to blow the issue out of proportion, so I won’t go
so far as to say that Xen is unsuitable and no one should use it.
However, this has definitely shaken our confidence in the platform;
and that is why I’m currently taking a serious look at Solaris.

Since we co-locate our own servers, this isn’t a Joyent vs. Engine
Yard (or other Xen-based host) issue, but it’s pretty much the same
question.

[1] Citrix Hypervisor - Server Virtualization and Management Software - Citrix


Regards,

John W.

I’m looing at these a bit more from the sidelines, but there are a
couple
appealing things about Solaris in general to me, which then the
virtualization choice is more of a follow-on… One is DTrace. This is
quite a cool tool for diagnosing things. The other is how system
crashes,
crash logs, and reboots work - it’s simply more robust here. ZFS is
also
extremely appealing. I do hope ZFS can be brought to Linux. I think
it’s
somewhat unfortunate that there is a Linux vs. Solaris battle. I happen
to
use Linux a lot, but Solaris has become much more appealing lately
because
of the tools and, at least in my perception, more robust handling in
very
heavy load and heavy hardware situations. If I were building up my own
setup I’d be looking very seriously at it, in fact, I’d probably be
using it
automatically, if it weren’t for the fact that I simply don’t know the
OS as
well/I’m not as comfy with it as I am Linux. Obviously you’ve got folks
like Google and others using commodity stuff, and Linux (and older Linux
from what I understand), and working just fine, but it depends on what
your
situation is. Most folks aren’t building their own hardware, and can’t
be
slapping in new boxes all the time.

As for available solutions and cases where you are hosting on someone
else’s
system, I think you have to factor in a lot more. I happen to be a
Joyent
customer, but based on my experience, I’m not sure I would choose them
to
host an important system. I would have to look at the accelerators more
closely, and understand the support a lot better. My experience is
solely
based on shared hosting, and that has worked ok for me as I mostly
experiment in that space, and want something cost effective, but for a
production system I’d be looking at a lot more. EngineYard is very
impressive to me, and I’d be looking at it very seriously for Rails
apps.
Part of that is not just technology and the available solution, but the
people and approach. Having met several Engine Y. folks, and knowing
who
they have on staff, and their dedication to the particular market
segment,
they are absolutley a top choice IMHO. It likely just depends on your
particular needs.

On 5/28/07, Jesse P. [email protected] wrote:

and other features in Solaris really make it shine for web applications.

On May 28, 2007, at 9:46 AM, John W. wrote:

Although I’m not the OP, I’ll take a stab at why Solaris’ VPS solution
interests me over Xen at the moment.


Chris B.
[email protected]

On May 28, 9:52 am, Jesse P. [email protected] wrote:

We’ve also had some reservations about the performance of Xen under
high server load. HP recently conducted a performance evaluation of
Xen compared to OpenVZ (aka Virtuozzo) and while Solaris obviously
isn’t in this report, the performance degradation that Xen faces upon
heavy server load really concerns me [1].

“We use the Xen 3.0.3 unstable branch”

I wonder why they did that a time when 3.04, 3.04-1, and 3.04 tip was
available?


– Tom M.

On May 28, 9:46 am, John W. [email protected] wrote:

The company I work for builds semi-custom web app tools which we then
spending all of its available CPU cycles logging a message that “time
went backwards”. This somehow “infects” all of the other virtual
servers on the same box. The servers don’t actually go /down/, they
just run out of processing power to handle any other task within a
reasonable amount of time. We have, thus far, been unable to determine
what specific conditions cause this behavior to start happening; so
even though it hasn’t occurred on our other (physical) production
servers yet, we have no reason to believe that it won’t.

What version of Xen were you using?

I notice that nobody has reported on that ticket since 3.02. We’ve
certainly never experienced that issue…

Current Xen is 3.1. 3.03, 3.04, 3.04-1, and an untagged 3.04 tip have
been available since then…

The thing we like about Xen is that you get your own machines.

They behave just like a real machine with respect to tuning, tweaking,
etc.

And, they’re Linux machines, which is, we believe, a good thing in and
of itself, while agreeing that Solaris is a very nice OS indeed.


– Tom M., CTO
– Engine Y., Ruby on Rails Hosting
– Support, Scalability, Reliability
– (866) 518-YARD (9273)

Is there someone willing to provide some feedback about hosting rails
apps on a joyent accelerator ?

Any feedback welcome.

regards,

– Thibaut

On 7/23/07, Thibaut Barrère [email protected] wrote:

Is there someone willing to provide some feedback about hosting rails
apps on a joyent accelerator ?

Any feedback welcome.

We’ve been hosting our main site on a Large accelerator for the last
four
months (www.bigfilebox.com) and it has been rock solid. It was easy to
set
up, and the performance has been great. The nicest feature is that
most of
our storage is on an NFS share that can be resized on demand, so as we
need
more disk space we just file a ticket and shortly after the disk space
is
increased.

I run my own personal websites and mail off a Medium accelerator.
Again,
that has been rock solid since I booted it for the first time, and
performance on it has been great.

So my own experience has been a hassle free, well performing Rails
infrastructure.

Marcus

I had a client who had a social networking site on joyent accelerator.
And in my opinion it was not best of hosting env i have experienced. I
have excellent experience with Xen based VPS providers rails machine
and engine yard. First of all i had problems configuring
mongrel_cluster script from init.d which i normally do on rails
machine. I had to set env path to some thing like /opt/… other wise
script just wont run. If i add my dissimilarity between linux and
solaris commands it was quite a pain to configure

On Jul 24, 11:37 am, “Marcus R.” [email protected]