Web Hosting Talk : Web Hosting Main Forums : Cloud Hosting : Rackspace cloud servers - As good as Rackspace dedicated?

[Xandrios]

Hi,

I noticed that rackspace now offers cloud servers for a fairly low price. This got me excited and I created an account, started a few instances..it seems to be running very well.

As most of you know, Rackspace has a reputation to be extremely stable, high SLA's, and..high prices. The prices for their Cloud/VPS servers are not high. Should I assume that stability has gone down as well? :-P

My idea is to use a rackspace cloud instance to run a MySQL server for a busy website. The website itself will run on dedicated servers across the country, however that requires the database to be centralized, hence a single central Rackspace cloud instance.

I hope this will work well. Rackspace has the reputation to be stable and fast, plus the option to allocate more resources when traffic or database size grows sounds ideal. It would eliminate the need for MySQL replication and such.

Question is: How are your long-term experiences with rackspace cloud servers? Are they indeed stable? Does anybody happen to have a hyperspin report to give me an idea of how stable they are?

As for speed, I ran the UnixBench tool on a 256MB instance, this resulted in a score of 39 points. This is roughly equal to a high-end Pentium 3. Not bad, altho it remains to be seen if these kind of speeds will stay during peak times. What are your experiences in this regard?

Thanks!

[FastServ]

From what I understand they bill you on actual cpu and disk I/O utilization if you go over a certain amount, so, watch out for hidden fees -- especially if you plan to run a busy DB server which will have lots of I/O.

[Xandrios]

Thanks Randy! :-)

I was not able to find anything related to this; only a set hourly rate and variable costs based on bandwidth usage. Do you have any link for more information? Thanks!

[tumble]

Since you are on there. toss in a database and check I/0 speed output ect.
http://sysbench.sourceforge.net/

Post up your numbers and lets check out thier MYSQL benchmarks.

[FastServ]

Originally Posted by Xandrios

Thanks Randy! :-)

I was not able to find anything related to this; only a set hourly rate and variable costs based on bandwidth usage. Do you have any link for more information? Thanks!

I might be confusing things (billing wise) with Cloud Sites.

Best to run some MySQL benchmarks though, I have heard that raw MySQL performance leaves something to be desired.

[sshepherd]

Rackspace Cloud Servers service doesn't charge per CPU cycle, however they're very expensive for bandwidth. I'm not seeing the advantage vs other solutions or competitors given some of the choices made in their cloud platform too.

Their Cloud Sites service is the one that charges per CPU cycle.

General advice is to separate out your SQL Server to a dedicated server via private VLAN for the best performance and savings. If your SQL load is going to be fairly low you'd be fine running that on your cloud server and moving it later if it got any higher. You just want to be careful about some cloud providers and their SAN file I/O performance and the impact that has on your read/writes.

[Winky]

Last time I checked, Rackspace cloud has no allowance for bandwidth. They have a utility style billing, meaning a certain cost per GB. This can add up to a lot. Make sure you make some accurate bandwidth projections before going this route.

[arisythila]

Yeah, They are VERY expensive for bandwidth. It's rackspace tho, What do you expect! :-)

[axcessi]

Why not use a VPS or dedicated server?

How would it eliminate your need for mysql replication?

I think you might be one of those people that are confused by the "cloud server" marketing term

[tallship]

Originally Posted by Xandrios

..it seems to be running very well.

That's good to hear

Originally Posted by Xandrios

As most of you know, Rackspace has a reputation to be extremely stable, high SLA's, and..high prices. The prices for their Cloud/VPS servers are not high. Should I assume that stability has gone down as well? :-P

Actually, they don't anymore, and yes, their network is aging and although stability is generally good on their nodes, quality has waned in the last couple of years.

Now, you do realize that there's no such thing as a 'cloud server', right? It's just a host, plain and simple. Rackspace will argue with you that it's a "Cloud Server, not a VPS", but in fact that's all it merely is, an OpenVZ VPS on a somewhat oversold node.

Here's a quick test for anyone considering services in the fable of the, "Cloud":

• Ask them if you can run Solaris, FreeBSD, OpenBSD, or DragonFly
• Ask the provider if you can do an install from your own iso image
• Alternatively, ask them if you can install your own template

If the answer is not to any two of those questions, then Run Forrest! Run! (Even if you have no plans on running anything but Linux, which of course is just fine).

That having been said, The reliability is much higher than many of the other startup small VPS companies actually offer. A high percentage of these companies are themselves customers of Rackspace, and have rented space in a rack where they put their own (or rented) hardware and install Xen or OpenVZ or ESXI, etc., etc.

These companies REALLY oversell, and you'll find that although a lot of them technically meet their 99% uptime SLAs, they're down all the time.

How is that possibile? Well, the 99%, if you read the terms of your SLA, doesn't include planned outtages, and they'll simply shoot off an email and you'll find yourself down five or six tiimes a month - instead of 3 or 4 times a year (which is still unnacceptable.

your best bet, if you can afford it, is to rent a half rack yourself, drop a box in, and break it into several pieces with Xen HVM, and if that isn't feasible , then ProxMox or ESXI. Then you can sublet the rest of the rack space out or add another box and charge people a fair price for their VPS's.

You'll actually do much better than break even if you simply tell all your customers that their services are 'self-managed'.

That whole "Fair use shared resources" schpeel is crap, and if you want proof just install a Zabbix agent and watch your blood pressure rise

You'll be logging into your VPS control panel seeing all that disk space, for example, or memory, or CPUs, and everything looks fine from there, while you're machine is actually getting slammed hard.

Originally Posted by Xandrios

My idea is to use a rackspace cloud instance to run a MySQL server for a busy website. The website itself will run on dedicated servers across the country, however that requires the database to be centralized, hence a single central Rackspace cloud instance.

yeah... sounds a little suicidal to me. First, I don't do MySQL, it has become junk. the good news is, MariaDB.org has a drop in replacement - Stronger, better, faster; just like Steve Austin, the Six Million Dollar Man

That's not why though. If you're incorporating such redundancy on the Apache front, then you need to do the same for your database server too. That should be a major premise of your architecture.

I hope this will work well. Rackspace has the reputation to be stable and fast, plus the option to allocate more resources when traffic or database size grows sounds ideal. It would eliminate the need for MySQL replication and such.

Originally Posted by Xandrios

Question is: How are your long-term experiences with rackspace cloud servers?

Not good. See above.

Originally Posted by Xandrios

Are they indeed stable?

Not really, and neither is Amazon - These are aging networks and they definately do not scale well for busy, mission critical applications. DNS and SMTP is fine (multiple MX servers and of course the RFCs dictate redundancy for DNS makes it work).

Drop your own hardware in a rack if you want stability (And control over your stability) - see above.

Originally Posted by Xandrios

As for speed,

Use bing (Not to be confused with Bing). You can get a copy here

http://fgouget.free.fr/bing/bing_src-1.1.3.tar.gz
Thanks![/QUOTE]

You're very welcome. I hope that helps

--
Registered Linux User #190795

- "Ask Bill why the string in [MS-DOS] function 9 is terminated by a dollar sign. Ask him, because he can't answer. Only I know that." - Dr. Gary Kildall.

[Orien]

Originally Posted by tallship

Now, you do realize that there's no such thing as a 'cloud server', right? It's just a host, plain and simple. Rackspace will argue with you that it's a "Cloud Server, not a VPS", but in fact that's all it merely is, an OpenVZ VPS on a somewhat oversold node.

Hm, are you saying that Rackspace is running OpenVZ? I definitely didn't know that.

[tallship]

Originally Posted by Orien

Hm, are you saying that Rackspace is running OpenVZ? I definitely didn't know that.

I actually have a reseller relationship with them, and softlayer, and on occasion, I set up some clients on those systems. I'm dilligent in interviewing my potential new clients, as to their needs, so I can assure them the level of service they require.

Most people either last a couple of months, and then they're gone, or they outgrow those types of services very quickly.

I got my IP blocks about twenty years ago under direct assignment from nic.ddn.mil - they're mine, and they're not going anywhere. I only provide services to serious enterprises that have mission critical needs, as it appears the OP does, and I let them know right up front that their services are self-managed, although I can indeed intervene if need be, contract permitting.

If they want more RAM, I give it to them, if they are running low on disk space, I give them more, and if they need a clone or clustering or other typical items that a growing company needs, I accommodate them.

As a provider, I need to have an idea of the needs of my customers even when they don't fully grasp it themselves. With a basic Rackspace so-called cloud server instance, you can't just go and do a kernel compile whenever you need to, yet some providers are beginning to provide PyGrub so you can can stay current in your Linux distributions indefinitely.

Best of all, I'm sitting right down on the pipe, one hop from the major peering points of all continents

They can come to me with whatever it is they want to run because again, it's all self-managed, they can build a clustermuck out of their hosts or they can have a nice, reliable service that runs for a couple of years at a time without needing to go down for a scheduled maintenance.

And if or when there's a major issue on their end, I'm an email or phone call or IRC chan away, able to intercede 24/7.

Case in point, I have a server in Sydney Australia that I use for some of my own mission critical needs down that way, and I rent it from a really good provider who has rackspace in Global Switch there. I like it because I'm one hop from it here in the states, it's a new datacenter (and the largest in Sydney), and I have a direct connection to pipenet as a result.

The other day, I was informed that they used up their /24 and had to go directly to APNIC since they now needed a /22. Guess what? I'm in the middle of a nightmare now, having to reconfigure various things on servers all around the globe, and most of those aren't even my own machines!

What am I supposed to tell those customers who depend upon the services I have running on that machine in Sydney? I'm finding out that many of them don't even know how to edit their iax.conf or sip.conf files and change the IP for the switch in Sydney!

Now, following about twenty minutes of downtime when the new IP address was taken up, and another upcoming twenty minutes of downtime in about a month from now when original IP is retired, I acutally have adequate time to fix all the phone switches for all the people I can find for free - but it's gonna be a MAJOR drag.

There's nothing wrong with Bochs, VirtualBox, or OpenVZ, depending upon the services you need it may be fine, but the OP has indicated an infrastructure such that it appears he should have things like Varnish out in front of his Apache Servers (Like Facebook uses), and more than one MySQL Server in a clustered environment, etc.

He's not going to need to be concerned with kernel compiles, but he might find himself in deep dookie when his database server, on random occasions, suddenly isn't available when important transactions need to occur.

There's nothing like bare metal, and if you can't do that, then certainly, I would suggest that a mission critical application not be hosted on a MASSIVELY marketed, resold service.

I know of several VPS providers that I personally find to be quite desirable, I won't bother mentioning any of them in this thread, and there are plenty of opinions as to whom one might choose.

But I'm always willing to chat about them privately and some are extremely affordable for even shoestring operations for the level of services they offer. My contact info is displayed on my profile and I can also be found in chan #tallship on freenode, which is my public general support channel. When I'm not there (Sleep beckons even as I write this), there's usually someone knowledgeable idling who can help too, and some of them are other independent providers themselves.

But yes, to reiterate, and answer your question of astonishment, they're simply selling OpenVZ VPS's

I hope that helps

--
Registered Linux User #190795

- "Ask Bill why the string in [MS-DOS] function 9 is terminated by a dollar sign. Ask him, because he can't answer. Only I know that." - Dr. Gary Kildall.

[dazmanultra]

Your best bet, if you can afford it, is to rent a half rack yourself, drop a box in, and break it into several pieces with Xen HVM, and if that isn't feasible , then ProxMox or ESXI. Then you can sublet the rest of the rack space out or add another box and charge people a fair price for their VPS's.

I can't disagree with this more. You talk about redundancy and reliability on the one hand, and then advocate putting all of his eggs in one basket.

Back to the main point of the thread, I would not advocate putting MySQL "in the cloud" purely because of the poor I/O performance. You could also have performance issues if there is a routing problem that resulted in increased latency between Rackspace's DC and your DC where your front end web servers are based.

[tallship]

Originally Posted by dazmanultra

I can't disagree with this more. You talk about redundancy and reliability on the one hand, and then advocate putting all of his eggs in one basket.

No I didn't, I recommended bare metal hardware for his farms - not VPS's, yet offered VPS's as a way to help pay for the physical machinery in the racks.

Try reading it again

Originally Posted by dazmanultra

Back to the main point of the thread, I would not advocate putting MySQL "in the cloud" purely because of the poor I/O performance. You could also have performance issues if there is a routing problem that resulted in increased latency between Rackspace's DC and your DC where your front end web servers are based.

He's going to have to put it somewhere (in Multiple places) that isn't in the same place as all the rest of his Apache servers

It's not a bad model, and is commonplace in the Enterprise - how you approach your tunneling is part of the solution.

--
Registered Linux User #190795

- "Ask Bill why the string in [MS-DOS] function 9 is terminated by a dollar sign. Ask him, because he can't answer. Only I know that." - Dr. Gary Kildall.

[rootserver]

What is cloud hosting?