There’s Good News and There’s Bad News
So the good news is yet to be determined – I’m sure there are lots of good things about the Rackspace’s move to OpenCloud… I mean seriously it has ‘open’ right in the title… that must be good. Right?! Unfortunately I’m here to report the bad news.
Here is a quick visual chart of the benchmark results:
As you can see the first generation Rackspace servers perform far better than the new OpenCloud serves. That is of course an understatement, basically it’s sucktown, U.S.A!!! Sigh…
I’ve been trying to engage Rackspace at every level I can to try to address this obvious downgrade, and hopefully avert a disaster to my beloved Rackspace! Here are some of the communications I’ve had thus far with the folks at Rackspace:
It was a pleasure chatting with you. The reason for this ticket is because you ran a benchmark test on your open cloud servers vs first gen servers. You have sent be a google docs page with your results, and it shows that the first gen servers are significantly better than the open cloud servers.
Although this shouldn’t be the case, and you were wanting to know if there was another benchmark testing suite that you could use to test the servers more. I am passing this over to a higher level tech to point you in the right direction. If you would like any additional assistance, please don’t hesitate to ask.
Thank you for your time.
Need an answer, quick?
Use rackspace chat, now available in the portal!
Great customer support as always! Now for the response, I’m sure it will be informative:
OpenCloud servers have different vCPU allocations than the FirstGen cloud servers, which explains the differing results you see when performing CPU-centric benchmarks. To view the vCPU allocation of our OpenCloud servers, simply go to http://www.rackspace.com/cloud/public/servers/pricing/ and expand the given server size.
A 512MB OpenCloud servers is allocated 1 vCPU, while all FirstGen Linux cloud servers are allocated 4 vCPUs (with the exception of 30GB servers that are allocated 8). Based on your results, the OpenCloud server performed as expected achieving 29% (over 25%) of the CPU performance of your FirstGen server.
Per our FAQ at http://www.rackspace.com/cloud/public/servers/faq/#overview_faq_11 the vCPU allocations were changed in order to unify the vCPU settings and scale CPU performance along with the size of the server.
If you have any other questions or concerns, please let me know!
Rackspace Cloud Support
… WHAT?! Wait, what? You’re telling me that you’ve downgraded my server and everything is as expected. WTF? Mark closed the ticket, so I lodged a complaint to Rackspace:
I am deeply troubled by the performance benchmarks I’m seeing form my new NextGen OpenCloud servers. I feel like my servers are being downgraded, and I’m ultimately being forced to pay more for the same performance.
I’m planning on publishing my benchmarking results and would like to give Rackspace an opportunity to respond and address my concerns – all of which I will publish, along with my commentary. You may reference Ticket Number: 604951 [above] for more details.
I have always received excellent customer support and value from Rackspace, but in this matter I do not feel that my interests and needs as a customer are not being met. I suspect that once more people are aware of these performance downgrades inherent in the NextGen OpenCloud servers they will think twice before becoming new customers. I, for one, have universally recommend to friends and colleges going with Rackspace cloud servers, that however will not be the case anymore.
The only satisfaction that I as a customer, and Rackspace proselytizer, would feel appropriate in this matter would be to charge the same amount of money for the same performance. If that means you assign 4 vCPUs as a minimum, then so be it – anything less would be a downgrade!
Thank you, Sean Conner
[UPDATE: Still no response to my original complaint as of October 17th, 2012]
Here is a chat I recently to probe about getting in touch with someone more directly, rather than the provided channels on the Rackspace Website:
Welcome to the Rackspace Cloud! My name is marshall.h, how may I help you?
marshall.h: hey there
marshall.h: how can i help today?
Sean Conner: Marshall, before you go, I have a gripe that I’d like to share with you.
Sean Conner: I’ve been experimenting with the new OpenCloud servers, and doing some benchmarking tests on them. And I’m sorry to say, it’s not good.
marshall.h: i hear that the legacy servers benchmark better
marshall.h: im not sure why that is
Sean Conner: one sec
Sean Conner: Sorry- call. So ya the new servers benchmark at about 27% of the old ones.
Sean Conner: It’s because the vCPU in the old ones are x4 and in the new ones are only x1
Sean Conner: My gripe it that I really feel like I’m getting a downgrade with the open cloud, and ultimately being made to pay more for the same performance.
marshall.h: yea i can understand
Sean Conner: I submitted feedback a week ago, asking for a formal response from someone with the power to effect change at that level, but nothing yet. Do you know of an effective way to contact upper management?
marshall.h: to be honest, just through ticket, or calling in
marshall.h: and asking to be escalated
Sean Conner: How high can escalation though that channel get? I would imagine the final tier of escalation through that route would be a systems administrator?
marshall.h: well we have managers that you can get escalated to, or account managers
Sean Conner: Do they have the power to assign 4 vCPUs to all of my OpenCloud servers and only charge me the same amount as my non-open cloud servers?
marshall.h: that i do not know of to be honest, i dont think they would be able to do that unless you get a larger server, which come with the x4 procs
marshall.h: like the 15gb servers
marshall.h: see even with the legacy servers
marshall.h: smaller instances dont use the whole 4 procs
marshall.h: for instance a 512mb slice will only use 3.1% of the 4 vcpus
marshall.h: with nextgen you use the whole 1x proc
Sean Conner: Well that’s puzzling, I don’t think that’s the whole story then – because when you test the old and the new side-by-side, the performance data would suggest what you said is in fact not the case.
Sean Conner: Anyway- it’s not important right now.
Sean Conner: I just wanted to the opportunity to vent a little. Thank you for listening to me.
marshall.h: if you need anything else, please be sure to let us know
Sean Conner: I will do, thank you. And have a great Saturday.
marshall.h: you do the same sir
Thank you for contacting the Rackspace Cloud Support Team
… To be continued
[UPDATE: I have some responses as of October 19th, 2012 — as follows]
The first gen and next gen cloud servers actually get the same amount of CPU cycles assigned to them by the scheduler, but in next gen, it is all on one virtual core, so some synthetic benchmarks which use multithreading can indicate a false result. I can assure you that you are getting similar to same performance as your first gen servers.
The reasoning behind this change is to simplify the cloud server platform, by having a single virtual core allotment for both Linux and Windows. Each server of a specific size gets a specific percentage of the host CPU, whether on first or next gen. The only difference is the lack of multi threading.
Thanks for being a Customer!
Benjamin [REMOVED], RHCSA
Rackspace Cloud Servers
Hello Benjamin, you’ll have to excuse me as I don’t fully understand how a benchmark can be “synthetic” — what would be a non-synthetic benchmark in this arena? Perhaps you could make some suggestions, preferably with clear metrics such as execution time.
I will attempt to come up with some examples of my own and post those on this ticket thread, as well as my blog post here: http://www.seancombinator.com/rackspace-firstgen-vs-opencloud
The other response:
We are reaching out to you regarding your recent feedback about the rackspace-firstgen-vs-opencloud server performance.
I have contacted you via phone at 646-450-9657 and was unable to reach you.
We would like to have our Tech Team talk to you about the Server CPU usage differences and to be able to answer any additional questions you might have.
We will continue to try to contact you and in the mean time if you have any additional questions or comments please update this ticket or contact us at Support: 1-877-934-0407
Cloud Account Manager
Hello Michelle, thank you for trying to follow up. The number I had in my account was out of date. My personal line is (—) — —-
That said, if you’re contacting me to rehash discussions already had I would prefer you address the real issue. To bring yourself up to date on my main concern you can read my blog post on the topic: http://www.seancombinator.com/rackspace-firstgen-vs-opencloud
After several back and forth phone calls and emails with Michelle and Patrick from Rackspace we agreed that the best thing to do would be post side-by-side results for FirstGen vs. OpenCloud servers for the larger sizes. The concession that Patrick and Michelle made was that while it’s clear that the FirstGen servers far out perform OpenCloud at the smaller sizes, performance is better for OpenCloud when you measure the largest server sizes available.
Here are the results of this investigation:
RackSpace Cloud Server Benchmarks – FirstGen vs. OpenCloud – Ubuntu – 4GB RAM
RackSpace Cloud Server Benchmarks – FirstGen vs. OpenCloud – Ubuntu – 8GB RAM
RackSpace Cloud Server Benchmarks – FirstGen vs. OpenCloud – Ubuntu – 16GB RAM
So basically these results confirm my original hypothesis that switching from FirstGen to OpenCloud means a very significant downgrade in performance, with the caveat that if you choose a 16GB (15GB according the pricing chart and Rackspace’s website) or bigger server configuration you will see a noticeable bump in some of the benchmark metrics.