Page 2 of 3 FirstFirst 123 LastLast
Results 16 to 30 of 33
  1. #16

    CloudSigma opinion from an independent cloud broker-dealer

    We have introduced several of our customers to CloudSigma, both those looking to do something simple, and those building complex architectures. The commercial way they are setup (choice of CPU, RAM, HDD and SSD, and 5 minute billing amongst other things) makes a lot of sense economically, and listening in on technical calls, I have heard clients say "Your platform rocks". Doing a price comparison with other providers is tough - if you just look at the CPU, RAM and Storage comparisons, they sometimes look slightly more expensive, but once performance optimisation has been done, our clients' experience has been that they are either slightly cheaper, or MUCH cheaper, depending on the use case. I have had a business relationship with CloudSigma for sometime now and have found them really professional and prepared to go the extra mile to get your business. For full disclosure, my relationship with CloudSigma is as a cloud broker-dealer for them. If you want that big discount for a 3 year commitment to use them, but don't want to prepay, we're who you come to speak to. We work with several clouds, so I am reasonably unbiased.

  2. #17
    Join Date
    Jan 2011
    Location
    Canada
    Posts
    934
    I agree that price comparisons are hard. It doesn't help that CloudSigma's virt is KVM and uses the standard linux scheduler compared to Amazon which has a very custom and locked scheduler running in Xen.

    From what I'm seeing, the hardware nodes are extremely lightly loaded (maybe trial accounts were segregated from the rest of the universe). So you get excellent bang for the buck as CPU spills over. My concern going forward would be that it also makes it hard to properly size your VM correctly when the node suddenly gets load. A general rule of thumb that one vCore is roughly a ECU helps, but I haven't been able to test that theory.

    I'll impose this caveat on my previous post saying it performs better.

  3. #18
    Quote Originally Posted by tchen View Post
    From what I'm seeing, the hardware nodes are extremely lightly loaded (maybe trial accounts were segregated from the rest of the universe). So you get excellent bang for the buck as CPU spills over. My concern going forward would be that it also makes it hard to properly size your VM correctly when the node suddenly gets load.
    I was testing thems performance and maybe you're right but my thinking that they have not so many customers yet in Las Vegas. Always good performance there. Get Zurich account and all bets off. Performance is 50% as good and changes from machine to machine (even on trial). I guess they have more customers cuz it's their first location.

    Robert, co-founder from CloudSigma answered this with claim it was due to split of drive from physical machine where instance was run. That does explain why disk performance is so low. But that is not answer to why CPU power is also 50%. I hear no response on this so I guess they might really overcommit their CPUs.

    They supposed to have new I/O system coming soon (but not SAN, so if it's not local, what will it be? well no matter what, I hope it fixes bad disk performance). But I guess no solution to slow CPU.

    I think new customers should go Las Vegas, you get awesome CPU but beware that it will degrade badly once they get lots more customer at that location.

    I did lots of benchmark and posted a very interesting and helpful conversation (Renich from LinuxCabal is so nice!) here:

    http://blog.woralelandia.com/2011/10...ou-cloudsigma/

    Btw their control panel awesome i think and complete control of servers and drives is super great. If they can fix their performance problems they will be hands down the only cloud you ever wanna use.

  4. #19
    Join Date
    Jan 2011
    Location
    Canada
    Posts
    934
    That's a good point about Zurich. I was testing strictly in Las Vegas.

  5. #20
    Join Date
    Jan 2011
    Location
    Canada
    Posts
    934
    juenca, looking at your unixbench results at Zurich, it looks like one vCore very roughly equivalent to one ECU when push comes to shove (maybe just a tad better).

    My results from EC2 in case you're curious

    m1.small (1 virtual core with 1 EC2 Compute Unit)
    Dhrystone 2 using register variables 5378544.9 lps
    Double-Precision Whetstone 925.9 MWIPS

    c1.medium (2 virtual cores with 2.5 EC2 Compute Units each)
    Dhrystone 2 using register variables 23458569.1 lps
    Double-Precision Whetstone 4263.9 MWIPS

  6. #21
    Quote Originally Posted by tchen View Post
    juenca, looking at your unixbench results at Zurich, it looks like one vCore very roughly equivalent to one ECU when push comes to shove (maybe just a tad better).

    My results from EC2 in case you're curious

    m1.small (1 virtual core with 1 EC2 Compute Unit)
    Dhrystone 2 using register variables 5378544.9 lps
    Double-Precision Whetstone 925.9 MWIPS

    c1.medium (2 virtual cores with 2.5 EC2 Compute Units each)
    Dhrystone 2 using register variables 23458569.1 lps
    Double-Precision Whetstone 4263.9 MWIPS
    OK, yah actualy that helps a lots. Which unixbench results are best to look at? I mean, unixbench gives 2 results scores when you have 2 more more CPUs. The first is always lower score.

    Can you say rough pricing of one ECU or the server sizes you posted here?

  7. #22
    Join Date
    Jan 2011
    Location
    Canada
    Posts
    934
    I'll always use the parallel tests because most things I do spawn processes galore or are multithreaded.

    My CS vs EC2 comparison table is below for one year contracts and assuming full utilization. It includes 8GB volumes, and about $5 for EBS IO. It also doesn't account for the VLAN or static IP so you need to tack those costs on. Spot (burst) prices are for the Las Vegas rates (0.011 CPU, 0.0154 RAM, 0.28 HDD)

    m1.small 1 YR - $36.25 vs $42.36
    m1.small spot - $29.46 vs $24.46

    m1.large 1 YR - $152.09 vs $155.34
    m1.large spot - $118.99 vs $83.76

    c1.medium 1 YR - $84.56 vs $80.06
    c1.medium spot - $61.66 vs $44.23

    t1.micro comparisons aren't really valid because of the way the scheduler throttles from 0ECU to 2ECU.

  8. #23
    Thanks for these price comparisons however there are a number of points to make:

    Firstly on CloudSigma you don't have fixed server instance sizes. As such starting with AWS and re-creating that server size in CloudSigma is a false methodology. What you should start by doing is creating a server that actually fits your needs in terms of CPU/RAM/Storage on CloudSigma THEN go and find the nearest fit on AWS. You'll find you will have to over-purchase at least one of CPU/RAM/Storage and perhaps more in order to fit your requirements. This is the purchasing waste that bundling introduces for customers.

    Secondly, most customers using burst are doing ad hoc computing but often re-using the same storage. It is a very rare use case that our customers use both CPU/RAM and storage both on burst. The more usual case is using storage on subscription and CPU/RAM on burst. Again, because most hosting providers over-provision storage in relation to CPU/RAM amounts, working backwards from AWS fixed sizes rather than working forward from your actually computing size requirements on CloudSigma introduces bias to the results.

    Finally, in CloudSigma you have a number of features that allow you to tailor performance to your needs. For example, you can vary the number of virtual CPUs separately from the aggregate core-GHz. For example, a 10GHz server could have 10 cores of 1GHz each or 5 cores of 2GHz each. This means you can match the threads to your application requirements. Likewise we offer VirtIO emulation which significant boost performance. If you run the benchmarks using VirtIO storage and networking you'll see major improvements with this change alone. We also offer SSD storage which provides a major performance boost for operations such as database work.

    Best wishes,

    Robert

  9. #24
    Hi Robert! I'm glad seeing you here!

    Quote Originally Posted by cloudsigma View Post
    Firstly on CloudSigma you don't have fixed server instance sizes. As such starting with AWS and re-creating that server size in CloudSigma is a false methodology. What you should start by doing is creating a server that actually fits your needs in terms of CPU/RAM/Storage on CloudSigma THEN go and find the nearest fit on AWS. You'll find you will have to over-purchase at least one of CPU/RAM/Storage and perhaps more in order to fit your requirements. This is the purchasing waste that bundling introduces for customers.
    I agree and thats why I really likes CloudSigma! But I think you must also accept that it makes much sense to making price point comparison at a CPU/RAM level that is equal between platforms. This lets you know if cost/resource ratio is similar or not.

    Other problem with your suggesting is that HOW do you actually propose to start by sizing a server on CloudSigma? If you read my benchmarks I posted on the link I listed you can see that performance changes LOTS depending on location (las vegas/zurich) and even depending on which machine your on at the same location! That makes sizing impossible.

    Don't get me wrong, I fell in love with the level of control in CloudSigma over resources and sizing and subscriptions!

    Quote Originally Posted by cloudsigma View Post
    Likewise we offer VirtIO emulation which significant boost performance. If you run the benchmarks using VirtIO storage and networking you'll see major improvements with this change alone.
    I did that but I didn't see many change between disk performance in both modes. But its said you will be making some storage system changes thats exciting! I only wonder why you don't respond to the CPU question.

  10. #25
    Join Date
    Jan 2011
    Location
    Canada
    Posts
    934
    Thanks Robert for coming out here and adding some salient points.

    Quote Originally Posted by cloudsigma View Post
    You'll find you will have to over-purchase at least one of CPU/RAM/Storage and perhaps more in order to fit your requirements. This is the purchasing waste that bundling introduces for customers.
    There definitely is much more flexibility in CloudSigma's provisioning. But the forced apples-to-apples comparison was just to compare relative price/performance.

    There are also performance variability downsides to packing VMs of disparate sizes, which are concerns alleviated to some extent via bundling. Bundling can be both good or bad.


    Quote Originally Posted by cloudsigma View Post
    Secondly, most customers using burst are doing ad hoc computing but often re-using the same storage. It is a very rare use case that our customers use both CPU/RAM and storage both on burst. The more usual case is using storage on subscription and CPU/RAM on burst.
    I think that muddles the point as a good deal of ad hoc computing is opportunistic. Keeping a volume on standby is just pure waste, especially if you're numbering in the dozens of servers. Configuration and coordination are typically left to things like Puppet or Rightscale, rebuilding the server from a base image. Hence, this is why the comparison was left as purely spot pricing.

    Besides, the difference in burst storage versus a one year sub was $1.40

    I think you're better served by mentioning that burst prices at CloudSigma are capped, as oppose to EC2 which goes as high as you're willing to pay - should it go higher, the instance is terminated abruptly. I much prefer CloudSigma's spot even if it is a bit more expensive.



    Quote Originally Posted by cloudsigma View Post
    Again, because most hosting providers over-provision storage in relation to CPU/RAM amounts, working backwards from AWS fixed sizes rather than working forward from your actually computing size requirements on CloudSigma introduces bias to the results.
    Actually, the comparison bias was removed in this case by using EBS. Ephemeral storage would be different, and would be more skewed in EC2's favor had I included it.

  11. #26
    Thanks for all the great feedback and useful discussion.

  12. #27
    Regarding VirtIO here is a couple of interesting links relating to a FreeBSD customer that uses us with things like MongoDB clustered database servers:
    How to upgrade to FreeBSD 9.0 with VirtIO: http://viktorpetersson.com/2012/01/1...0-with-virtio/
    Benchmarking VirtIO networking versus alternatives: http://viktorpetersson.com/2012/01/2...der-freebsd-9/
    Tuning up VirtIO networking: http://viktorpetersson.com/2012/01/2...etwork-driver/

    Hopefully he'll post something on storage because I know our customers have seen significant differences in performance with VirtIO (in a positive way). In some cases more than doubling performance.

    In general benchmarking is a poor fit compared with real computing. It tends to break things in the way it behaves giving some deceptive results. We always advice customers to do real world deployments and loads to see how things compare. That's one reason why we offer free trials and also upgrade capacities for customers needing to evaluate something needing more time or capacity. We have a lot of HPC customers and I can tell you we come in on a price/performance favourably.

    To give you an example, check out https://www.cloudsleuth.net/web/gues...-provider-view and select Europe. Bear in mind this is the cloud with 'lower' performance compared with Las Vegas apparently. If you are running a web service from our Zurich cloud serving Europe, we are significantly faster than anyone else. Having only opened the Las Vegas cloud recently, we are in the same process of optimisation over there too and expect to deliver faster and faster performance for external access as that process progresses.

    Regarding the query about CPU, yes, you do get to pick up the slack on spare CPU in our cloud because of the way KVM works. That is of course a bonus as opposed to being robbed if that extra CPU isn't there :-) We had some people gaming the way our cloud works with regards to CPU recently in Zurich. We've now made the necessary changes to prevent such gaming of our cloud in the future so you'll see CPU performance in Zurich improve as a result over the next few weeks.

    Best wishes,

    Robert

  13. #28
    Just want to add my experience on cloudsigma...

    I created a trial account which supposed to last 7 days on their Las Vegas pool. However I cannot access it anymore on day 2 and not even getting a mail from them.


  14. #29
    Join Date
    Feb 2012
    Posts
    34
    I am trying their 7-day trial in zurich, i'm very impressed by their control panel, much power here, loads of drive images/CD/DVD's to install att a click. Many drive-mounts to choose from and much can be customized unlike many other cloud VPS panels i've tried. And implementing the fresh HTML5 based VNC Client really shows that these guys want to lead the pack...

  15. #30
    @jc10 If you registered without using real details like your actual name, actual address etc. then your account is liable for deletion. Likewise use of the trial for things like gaming is also prohibited.

    If you want another trial please send me your details at robert@cloudsigma.com and we will gladly organise this for you.

    Kind regards,

    Robert
    CTO
    CloudSigma

Page 2 of 3 FirstFirst 123 LastLast

Similar Threads

  1. Replies: 19
    Last Post: 11-03-2011, 04:30 AM
  2. Need your opinions! Which looks better?
    By Frosty in forum Web Design and Content
    Replies: 31
    Last Post: 01-17-2009, 08:53 PM
  3. Rackmounted - Any Opinions? Also need opinions on packages.
    By Bojangles6 in forum Dedicated Server
    Replies: 7
    Last Post: 11-29-2006, 06:28 AM
  4. looking for opinions
    By icehole in forum Web Site Reviews
    Replies: 9
    Last Post: 05-15-2002, 04:07 PM

Related Posts from theWHIR.com

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •