DO Ideas 2

More powerfull cpu

In my case is more significant the power of a single CPU core than the total number of CPU. (In my current dedicated server I have disabled the hipertheading)

I'm testing you're server in order to transfer a performance critical application I have in a dedicated server. My current bottle neck is on disk io so I thought that your ssd based service will solve my performance issues but:

Although the SSD increases the respond time of the first run of a query, all subsequent queries are CPU bounded and are significant slower than my current non-SSD dedicated server.

  • tk
  • Sep 11 2018
  • Shipped
  • Sep 11, 2018

    Admin Response

    Hi, We just have launched High-CPU Droplets which are powered by some of Intel's most powerful processors available and optimized for compute intensive workloads. You can learn more about these Droplets here: https://blog.digitalocean.com/introducing-high-cpu-droplets/ Feel free to email me directly if you have any questions.
  • Attach files
  • Mike Hayes commented
    September 11, 2018 18:51

    I agree with Anonymous that I did not see much of a performance increase with the $40 high-cpu plan. At best it seems comparable to 2 $5 droplets.

  • Jake Cattrall commented
    September 11, 2018 18:51

    We recently had to ditch DO for high CPU due to the 20gb storage limit. The volume storage overhead would have been impossible to use as we had no control over where our server stores it's files.

  • Anonymous commented
    September 11, 2018 18:51

    these HIGH-CPUs are more or less marketing BS. My team spent a lot of time testing them and the best we got was 5% increase in the performance compared to same # of CPUs in normal DO droplets.

    here are the results of AB between normal and High-CPU node:

    ------------ 8Gb ------------
    Concurrency Level: 20
    Time taken for tests: 113.161 seconds
    Complete requests: 20000
    Failed requests: 0
    Total transferred: 183089550 bytes
    HTML transferred: 179849550 bytes
    Requests per second: 176.74 /sec (mean)
    Time per request: 113.161 [ms] (mean)
    Time per request: 5.658 [ms] (mean, across all concurrent requests)
    Transfer rate: 1580.04 [Kbytes/sec] received
    Connection Times (ms)
    min mean[+/-sd] median max
    Connect: 3 3 1.0 3 28
    Processing: 31 110 26.9 113 273
    Waiting: 30 106 26.5 109 260
    Total: 35 113 26.9 116 276
    Percentage of the requests served within a certain time (ms)
    50% 116
    66% 126
    75% 132
    80% 135
    90% 144
    95% 153
    98% 164
    99% 173
    100% 276 (longest request)
    ------------ c-4 ------------
    Concurrency Level: 20
    Time taken for tests: 108.060 seconds
    Complete requests: 20000
    Failed requests: 0
    Total transferred: 183089432 bytes
    HTML transferred: 179849432 bytes
    Requests per second: 185.08 /sec (mean)
    Time per request: 108.060 [ms] (mean)
    Time per request: 5.403 [ms] (mean, across all concurrent requests)
    Transfer rate: 1654.61 [Kbytes/sec] received
    Connection Times (ms)
    min mean[+/-sd] median max
    Connect: 3 20 10.7 19 85
    Processing: 19 88 30.0 84 513
    Waiting: 18 82 29.4 79 512
    Total: 38 108 29.9 104 555
    Percentage of the requests served within a certain time (ms)
    50% 104
    66% 116
    75% 125
    80% 131
    90% 147
    95% 164
    98% 182
    99% 194
    100% 555 (longest request)

  • George_centminmodcom commented
    September 11, 2018 18:51

    definitely needed as DO is now lagging behind in the 3yrs this suggestion was posted benchmark proof at https://community.centminmod.com/forums/virtual-private-server-vps-hosting.26/?prefix_id=45

    time to see DO update to the new Xeon Broadwell-EP E5 26xx v4 processors http://www.anandtech.com/show/10158/the-intel-xeon-e5-v4-review :)

  • Ted Wood commented
    September 11, 2018 18:51

    Yes, please include 2 CPUs in the $10 plan.

  • Ted Wood commented
    September 11, 2018 18:51

    Some people here have brought up the cost of droplets. $5 for a starting price. Cheap right? Well, add in the 20% for backups, and then convert to Canadian (my local currency), and it's no longer a $5 droplet. It's closer to $10 for me. Multiple that by 20 small droplets with limited resources and I'm paying $200 a month. That's far more than I'd pay for one large VPS with generous resources hosting many websites. I no longer want the shared-VPS model for my company. So my point is that the low DigitalOcean pricing actually comes at a significant cost that's not nearly as affordable as it first seems.

  • Ted Wood commented
    September 11, 2018 18:51

    I've tested out DigitalOcean for a year before bringing over the rest of my sites. I finished transferring the last of the sites (20 of them) in August, but I've been experiencing several server outages lately. This is likely due to PHP+Nginx not having enough memory to handle the load from some of these sites. Digital Ocean held well during my evaluation and then has not been providing the value that was initially seen. Now I'm faced with doubling my hosting costs just to avoid outages. Not exactly the experience that I was expecting.

    I've been evaluating competing services, mainly Vultr and RamNode, for some of the higher-traffic sites that are experiencing problems. RamNode in particular offers more CPU for less cost than DigitalOcean. I hope that DigitalOcean can keep their droplets competitively priced to keep my sites here.

  • Dylan Wilson commented
    September 11, 2018 18:51

    While I like the idea, what I like even more is DO's simplicity, no need to configure SSD, Cores, RAM etc... which in turn give simple pricing.

    Being able to configure your droplets would be nice but honestly, I would prefer configured options.

  • Oh Jia Jun commented
    September 11, 2018 18:51

    Another way to say this is a fully customisable VPS :)
    Eg. 2 Core and 256MB ram has the same pricing as 1 Core+512ram.
    I support this!

  • Anonymous commented
    September 11, 2018 18:51

    It would be nice to see a 10$ plan with 2vCPU, i need a cloud application that needs to be able to run as many threads as possible, not a cloud storage service

  • Ted Wood commented
    September 11, 2018 18:51

    One way to work around the low number of cores is to set up multiple droplets with Private Networking... one for application, one for database, one for memory-based caching, etc. That way, $15 gets you 3 cores to serve your site, for example. By moving away from the "one server does it all" mindset, you can actually get tremendous value and performance from DO's 1-core plans.

  • Ted Wood commented
    September 11, 2018 18:51

    I was excited to be moving my entire business to Digital Ocean away from a Rackspace reseller. I thought the price-per-droplet would be better than what I am currently using. I failed to look closely at the CPU count of each server. Despite sites being faster under light load when running on DO droplets, things begin to change drastically once there's a heavier load compared to my previous hosting. I peg this at the CPU difference. My previous production server (20GB HHD, 500MB RAM, and 4 cores), vs only 1 core at Digital Ocean. I would actually need to spend more per month at Digital Ocean to get the same 4 cores. I take full responsibility for overlooking the CPU performance, as I thought that DigitalOcean was a dream come true. Some dreams are too good to be true, I guess. :) I'll continue my migration to DO in hopes that they strike a better balance of price-per-core. I don't need to massive RAM and SSD that comes with the high-core servers.

  • Mikhail Kolesnik commented
    September 11, 2018 18:51

    Is there any proof that DO's single core is much slower than Linode's 8 on some popular web stack? UnixBench actually gave me the impression that DO is overall faster despite of the number of cores. At the end it does not matter much if you have one or 32 cores if that single CPU is really fast.

  • Anonymous commented
    September 11, 2018 18:51

    It's a good idea to increase CPU ...I didn't use Linode, but, even so, I would want DO to do it.

  • Haydrion commented
    September 11, 2018 18:51

    It is also annoying that people are like to do the Linode vs Digital Ocean topics, digi is new, Linode is here for years. They are leading and that will always be how much Digital Ocean is trying to do, they will never get the level of Linode

  • Haydrion commented
    September 11, 2018 18:51

    Linode is just better than Digital Ocean ! End of Story ! Why you pay more by Linode is because on Linode you have more freedom than by Digital Ocean. $5 for a VPS .. serious .. don't expect amazing things or even better things that you have by Linode

  • Evandro Camargo commented
    September 11, 2018 18:51

    Well, maybe they could keep the current plans (for those of us who actually fit into it.) and make the "custom" plans at fair prices (not only fair to customers but to their infrastructure too.).

    That way we have what we want and they get what they need to not overwhelm the structure and still provide everyone with a reliable level of service.

  • Bob Monteverde commented
    September 11, 2018 18:51

    I would assume this would be tricky to implement. It would be much harder to divide of server resources. Would like to hear from someone at DO about this, I'm sure they have put quite some thought into it. From a user perspective it would be ideal to choose RAM, HD Storage, Cores, and Bandwidth independently.... but again, probably a very hard thing to do from their perspective.

  • Bob Monteverde commented
    September 11, 2018 18:51

    While this would be amazing, it may not be practical at the current cost. On the other hand, I find it very useful to have at least 2 [v]CPUs per a droplet. Having to pay for a $20 a month droplet, when I don't need the extra RAM is a little steep.

    I would be more than happy if you just gave the 1 GB plan 2 cores. It makes sense to keep the 512mb plan 1 core.

    I suppose there is still an issue with the larger plans having few cores, but my vote still is to just boost the 1gb plan to 2 cores.

  • Teshoo Lama commented
    September 11, 2018 18:51

    The way CPU cores scale through your price plans penalises your bigger customers. At things stand, Bitcoin miners pay $5 per core, while on the 8GB plan we're paying $20 per core. I understand you're primarily competing on RAM, but charging 4x more per core is just extreme.

  • Rodrigo Sandoval commented
    September 11, 2018 18:51

    I really want to move to DO, but at this point is useless with the current configurations, I understand the whole thing about pricing and everything else, but being able to create your own droplet with custom resources sounds better than being forced to have 160Gb of HDD when you don't need it, or 16Gb of RAM, when I only need 8 cores and 2Gb of RAM

  • Anonymous commented
    September 11, 2018 18:51

    I think increasing ram would be better from 512 to 728 or something.

  • Vincent commented
    September 11, 2018 18:51

    From what I see on Linode, you are on a server with cpu with 8 physical core but you have access to 1 virtual core like Digital Ocean.

  • Gaurav commented
    September 11, 2018 18:51

    Agreed

  • Jeffrey commented
    September 11, 2018 18:51

    Linode is more expensive, so don't expect DigitalOcean to double the amount of cores without charging you more. It doesn't just work like that.

  • Simon West commented
    September 11, 2018 18:51

    Linode also seems to be considerably more expensive than Digital Ocean.

  • Sjors commented
    September 11, 2018 18:51

    This is why I want to host another application at Linode. The SSD/HDD comparison doesn't matter since I'll be caching files heavily anyway.

  • Moisey Uretsky commented
    September 11, 2018 18:51

    @sharath win:

    Thanks for the feedback, however the situation is a bit more complex. When it comes to a cloud or a VPS provider there is no "apples" to "apples" comparison of CPU possible. This is because unlike RAM which is strictly segregated, CPU and disk IO are not strictly segregated but shared.

    This means that it is impossible to really accurately predict how that utilization will playout in a production environment with mixed workloads from different customers on the same hypervisor.

    Our plans for RAM do scale linearly because this is the resource that we have absolute control over, we also scale our HDD space linearly for the same reason, except the smallest plan starts with a bit more HDD space as a bonus because if we scaled linearly there it would come with 10GB to start and we felt that was a bit on the small side so we bumped it up.

    When it comes to CPU it is important to not forget that in cloud environments or VPS it is not always the CPU that gets taxed, but instead the bottleneck becomes the disks. This is because if you review your CPU usage and your applications you will see that often times your CPU usage is also tied to reading/writing from the disk. As a result when disk IO runs out your CPU usage spikes. This is because an application is "waiting" on the disk to either finish a write or a read before it can process the next instruction, as a result your CPU usage begins to increase even though there is actually more CPU available.

    This is why we went with an all SSD cloud because we know that more often than not most customers workloads will tie up the disk before they tie up the CPU. So while we may provide fewer cores by having faster disks underlying the infrastructure you get more CPU. This is because even disk IO isn't completely saturated all of your read and write requests are finishing faster as well so the CPU can move on to the next instruction.

    That is why the best way to really compare two hosts is to run a production environment on each and compare the price to performance. Running benchmarks is helpful but unfortunately with benchmarks they aren't really mimicking a real production workload because they usually just like to hammer things in a very predictable manner, where as a production setup has much more randomness built into it.

    The way that we have setup our infrastructure is to allow for bursts in CPU to be processed faster again with the idea that if there are also SSD drives that will have a larger workload clear through faster and then cause less overall contention. So while you may get less "cores" you get more of the core for processing.

    The last item to consider is your actual application. Depending on your stack or application it may not be very multi-threaded, so even if you have 8 cores you end up really utilizing only one.

    Ultimately I would recommend setting up a second app, web, db, or whatever server and running it in production on another provider, whether it be us or someone else, and then compare the CPU utilization and ms response times against price.

    Think of it like this, you can have two cars that have 500hp, but why does each one accelerate 0-60mph differently? That's because many factors come into play such as aerodynamic drag, gear ratios, powerloss through the drivetrain, grip, rolling resistance, etc.

    It's the same here, because there is no specific CPU unit, unlike RAM where 1GB = 1GB, this means that a direct comparison isn't possible.

  • Luka commented
    September 11, 2018 18:51

    @Sharath Win: Agree to the letter...

  • Sharath Win commented
    September 11, 2018 18:51

    I just did a basic comparison of the resources offered for every price point in DigitalOcean vs Linode.

    Linode resources & prices are more predictable for every price point. Everything (RAM, 8 CPUs Priority, Disk, Bandwidth) Doubles for every price point and very easy to understand and plan on the capacity for resources.

    Digital Ocean prices are definitely low cost when RAM alone is considered for the comparison. However the Disk, CPUs and the Bandwidth offered is no way in comparison to what Linode offers and its difficult to understand how that resources are set for each plan.

    Please add CPUs also for the comparison to help understand the price | resource comparison
    https://www.digitalocean.com/price-comparison-chart

    It would be great if DO can offer better resources or alternate way of getting the resources such as CPUs, SSDs and Bandwidth.

    Just do a simple analysis for $80 plan, the resources offered are:
    8GB Memory
    4 Cores
    80GB SSD Disk
    5TB Transfer

    For same $80 DigitalOcean gives out more resources on Entry level plans ... Not sure why DO doesn't realize this and why it penalizes the users who shows interest in bigger plans!

    If we take $5 droplets (512MB Memory, 1 Core, 20GB SSD Disk, 1TB Transfer) - for $80 16 droplets can be purchased, with the total resources across all droplets:
    8GB Memory
    16 Cores
    320GB SSD Disk
    16TB Transfer
    possibly on different physical hardwares - even super!

    When the above is possible, why can't DigitalOcean offer the similar the resources Doubling for every price point? It will be the same cost for DO and its more predictable for the customers!

    If you offer low cost plans with more resources, why penalize the users when they opt for bigger plans? I think it should be discounted when we go for bigger plans than penalizing on hardware resources.

    Are you encouraging all your users to be on $5 plans?? This will waste lots of computing resources to have Linux run for every single instance - too much of unwanted CPU, RAM, Disk wastage for each Droplet. "LETS GO GREEN" and save the computing resources by encouraging the users to go for bigger plans with better prices|resources.

    Hope you guys do some thinking and apply some logic to your pricing and plans and resources - that will make people go for bigger plans than fiddle with all these smaller plans!

    Thanks

  • Sharath Win commented
    September 11, 2018 18:51

    Linode provides 8 CPUs with low priority share for every plan.

    Would be great if you can offer something in comparison ... its difficult to rely on 1 Core server specifically when we run Rails apps.

    How can we get more CPUs per a given plan?

  • Anonymous commented
    September 11, 2018 18:51

    Rackspace also does this right (and SliceHost before being acquired by Rackspace). They do a good job at splitting up CPU and also bandwidth. You have to remember the bigger your server, more you pay, the bigger the pipe you should have.

    One of the BIGGEST problems with hosting like this is resource sharing. If you have a client who has an application that becomes problematic...Or they just are downright abusive...Others will suffer and that's really not good. I understand that sometimes this means a user can get more CPU than what they pay for if others aren't using it and that's cool and all...But we have to put a priority on consistency first.

  • Josh commented
    September 11, 2018 18:51

    I have to agree with Roger here. That is the one thing that makes Linode more appealing at the moment. Overall everyone gets the maximum potential of the CPU if other nodes are not using it, but those that pay higher get a higher priority in that process. So performance is comparable based on the idea of potential rather than predefined constraints.

  • roger pack commented
    September 11, 2018 18:51

    Just to make sure I'm being clear, I believe the original poster here is actually requesting an option to spend more money for "faster cores."
    My suggestion is a bit different here, it's actually to "do what linode does" here and, given a box with 8 cores and 8 single core accounts on it, guarantee each account "at least one core at all times" but, if one account's 7 neighbors aren't utilizing their cores, to allow the extra, underutilized cores to be used by each account. Effectively multiplying cpu power for each account, without adding extra cost.

  • Anonymous commented
    September 11, 2018 18:51

    Agreed, our app runs OCR, which is totally CPU dependent, so having a fast CPU is imperative before making a switch.

  • Shelby DeNike commented
    September 11, 2018 18:51

    Would love the ability to add more cores to any plan as needed as I require more CPU than I do disk or bandwidth.

  • Ivan Lagru commented
    September 11, 2018 18:51

    This sounds great, the upgrades should apply to the single core and dual core instances I guess, right?

  • Es Cendol commented
    September 11, 2018 18:51

    It says Planned, any one have getting upgrade on CPU?

  • roger pack commented
    September 11, 2018 18:51

    definitely suggest that by default, images have some type of swap enabled...

  • Banh Mi Cua Em commented
    September 11, 2018 18:51

    WHat you pay is what you get, buddy.. This is more like VPS not a real CPU from dedicated. But it cheap.

  • LazyTiger commented
    September 11, 2018 18:51

    You could create a swapfile on your ssd. There are tutorials on the web describing how.

    @xavier: I think it's better to let the user add everything optional (like a swap-file) that takes additional space on the disc.

    I also agree with J. Rollins: You can't expect fast desktop power etc. from a $5 plan with 0,5 GiB RAM.

  • Anonymous commented
    September 11, 2018 18:51

    Is the update active yet? Looking forward for it :)

  • Matt Razza commented
    September 11, 2018 18:51

    What really kills our performance on the smaller EC2 instances (and I think it's the issue with DO as well) is steal time. We can have prolonged CPU bursts and the hypervisor starts taking cycles away from our VM. The EC2 medium instances have a higher CPU priority so we get basically zero steal time.

  • Chirag Patel commented
    September 11, 2018 18:51

    Hey guys. I'm at LiNode at the moment and only reason holding me back is the CPU juice. Can you please do this sooner. We can't wait to be on fully SSD environment. :)

    Keep the good work, its great for us (the customers)

  • Johnneylee Rollins commented
    September 11, 2018 18:51

    You can always just not use a 512mb VPS server as a desktop machine. Ever consider using it as a VPN or something?

  • Scott R. commented
    September 11, 2018 18:51

    @kenn

    If you use snapshots to build identical loadbalanced servers, dynamically created or destroyed, DO has a major advantage

  • Kenn Ejima commented
    September 11, 2018 18:51

    CPU Price Performance: DigitalOcean vs Linode

    https://gist.github.com/kenn/5228906

    tl;dr

    If you have multiple app servers (like Rails), Linode is a better option. Like, Rails server x 4 (x8 cores each) + 1 DB server.

    DigitalOcean is better and well balanced when you only have one server and is more oriented toward scaling up rather than scaling out.

  • Moisey Uretsky commented
    September 11, 2018 18:51

    Not sure of the sizing of the medium instances without specific metrics, but if you are running a single threaded application as you mentioned getting more cores won't increase performance as the remaining cores will be sitting idle.

  • Matt Razza commented
    September 11, 2018 18:51

    We're looking to spin up and down instances of a real-time single threaded application to scale with compute demands. We've found Softlayer and EC2's medium instances provide a solid experience but we haven't been able to get the same performance out of digital ocean - this is completely CPU bound and getting additional cores provides no improvement as the application is largely single-threaded.

  • Scott R. commented
    September 11, 2018 18:51

    Technically, for the price of linode with 1GB of ram, DO gets you 2 cores at full utilization, with 2GB of ram

    DO has more ram, less CPU

    Depending on your tasks, particular with caching, you can do tradeoffs

    Running a website?

    APC, static caching, and varnish can eat a lot of ram, but it reduces CPU usage

  • Moisey Uretsky commented
    September 11, 2018 18:51

    If you're having an issue please comment here so we can get more feedback to understand customer usage =]

    What stack/app are you running, where is the limitation/bottleneck that you are hitting, etc.

    Thanks

  • Matt Razza commented
    September 11, 2018 18:51

    Thanks for the response. Hopefully we can get more information soon. Would love to use your service but the CPU just isn't cutting it atm.

  • Moisey Uretsky commented
    September 11, 2018 18:51

    If we scaled up the resources on the existing plans and hypervisors it could lead to worse overall performance for customers.

    By the same token to accomodate new plans we would need to provision new hardware, which would mean that our inventory process management is now fulfilling requests for different types of servers based on the usage of different types of plans which makes it much harder to scale operations because instead of keeping track of just "usage" we have to now track usage of multiple product lines.

    The other difference is that cores do not translate to "cpu speed", they do translate to the ability to parallel process better but smaller plans do come with less resources so having more cores wouldn't necessary provide a larger boost for a particular application.

    So it's a bit more complex than it first seems.

    Thanks

  • Moisey Uretsky commented
    September 11, 2018 18:51

    We're looking into this request and seeing what we can do about it, no clear ETA on it yet, if we go down this route we will be providing updates regarding it and we are also investigating a new engineering tab that will keep all of our release information easily browse-able as well as current work so that customers are better informed of active and planned development.

    Thanks

  • Andreas Madantzis commented
    September 11, 2018 18:51

    Thanks for indicating that, i m having a very similar situation

  • Morthawt commented
    September 11, 2018 18:51

    Moisey Uretsky, will you be notifying customers of when they should reboot their servers to get better CPU speed?

  • Matt Razza commented
    September 11, 2018 18:51

    Also looking for more info on this.

  • Eugene ElJefe Cook commented
    September 11, 2018 18:51

    Any update on this? Was this done?

  • TribalInstincts commented
    September 11, 2018 18:51

    Can I just go ahead and add all 10 votes here? This is huge for me right now.

  • mohd sadiq commented
    September 11, 2018 18:51

    Please, when making an cpu updates , write a blog post or email existing users.

  • Ralph Tang commented
    September 11, 2018 18:51

    Hi,

    Can I confirm that Digital Ocean will be making CPU speed upgrades in the next week or so? I'm really looking forward to it!

  • Di commented
    September 11, 2018 18:51

    Linode just recently went rambo on the CPU core giving every plan the ability to burst to 8 cores. Not saying that it's a good idea, but being able to burst to 2X the current number of cores seems reasonable.

  • Moisey Uretsky commented
    September 11, 2018 18:51

    We will be making some CPU updates on the smaller plans in the next 2-3 weeks =]

    Any existing customers will be able to upgrade with just a shutdown on the command line and a "Boot" from the control panel.

  • rogerdpack commented
    September 11, 2018 18:51

    me too, it seems odd to me that linode lets you "use your neighbors cpu" (up to 8 cores) but with DO you're stuck with always exactly 1...

  • xavier commented
    September 11, 2018 18:51

    well you can alway make swaps but its better for those people who dont know how to make them if it comes premade in the plan. what i really want is a better cpu

  • Morthawt commented
    September 11, 2018 18:51

    Better CPU's are always going to improve everything.

  • xavier commented
    September 11, 2018 18:51

    awesome =]

  • Moisey Uretsky commented
    September 11, 2018 18:51

    Hi Xavier,

    We're going to look into it =]

  • xavier commented
    September 11, 2018 18:51

    im not really asking for much here just raise the cache sizes in the cpus a little more and at least 124mb - 200mb swap since my old provider has unmetered bandwith 2 cpus 512mb plan with a 1gb swap all for just 1 dollar more the only thing about it was the speed was a little lower then the one provided here