Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


DO High CPU Droplets
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

DO High CPU Droplets

peixotormspeixotorms Member
edited July 2017 in Reviews

Hi there,

The other day I read about this:
https://blog.digitalocean.com/introducing-high-cpu-droplets/
and then I thought... finally!

So I launched a droplet and did a quick test to compare it with another similar dedicated instance I have from Vultr.

@jarland I heard you are at DO, so I would suggest, for the High CPU to actually be high performance. Something like an E3-1275v6 for instance, would be faster than Vultr and DO could take the lead on this.

I had high hopes, but it's a bit disappointing.
It's true the cpu's are different, as well as pricing... but I am comparing the same offer type (dedicated cores).
It might not seem like a lot of difference, but PHP really benefits from fast CPU's.
I wish that the High CPU offers were actually the fastest available, not just something that is slightly better.

Check this out:

Vultr Dedicated Instance:
Virtual 714389bda930 @ 3.50GHz (2 Cores)

NGINX Benchmark 1.0.11: Average: 34472.29 Requests Per Second
PHPBench 0.8.1: Average: 489675 Score

# MySQL: SELECT BENCHMARK(1000000,ENCODE('hello',RAND()));
1 row in set (7.2 sec)

Digital Ocean High CPU:
2 x Intel Xeon E5-2697A v4 @ 2.60GHz (2 Cores)

NGINX Benchmark 1.0.11: Average: 27072.47 Requests Per Second
PHPBench 0.8.1: Average: 321515 Score

# MySQL: SELECT BENCHMARK(1000000,ENCODE('hello',RAND()));
1 row in set (8.6 sec)

And a Linode 8GB, $40/month (a normal instance)
4 x Intel Xeon E5-2680 v3 @ 2.50GHz (4 Cores)

NGINX Benchmark 1.0.11: Average: 15512.15 Requests Per Second
PHPBench 0.8.1: Average: 286376 Score

# MySQL: SELECT BENCHMARK(1000000,ENCODE('hello',RAND()));
1 row in set (10.2 sec)

Linode, Vultr and DO score roughly the same in the normal instances, there's not much benefit among offers, except for the extra services. In that Case, DO is better... but offers less RAM.

DO also took 15+ hours to reply to some of my tickets... so there's something to improve there also. Vultr took 15 minutes, and Linode 1 hour.

Thanked by 1jar
«1

Comments

  • edited July 2017

    The 2 or 3 times I had a linode outage (hw failure ) they have answered in like less than 30 mins. Also they helped very quick when their xen to kvm simply fucked up

  • jhjh Member
    edited July 2017

    The high CPU plans are a bit disappointing. It's to be expected though - CPU is expensive. DO's product line has improved a lot recently, though, with different instance types, cloud firewall, volumes etc.

    Linode's support is better and the price/performance is better (in my experience). Linode also have a new control panel - the old one was a bit disorganised. Linode are also slowly rolling out volume storage.

    For European customers, DO have cleverly made volumes available in FRA and high CPU instances in AMS so you have to choose one ;)

  • rm_rm_ IPv6 Advocate, Veteran
    edited July 2017

    peixotorms said: Vultr Dedicated Instance:

    peixotorms said: Digital Ocean High CPU:

    The Vultr dedicated instance also costs $60, compared to DO's $40. Funny how you omitted prices for the first two to not show this, perhaps else your comparison would look way too stupid?

    Thanked by 2philipc Pwner
  • FranciscoFrancisco Top Host, Host Rep, Veteran

    The only answer is to get a slice.

    I may be biased though.

    Francisco

  • jarjar Patron Provider, Top Host, Veteran
    edited July 2017

    Thank you for providing that review and comparison. I'll definitely make sure it's seen by the right people.

    On the support wait time, that's my bad. It's my job to get that down. I'm not alone, but I still place the weight of it on myself. We are getting it down but then it creeps back up again. It's a strategy issue. Scaling out a support team is a difficult but rewarding task. I don't want us to do the things everyone else does to scale, support consistently becomes either terrible or unsustainable at so many companies when scaled. Some pull it off, I want us to be among them. In the meantime, I apologize for the wait time, and know that I do consider it my problem to solve.

    Thanked by 2inthecloudblog bap
  • @peixotorms You should see what I did. I got the highest instance ($640/min High CPU) and it was disappointing. It took more time to do a copy than VMHaus' NVMe on a plain VPS, and most importantly: it couldn't mine.

    grrr https://serverscope.io/trials/MNMD

  • jarjar Patron Provider, Top Host, Veteran
    edited July 2017

    @doghouch said:
    @peixotorms You should see what I did. I got the highest instance ($640/min High CPU) and it was disappointing. It took more time to do a copy than VMHaus' NVMe on a plain VPS, and most importantly: it couldn't mine.

    grrr https://serverscope.io/trials/MNMD

    It's important to note that this isn't meant to be or marketed to be the highest performing server around. It's meant to be the most consistent performing product offered, where neighbors cannot dynamically impact your available CPU resource. It's for people who need consistently available CPU, not necessarily for people who need the absolute minimum MS spent on performing actions in the moment.

  • @jarland said:

    @doghouch said:
    @peixotorms You should see what I did. I got the highest instance ($640/min High CPU) and it was disappointing. It took more time to do a copy than VMHaus' NVMe on a plain VPS, and most importantly: it couldn't mine.

    grrr https://serverscope.io/trials/MNMD

    It's important to note that this isn't meant to be or marketed to be the highest performing server around. It's meant to be the most consistent performing product offered, where neighbors cannot dynamically impact your available CPU resource. It's for people who need consistently available CPU, not necessarily for people who need the absolute minimum MS spent on performing actions in the moment.

    That's a lot to pay for dedicated cores. :(

  • eva2000eva2000 Veteran

    @peixotorms what nginx benchmark and phpbench scripts/parameters are you running ?

  • jarjar Patron Provider, Top Host, Veteran
    edited July 2017

    @doghouch said:

    @jarland said:

    @doghouch said:
    @peixotorms You should see what I did. I got the highest instance ($640/min High CPU) and it was disappointing. It took more time to do a copy than VMHaus' NVMe on a plain VPS, and most importantly: it couldn't mine.

    grrr https://serverscope.io/trials/MNMD

    It's important to note that this isn't meant to be or marketed to be the highest performing server around. It's meant to be the most consistent performing product offered, where neighbors cannot dynamically impact your available CPU resource. It's for people who need consistently available CPU, not necessarily for people who need the absolute minimum MS spent on performing actions in the moment.

    That's a lot to pay for dedicated cores. :(

    Is it a lot per hour? If you only need it for an hour and you pay a monthly fee then you've overpaid by going with a monthly fee only provider. Degrees of need and use cases vary, it's important to remember that. If you just want a monthly rate dedicated core then yeah, there are cheaper options available. I don't know what's cheaper per hour but that's up to each person to shop and pick what they think is best.

    There will always be cheaper options available. I have a 512mb openvz with 4 IPv4 for $5/year. Cheaper always exists. Being the cheapest thing available is one goal, but not one that everyone shares. This has not been marketed in any way as the lowest price per month for the server resource. This fits a need that several of our clients had, and it's marketed as available for anyone with similar needs. Totally acceptable and understandable if that isn't you.

    If a slice is what you need, look no further than @Francisco. Everyone has different needs, and I wouldn't frame myself as having something to offer as an alternative to it. It may seem that way on the surface, and shares many qualities, but it's really not pushed for the same use case. Compare us to AWS, but not to BuyVM. Only Fran can meet your needs, when Fran is what you need ;)

  • FredQcFredQc Member
    edited July 2017
  • HxxxHxxx Member

    I think that the main benefit is that you can abuse the CPU a lot more than with a normal instance or a vultr instance. @jarland might correct me here.

  • jarjar Patron Provider, Top Host, Veteran
    edited July 2017

    @Hxxx said:
    I think that the main benefit is that you can abuse the CPU a lot more than with a normal instance or a vultr instance. @jarland might correct me here.

    Basically yeah. I've met several clients who perform a daily or weekly task and they need to ensure that it's completion time is predictable. With shared CPU cores their perfectly valid complaint is that it's anything but that. You can't predict neighbor usage accurately. With these they are able to offload the task for the hours needed and then spin the droplet down. Cost is barely noticeable for them in this use case, and the predictability of the task is of great benefit to them. To me, this is the picture perfect use case.

    Granted you can use it for anything, but just to speed up a WordPress site, for example, seems to me like an inefficient use of money and resource for the job.

    Thanked by 1Hxxx
  • niknik Member, Host Rep

    The big difference is that DO isn't able to use E3s because they offer up to 32 Cores. Vultr and Linode are not even close to that.

  • WSSWSS Member

    @jarland said:
    Granted you can use it for anything, but just to speed up a WordPress site, for example, seems to me like an inefficient use of money and resource for the job.

    It's WordPress based- they're already used to getting screwed for an inferior product.

  • jarjar Patron Provider, Top Host, Veteran

    @WSS said:

    @jarland said:
    Granted you can use it for anything, but just to speed up a WordPress site, for example, seems to me like an inefficient use of money and resource for the job.

    It's WordPress based- they're already used to getting screwed for an inferior product.

    If you say so :P

    Base WP is pretty nice.

  • WSSWSS Member

    @jarland said:
    Base WP is pretty nice.

    At least for the 10 seconds until the latest exploit hits it. Heyy - don't you like CPanel too? ;)

    Thanked by 1jar
  • williewillie Member

    I don't think this is sold as a pure cpu resource (it's ridiculous compared to cheap dedis) but rather, as a big server instance inside DO (e.g. for a shared database). It's not for low end customers at all, but for business users whose main expenses are salaries, office rent, COGS etc. Server costs are down in the noise even if they're 10x what an LET user would spend on comparable iron. So they want something that plays well with their other stuff that's all under the same umbrella. In my sector the default host for that is AWS, which is even more expensive than DO, but people keep buying it.

    Thanked by 1jar
  • HxxxHxxx Member

    DO was never for the low end scum. Only LET thinks it is but you are paying 10 dollars per 1GB RAM, that's premium price and prety much the standard nowadays.

  • jarjar Patron Provider, Top Host, Veteran

    @willie said:
    In my sector the default host for that is AWS, which is even more expensive than DO, but people keep buying it.

    Yep. Which is why I think they're a more fun competitor than the low end. I mean, who needs to compete on the low end anyway when we've got Fran? Market cornered ;)

  • HxxxHxxx Member
    edited July 2017

    Agreed @jarland

    If only @francisco had more inventory...

  • filefile Member

    Mmm slices... of pie. Want pie now.

  • FranciscoFrancisco Top Host, Host Rep, Veteran

    @Hxxx said:
    Agreed @jarland

    If only @francisco had more inventory...

    Stock is going to be a real issue if I do the storage product I've been yammering on about.

    But, I've made this thread about me more than I should have, keep it on DO to help @jarland pay for his new tech.

    Francisco

    Thanked by 1Foul
  • J1021J1021 Member

    @jarland does DO still copy snapshots and backups
    to glacier?

  • jarjar Patron Provider, Top Host, Veteran

    @kcaj said:
    @jarland does DO still copy snapshots and backups
    to glacier?

    Snapshots yes but backups no. Since they get replaced weekly and typically end up on different NAS boxes. I expect that conversation to be revisited in the future, as far as backups going to glacier or not.

    Thanked by 1vimalware
  • williewillie Member
    edited July 2017

    .

  • @jarland do you know when DO will allow us to download our own snapshots. That is one of the features that keeps me from coming back. Like, I don't care about the price anymore it can still be $10 per 1gb of RAM. I just really want to be able to download my snapshots.

  • @jarland thanks for the reply. Glad it can be seen by the right people. If these are meant to be simply an offer for dedicated cores (with slightly better performance), then I suggest you look into having HPC instances geared towards the fastest performance. I definetly need faster (est) CPU's all the time, and I don't mind paying for it. And I'm sure there is a market for it too. As long as there are clients willing to pay hundreds of dollars for the EC2 crap, there is a demand for fast cores.

    @rm_ not sure where your love comes from... but I did say "pricing is different" on my post. The offer is similar (dedicated cores). Sure Vultr is more expensive, but so is Google Cloud and it doesn't necessarily means that more expensive is faster. The benchmarks are for single cpu core performance and both offers are dedicated cores. Price is irrelevant as well as disk, network or memory.

    @Francisco I'm still a bit afraid of having production sites on Linode, since I was also affected by this at that time: https://blog.linode.com/2016/01/29/christmas-ddos-retrospective/
    Nevertheless, I do use them for development extensively.

    @doghouch This is transversal. If you try multicores on Google Cloud, the performance of 1 vs 10 cpu cores is not linear. It's faster to have 10 instances with 1 cpu core, than 1 machine with 10 cores.

    @eva2000 I installed Ubuntu 16.04, latest Nginx mainline and PHP 7.0 for these (PHP 7.1 not supported). Benchmarks are from phoronix-test-suite.com but the mysql is just a query I run after installing percona 5.7, because this tells me immediately if the cpu is fast or not.

    @nik That is also true and it's great, however more cores are not necessarily better. For example, 2 modern fast cores can do the job of 16 "old" cpu cores faster (and I've tested this extensively). I would rather have 8 super fast cores than 16 slow cores... and for scalability, there are other solutions than to just scale up.

    Thanks for all the comments.

    Thanked by 2jar eva2000
  • eva2000eva2000 Veteran

    peixotorms said: but the mysql is just a query I run after installing percona 5.7, because this tells me immediately if the cpu is fast or not.

    thanks gives me more ideas for adding to my benchmark scripts :)

    Thanked by 1inthecloudblog
  • rm_rm_ IPv6 Advocate, Veteran

    peixotorms said: I did say "pricing is different" on my post.

    Yet you've hidden the actual difference where it was the most important, but "for some reason" had no problem showing that the 3rd contender is $40/month right there in the post.

    peixotorms said: Price is irrelevant

    Then you can just buy one of these https://www.online.net/fr/serveur-dedie/dedibox-wopr and call it a day.

Sign In or Register to comment.