Howdy, Stranger!

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

Sign In with OpenID
Advertise on LowEndTalk.com

In this Discussion

VPS IO Benchmark Competition

VPS IO Benchmark Competition

serverbearserverbear Member
edited July 2012 in General

Thought it'd be interesting to run a little competition to see which LEB has the best IO per $ spent. If you score 57 MB/s & your box is $5/month then your IO per $ spent is 11.4.

This is the first time we've run a competition, if you guys like the idea then we're happy to do something recurring themes (i.e. UnixBench/GeekBench/IOPS scores).

IO Measurement

We will be measuring the IO from the following dd command: dd if=/dev/zero of=sb-io-test bs=64k count=16k conv=fdatasync

How to Enter

Simply run the following command on your box to enter, this will complete a full benchmark with ServerBear (not just dd):

wget -N https://raw.github.com/Crowd9/Benchmark/master/sb.sh&&bash sb.sh 'HostName - LEB Competition' 'PlanName' '[email protected]' 'Price'

View the benchmark script (if you need to see what we are executing): https://github.com/Crowd9/Benchmark/blob/master/sb.sh

Replace:

  • HostName with the name of your host (i.e. Hostigation), you must include the LEB Competition so we know you are entering.
  • Plan Name with the name of your plan (i.e. OVZ-128)
  • Your Email address (you can leave this out if you want, but I will need this to send you the report + email you if you win)
  • Price - Just include the monthly cost of the plan (in USD)

Why can't we just post our own DD numbers? - If we just asked you to run the dd command on your box, there's no way to regulate cheating or fudging of numbers.

Results:

I will update this thread with the scores, ranked by IO per $ spent. Once completed I'll aggregate it up to an individual host so you can see which hosts provide the best IO value. The competition runs until Midnight GMT on the 31st August 2012.

Prizes:

  • 2 x winners (one of which must be non-ssd) will get their winning LEB paid upfront for the next 2 years (paid via Paypal)
  • 2 x runners (one of which must be non-ssd) up will get their runner up LEB paid upfront for the next 1 year (paid by Paypal)

Rules:

  • You must benchmark a LEB (base plan monthly cost must equate to less than $7/month)
  • You can only benchmark a single box once, so any tweaks must be made before you run it
  • Multiple entries from the same user are accepted but must be different LEB's & IP addresses
  • All entries will be checked to ensure the system specs match up with the plan submitted
  • The benchmark is high CPU & IO intensive, we cannot be held responsible for any issues it may cause with your host.
ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
Tagged:
«1

Comments

  • TazTaz Disabled

    Sata and ssd would differ a lot. So ssd vps will win for sure. How ever, what do comes in mind is that, while ssd providers will provide small diskspace (5-10gig) a leb provider can provide double specs for the same amount of storage.

    For example ssd provider x does $5 for 5gb space while average provider y does 15gb for $5. So unless, You are comparing apple to apple, this competition is not going to be fare.

    Time is good and also bad. Life is short and that is sad. Dont worry be happy thats my style. No matter what happens i won't lose my smile!

  • miTgiBmiTgiB Member

    @NinjaHawk said: So ssd vps will win for sure.

    Don't be so sure of that ;)

    Hostigation High Resource Hosting - SolusVM OpenVZ/KVM VPS
    Thanked by 2TheHackBox Maounique
  • miTgiBmiTgiB Member

    @serverbear said: which hosts provide the best IO value.

    Why are all these testfiles being downloaded, what part of IO testing does that fulfill?

    Hostigation High Resource Hosting - SolusVM OpenVZ/KVM VPS
  • TazTaz Disabled

    @miTgiB can you beat gig/s

    Time is good and also bad. Life is short and that is sad. Dont worry be happy thats my style. No matter what happens i won't lose my smile!

  • serverbearserverbear Member
    edited July 2012

    @NinjaHawk In the spirit of fairness, we'll do two winners & two runners up. One of each must be non-ssd.

    @miTgiB It's our standard testing suite, we haven't built a script purposely for this competition (as stated above).

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • miTgiBmiTgiB Member

    @NinjaHawk said: can you beat gig/s

    Only 1GB/s+ dd result I've seen was not a LEB

    Hostigation High Resource Hosting - SolusVM OpenVZ/KVM VPS
  • TazTaz Disabled

    @miTgiB check aging ;) someone got it.

    Time is good and also bad. Life is short and that is sad. Dont worry be happy thats my style. No matter what happens i won't lose my smile!

  • dnomdnom Member

    @miTgiB said: Only 1GB/s+ dd result I've seen was not a LEB

    IIRC I've seen quickweb dd results at 1GB/s but I think that's just because it's writing to cache.

  • All these i/o tests are going to kill the innocents :)

    mmm, I will add a check box in the order form: [ ] Check this if you plan to run i/o tests more than 2 time / year

    then will place that users on raid0 overcached nodes :)

    IperWeb & Prometeus, Hosting Provider since 1997. iwStack cloud infrastructure
    Thanked by 1djvdorp
  • TazTaz Disabled

    @prometeus LOL. Imagine what will happen if all users from a specific node decides to run IO test at the exact same time? EVEN SSD Would cry :P

    Time is good and also bad. Life is short and that is sad. Dont worry be happy thats my style. No matter what happens i won't lose my smile!

  • Added a disclaimer to the bottom of the competition. Just in case :P

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • jcalebjcaleb Moderator

    How to send entry? Just post in this thread?

  • @jcaleb you don't have to post them here (I have the email addresses), but you can if you want :)

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
    Thanked by 1jcaleb
  • My money is on ipxcore 0.50 boxes would be in 1-200 io per $

  • jcalebjcaleb Moderator

    @hayhayday said: My money is on ipxcore 0.50 boxes would be in 1-200 io per $

    Does special deal with the provider counted?

  • @jcaleb I will be using the standard price listed on the website for the plan.

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • yomeroyomero Member

    Where you will be doing this crap? I don't want to be in the same nodes... ¬¬

  • Nick_ANick_A Member

    Oh no...

    RamNode: High Performance SSD and SSD-Cached VPS
    Atlanta - Seattle - Netherlands - IPv6 - DDoS Protection - AS3842
  • jcalebjcaleb Moderator

    Are there any other disk io test with no ouch on the node?

  • TazTaz Disabled

    Poor nodes are going to get raped. Stop this inhumanity! Stop Node Abuse. #SaveMyHD

    Time is good and also bad. Life is short and that is sad. Dont worry be happy thats my style. No matter what happens i won't lose my smile!

  • jcalebjcaleb Moderator

    @NinjaHawk maybe run another contest, for those who can prove they did not dd their boxes for the longest time =)

    Thanked by 1Taz
  • jcalebjcaleb Moderator

    I just thought... since @Damian has the cheapest box at 50cents, his boxes would be the most logical entries.... Hope no downtime for IPXCore due to excessive io =)

  • Not seen any IPXCore yet, although it would be cheap for me to pay the winner on that one ;)

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • @serverbear said: Not seen any IPXCore yet, although it would be cheap for me to pay the winner on that one ;)

    I am running it now, hope @Damian wont suspend my box for it :p

    I use http://tuxlite.com to configure all my VPSes and I love it!

  • rolz14rolz14 Member

    Running now on one of my xen vps

  • @serverbear said: Not seen any IPXCore yet, although it would be cheap for me to pay the winner on that one ;)

    Result is here: http://serverbear.com/benchmark/2012/07/31/b9Nhro3ZjzyNWWUZ

    Seems like its not live yet? Cant find it in search with name, or when sorting on price (0.5 usd/month)

    I use http://tuxlite.com to configure all my VPSes and I love it!

  • @djvdorp There's a manual check before summarised data appears in the reports. Few reasons why, but in order for it to be useful I need to add all the plan data for that host.

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • @serverbear said: @djvdorp There's a manual check before summarised data appears in the reports. Few reasons why, but in order for it to be useful I need to add all the plan data for that host.

    I see. All info needed is here: http://www.lowendtalk.com/discussion/3384/ipxcore-32mb-openvz-vps-for-50-cents-per-month/p1

    I use http://tuxlite.com to configure all my VPSes and I love it!

  • @Serverbear doesnt seem to e-mail me once it's finished...

  • DamianDamian Member

    I don't mind people entering the contest with our services, however, the disk arrays are tuned for random I/O, not massive block writes, hence why the interactivity of the service 'feels' fast despite not returning 'impressive' numbers. I don't think we'll win.

    I have yet to determine how writing a contiguous 1gb file is indicative of anything anyone is going to do with their VPS, since I can't think of any 'normal' processes that would write a contiguous 1gb file. It is, however, the standard LET test for some reason. Such is life.

    Not the biggest, just the best: IPXcore
  • @Damian people are people. They like big, and fast things.

  • Also, it puts providers who have ultra low-end boxes (Sub 128MB RAM) at an advantage, as they'd obviously be cheaper.

    How about a ram standardisation? Multiply/Divide the price as if you're 'stacking' packages together until you reach the 'standard' amount of RAM. e.g. Supposing the standard is 256MB, and you're using a BuyVM $15/year box with 128MB RAM, you'd then multiply the price by two giving you the stacked 256MB.

    If that sounds confusing, it seemed better when I thinking about it :<

    Thanked by 1djvdorp
  • @eastonch PM me the command you used to bench. There's one here with the email in the wrong place.

    @ElliotJ I'll post an update after a few days, lets see if the results surprise us or not :)

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • @serverbear I just got the e-mail... lol. First time it's ever sent one from the 10 i've ever done.

  • @eastonch Strange, you got some sort of Government server with no access to the outside world? :P

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • Possibly. @serverbear.

    My ShardHost Test VPS is actually doing rather well with HDD's racking 184MB/s :}

  • qjqqjq Member

    can you make it work on freebsd?? like not assuming either yum or apt and using gdd

  • @qjq Only supports on CentOS, Red Hat, Debian & Ubuntu atm. You might be able to get it to run if you install the packages yourself first.

    @miTgiB Seems to have the highest IO I've seen so far.

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • miTgiBmiTgiB Member

    @serverbear said: Seems to have the highest IO I've seen so far.

    Well, the IO won't be changing, but I am resubmitting a new test as I forgot to downgrade the container from 1gb to 512mb, so gotta follow the rules ya know ;)

    Hostigation High Resource Hosting - SolusVM OpenVZ/KVM VPS
  • happelhappel Member

    @serverbear said: PM me the command you used to bench. There's one here with the email in the wrong place.

    The page on github states a different order of parameters than your instructions in this thread.

  • Nick_ANick_A Member

    @Damian said: I have yet to determine how writing a contiguous 1gb file is indicative of anything anyone is going to do with their VPS, since I can't think of any 'normal' processes that would write a contiguous 1gb file. It is, however, the standard LET test for some reason. Such is life.

    Good point. Real usage is the only true test.

    RamNode: High Performance SSD and SSD-Cached VPS
    Atlanta - Seattle - Netherlands - IPv6 - DDoS Protection - AS3842
  • avantiavanti Member

    I don't really get this, what does this test prove? That some providers will have newer hardware than others hence score higher? Surely thats obvious!

    The proof is in using the service, I would rather have an average server with good support and reliablility which works than a provider who can wave a top benchmark in front of me but wont answer a ticket or has frequent downtime.

    Just my 2c

  • my AWS EC2 will win cos it's Free

    Laf

  • @happel Both syntaxes work afaik, I'll double check with the guys when I get into the office to make sure.

    @avanti We're working on trying to simulate real world usage. It's not easy & involves a much longer test period (plus everyone uses their box for different things).

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • zserozsero Member

    Guys, this test is totally stupid as it is now. Please read an Anandtech SSD review to understand what is a proper way to measure I/O performance. dd is not! The only thing this test does at the moment is make some load for some poor LEB provider, without providing any real value about I/O performance.

    OK, for example from the Samsung 830 review: http://www.anandtech.com/show/4863/the-samsung-ssd-830-review/2#1BLQx9dqD5eYCM6g.99

    This paragraph is included in all review:

    The four corners of SSD performance are as follows: random read, random write, sequential read and sequential write speed. Random accesses are generally small in size, while sequential accesses tend to be larger and thus we have the four Iometer tests we use in all of our reviews.

    A proper test would be to record a typcal workload for a LEB server and play it back, like in Anandtech. If this is too complicated, at least use ioping, iometer, etc.

    For an other reference, look at StorageReview: http://www.storagereview.com/samsung_ssd_830_review_256gb

    Thanked by 1djvdorp
  • A proper test would be to record a typcal workload for a LEB server and play it back, like in Anandtech. If this is too complicated, at least use ioping, iometer, etc.

    The reports have IOPS data, I'll post IOPS per $ spent too for reference.

    ServerBear - Easy UnixBench/dd/IOPS/FIO (NEW) & Network Benchmarks | Example Report | Compare Low End Boxes
    Gleam - Run kick-ass viral competitions & rewards to grow your userbase. Free until Sept.
  • @serverbear, IOPS of ioping -RD I think is the best from the tests done, but the question caching is a good one. A well implemented SSD cache with HDD drives should be awarded almost as much as a pure SSD solution, but I don't know what does ioping return in such cases.

Sign In or Register to comment.