Howdy, Stranger!

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

Subscribe to our newsletter

Advertise on LowEndTalk.com

Latest LowEndBox Offers

    › SSDNodes RAM is not fully utilized
    New on LowEndTalk? Please read our 'Community Rules' by clicking on it in the right menu!

    SSDNodes RAM is not fully utilized

    pkrpkr Member

    I have an "8GB RAM" VPS from SSDNodes. When I got that VPS, the RAM usage was ~7gb and CPU usage was very low. As a result, the VPS was really fast. But after 10-15 days, I found that RAM usage has come down to ~1gb and CPU usage has gone up, which has affected the speed of the website. I have not changed anything on my website.
    I am not sure what is the problem. Can someone please help me. How can I fix it?

    Thanked by 1uptime

    Comments

    • seriesnseriesn Member, Provider
      edited November 9

      What did their customer support say? It could be many things or combinations of multiple things.

      Could be them, could be you. Could be both?

    • What CPU and what's ur top show

      relentless collector of highest clocked, highest performing KVM/NVMe/Gbit VPSes at the most competitive rates. just to hard idle them. zero knowledge on coding/programming; a mere hobbyist.

    • Problem is RAM? Move to RamNode.

      Thanked by 2cybertech vyas11
    • @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      Best to put in a ticket with their support - but probably best not to expect too much from them. If you do the math for the resources they advertise for a given price, heavy overselling seems inevitable.

      Thanked by 2ITLabs feezioxiii

      the Amitz.party lives on!

    • @uptime said:
      @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      Best to put in a ticket with their support - but probably best not to expect too much from them. If you do the math for the resources they advertise for a given price, heavy overselling seems inevitable.

      How this works? Isnt KVM supposed to have good isolation of resources

      relentless collector of highest clocked, highest performing KVM/NVMe/Gbit VPSes at the most competitive rates. just to hard idle them. zero knowledge on coding/programming; a mere hobbyist.

    • ITLabsITLabs Member
      edited November 9

      @uptime said:
      If you do the math...

      Earlier today we did the math and things went wrong.

    • seriesnseriesn Member, Provider

      @cybertech said:

      @uptime said:
      @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      Best to put in a ticket with their support - but probably best not to expect too much from them. If you do the math for the resources they advertise for a given price, heavy overselling seems inevitable.

      How this works? Isnt KVM supposed to have good isolation of resources

      Albeit,
      It can be oversold, and if done recklessly, can and will cause chaos.

      Will I do it?. Nah. I am too much of a woss according to people that knows me. I like my life simple.

      But, it is definitely possible and thus you see 8GB deals for 5 bucks a month now a days :) I have seen bunch of reputable providers, who publicly admitted of doing such. As long as nodes are monitored and well maintained, I see no issue with that. But if you have neighbors, who are hungry and instead of idling, actually uses their resource, now we got problem.

    • uptimeuptime Member
      edited November 9

      cybertech said: How this works

      I think one way disk can be oversold with KVM may be by using LVM for the node's underlying storage. Like many things in life, this might actually work smoothly enough - until it doesn't anymore.

      As for ram - probably a couple ways that could happen. Most basic would be for the node to use swap space (preferably on a fast NVMe volume). And there is also ram compression to reckon with - see zram:

      After four years in Linux' driver staging area, zram was introduced into the mainline Linux kernel in version 3.14, released on March 30, 2014. From Linux kernel version 3.15 onwards (released on June 8, 2014), zram supports multiple compression streams and multiple compression algorithms. Compression algorithms include LZ4 and LZO. The default is LZ4, which is faster at compressing/decompressing, but does not compress quite as efficiently as LZΟ. Like most other system parameters, the compression algorithm can be selected via sysfs.

      Long story short - these are not necessarily terrible things for a competent provider to make good use of. But it certainly adds however many extra layers of potential complication - especially when things go wrong. I suspect some inexperienced providers may easily get sucked into these sorts of shenanigans and eventually end up way over their heads as they oversell themselves into oblivion. Would recommend to carefully consider the price/resource ratio to gauge likely level of overselling - and maybe discuss with a potential provider if you have concerns. Some who dabble in these dark arts may be more willing to be transparent about their "secret sauce" than others, so that may be another indicator for your thoughtful contemplation when evaluating a provider.

      tl;dr: how it works is basically you pays your money and you takes your chances! :)

      the Amitz.party lives on!

    • jarjar Provider

      @uptime said:
      @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      Best to put in a ticket with their support - but probably best not to expect too much from them. If you do the math for the resources they advertise for a given price, heavy overselling seems inevitable.

      I have it on good authority that they're a pretty legit outfit, I'd examine software stack first.

      Thanked by 1uptime
    • pkrpkr Member

      @seriesn said:
      What did their customer support say? It could be many things or combinations of multiple things.

      Could be them, could be you. Could be both?

      They just sent me the output of the "free" command. Yes, there are a few MBs in the cache, but VPS now using just 1.5/8 gb seems fishy to me. The same website was hosted on another provider's VPS and it was using ~3.7/4 gb and some swap too, so I decided to go for 8gb RAM of SSDNodes. The website gets <1000 pageviews, so high CPU usage concerns me. But, I cannot do anything now as I have paid for 3 years. :(:(:(

    • @pkr can you give details of the applications you have on the box? Difficult to diagnose without details.

      Avoid scams and stay updated with legitimate great deals: LowEndBoxes Review | The LEBRE Whitelist of non-scammy lowend hosts | Join the #lexit party sans LEB scams.

    • seriesnseriesn Member, Provider

      @pkr said:

      @seriesn said:
      What did their customer support say? It could be many things or combinations of multiple things.

      Could be them, could be you. Could be both?

      They just sent me the output of the "free" command. Yes, there are a few MBs in the cache, but VPS now using just 1.5/8 gb seems fishy to me. The same website was hosted on another provider's VPS and it was using ~3.7/4 gb and some swap too, so I decided to go for 8gb RAM of SSDNodes. The website gets <1000 pageviews, so high CPU usage concerns me. But, I cannot do anything now as I have paid for 3 years. :(:(:(

      Boss man,
      Please share results of the followings(preferably screenshots as text copy paste will be hard to read ) :

      top
      htop

    • pkrpkr Member

      @cybertech said:
      What CPU and what's ur top show

      CPU: 2x CPU E5-2650 v4 @ 2.20GHz

      systemd, php-fpm, mysqld and nginx together are using ~12% of the CPU. So, it seems that I am in a bad neighborhood.

    • pkrpkr Member

      @ITLabs said:
      Problem is RAM? Move to RamNode.

      I have paid to SSDNodes for 3 years because the cheap price was available only for the 3-year contract. I also cannot get a refund as they have a 14-days refund policy. :(

    • pkrpkr Member

      @uptime said:
      @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      Best to put in a ticket with their support - but probably best not to expect too much from them. If you do the math for the resources they advertise for a given price, heavy overselling seems inevitable.

      Seems true.

    • @pkr said:

      @cybertech said:
      What CPU and what's ur top show

      CPU: 2x CPU E5-2650 v4 @ 2.20GHz

      systemd, php-fpm, mysqld and nginx together are using ~12% of the CPU. So, it seems that I am in a bad neighborhood.

      As seriesn mentioned would be good to have a look at those screenshots. 12% isn't really high use?

      relentless collector of highest clocked, highest performing KVM/NVMe/Gbit VPSes at the most competitive rates. just to hard idle them. zero knowledge on coding/programming; a mere hobbyist.

    • pkr said: which has affected the speed of the website.

      How does it affect your website?
      Any benchmark?

    • pkrpkr Member

      @greattomeetyou said:

      pkr said: which has affected the speed of the website.

      How does it affect your website?
      Any benchmark?

      I use Google's pagespeed/lighthouse/mobile-friendly test, GTmetrix, and webpagetest to see the response time of my website. These applications are not happy. :(

    • jarjar Provider

      uptime said: @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      I found a reference to what I was talking about before, check this beefy sucker out and that was two years ago:

      Enough up front and you can really sell some hardcore ram.

      Thanked by 3uptime poisson elliotc
    • jar said: hardcore ram

      um ... giggity!

      Thanked by 1jar

      the Amitz.party lives on!

    • @jar said:

      uptime said: @pkr - I wouldn't be surprised to find SSDNodes relies heavily on thin provisioning and/or other methods for ram "compression" (such as zram ... this may or may not explain the increased CPU utilization you're now noticing.

      I found a reference to what I was talking about before, check this beefy sucker out and that was two years ago:

      Enough up front and you can really sell some hardcore ram.

      Or sell even more , hardcore + softcore

      IYKWIM

      Thanked by 3jar uptime ITLabs

      relentless collector of highest clocked, highest performing KVM/NVMe/Gbit VPSes at the most competitive rates. just to hard idle them. zero knowledge on coding/programming; a mere hobbyist.

    • vimalwarevimalware Member
      edited November 9

      Maybe it takes time to warm up your caches.
      What is the application stack? Are you using a php fpm worker pool?
      Just set a high static number of workers in fpm config (default is not 'static') if you just want to see more RAM being reserved by your stack.

      Post 30s worth of output of 'vmstat 1' (screenshot would be easier to read)

      Down with the sic_K_ness...
      The doctor might be going K-razy with the low end spirit. 😉

    • pkrpkr Member

      @vimalware said:
      Maybe it takes time to warm up your caches.
      What is the application stack? Are you using a php fpm worker pool?
      Just set a high static number of workers in fpm config (default is not 'static') if you just want to see more RAM being reserved by your stack.

      Post 30s worth of output of 'vmstat 1' (screenshot would be easier to read)

      I played with different configurations, but nothing worked. I moved the same website to another host, the RAM usage has now reached 4.2/5 gb on that VPS and the cache is not cleaned up. I am now 100% convinced that SSDNodes is doing something wrong or there are some issues with their configurations. That 8gb RAM is not real; maybe swap/zram/something else.
      The problem with SSDNodes's VPS is the cache is cleared once RAM usage reaches ~2gb. Why will it happen when the VPS has 8gb RAM?

    • @pkr said:
      The problem with SSDNodes's VPS is the cache is cleared once RAM usage reaches ~2gb. Why will it happen when the VPS has 8gb RAM?

      Oh that definitely smells.
      I would be pissed if a host regularly cleared my pagecache on a KVM (via KSM?)

      @SSDNodes

      Down with the sic_K_ness...
      The doctor might be going K-razy with the low end spirit. 😉

    • jackbjackb Member, Provider
      edited November 10

      @vimalware said:

      @pkr said:
      The problem with SSDNodes's VPS is the cache is cleared once RAM usage reaches ~2gb. Why will it happen when the VPS has 8gb RAM?

      Oh that definitely smells.
      I would be pissed if a host regularly cleared my pagecache on a KVM (via KSM?)

      @SSDNodes

      Ksm wouldn't do that. The VM would report the memory as used and it would be transparently merged with other identical pages on the host.

      Whatever this is, it is likely being triggered within the VM. Maybe a script on cron, or maybe some sysctl tweaks.

      I am not aware of any sysctl tweaks that would do this.

      Thanked by 1vimalware

      Afterburst - Awesome OpenVZ&KVM VPS in US+EU

    • TheLinuxBugTheLinuxBug Member
      edited November 10

      Easiest way to test this I can think of is boot the VM and without anything running, cd to /dev/shm and then dd/copy a 7.5GB file there. /dev/shm allows you to write regular files into memory for storage, so if you can't even write or copy the full 7.5GB there (leaving a little room for overhead there) then it sounds like they are likely over provisioning ram in some way for sure.

      (You will know if it fails, it will just report it ran out of space while writing..)

      my 2 cents.

      Cheers!

      Thanked by 3seriesn uptime MrH

      Have an Allwinner H3 device? Android? Check out H3Droid! | Lichee Pi Zero - The 6$ SBC | #SYSarm - Get It! | Atomic Pi - $35 x86 SBC
      20+ Years IT Experience in Linux/Windows Hosting, Administration and Development

    • Their servers cost is around $2000 per month( yes they use so good servers) so with 1tb ram , so 2 bugs per ram, if you have a service of 8gb ram, only their cost is around $16, and the selling price is $4. Dont need to test to know that of course its oversold in many ways.

      I tried one year of their 16gb plan, and of course they never are near of their website benchmarks. They are better than others ( because use very robust servers ) but of course their resources as ram arent only yours.

    • jarjar Provider

      optisoft said: but of course their resources as ram arent only yours

      I have to say I'm impressed if they've managed to significantly oversell ram on kvm. I've never found a reliable way of doing it beyond just a few gigs, and even then it wasn't really worth it for the cpu cost of ksm.

    • jar said: I've never found a reliable way

      Sell very big ram for people that use small ram is a good way :blush:

      "Magnum XXXXL" plans sounds sexy

    • jarjar Provider

      optisoft said: Sell very big ram for people that use small ram is a good way

      I mean like kvm tends to take pretty much all of the memory that you allocate regardless of usage, and I never found ksm able to recover an amount close to what wasn't used inside the guest.

      Maybe there's a way that I'm not aware of, would be interested to learn it.

    • pkrpkr Member

      @optisoft said:

      >

      They are better than others ( because use very robust servers ) but of course their resources as ram arent only yours.

      >

      True. Their servers are really robust. That's why I have been using their VPS for ~4 years. But this problem popped up recently. Never had this issue in past.

      Thanked by 1vimalware
    • I bought a 10x performance 24G VPS last year for $191.99.
      Now the price is $149 with more disk and cpu cores.

      I'm worried about how they make the business sustainable, yes the memory and ssd price is dropping significantly, but moore's law would not work every time.

    • I wonder if his app is only using 2gb and SSDNodes is using ballooning to reclaim the unused portion where is other hosts aren't reclaiming.

      So he believes the app is using more, but it actually is not and is only reserving it.

      This method of ballooning might also account for cpu usuage.

      Without him posting some tests we ask for all we can do is guess. It is really all up to him to run the tests and post the screens.

      The Bun | Primary DNS | Small Web <--- Do not ask questions, just go for it.

    • @pkr said:

      @optisoft said:

      >

      They are better than others ( because use very robust servers ) but of course their resources as ram arent only yours.

      >

      True. Their servers are really robust. That's why I have been using their VPS for ~4 years. But this problem popped up recently. Never had this issue in past.

      I've used them too for quite awhile and ocassionally had my VPSs slow down and get bogged up. Twice it was due to some issue with the (I don't know the terminology here..) system they were using: once it was some update problem with their Virtuozzo 7 upgrade, and the other time was an issue related to transitioning from Virtuozzo to a proper KVM. Both times there was no notice that an upgrade or change was going to happen. But both times they got it fixed. And both times I had more than one service with them, and the other service was fine (already having been upgraded or changed before I started using it. I guess). So definitely open a ticket, they may be doing something to "improve their service". * I actually have a 24Gb Ram service with them right now and it has performed superbly.

    Sign In or Register to comment.