Page 1 of 2 12 LastLast
Results 1 to 25 of 30
  1. #1
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932

    Question Any idea why network speeds would be RAM dependant on Xen?

    We've had some complaints about our network speeds on our Xen VPSs in the past and we can't figure out why the amount of RAM a VPS has will impact the download speeds. We've configured the same exact VPS with 4 different amounts of RAM (same exact OS, sub-20MB usage in free -m and never uses swap for any amount), here are the results (using curl and wget return the same numbers):

    • 96MB - 2.9MB/s
    • 128MB - 4.9MB/s
    • 256MB - 10.3MB/s
    • 512MB - 10MB/s


    I would understand if the RAM was maxing out during the transfer but since no swap is being used it doesn't look like that is the issue.

    The only other difference between the plans is the amount of disk space offered but I don't see how that can be a factor since even with our lowest plan we still have 80% free according to df.

    Any ideas?
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  2. #2
    If you had showed me a 1gb, 2gb, 4gb and 8gb ram VPS, I would scratch my head. With 96mb, 128mb, 256mb, 512mb, no, it does not surprise me.

    Buffers used by the OS to speed up everything from disk access to tcp, there just isn't ram for that with so little available. Maybe you could say, ok, I'm not "out of ram" so use more for these buffers, but linux is not going to do that without tweaking. The sensible default is that when you're constrained so much on ram like that, not to waste any on anything that's not absolutely necessary.

    From the look of it, you get the correct performance with 256mb and up. Do you really want to be dealing with VPS's smaller than that anyway?
    IOFLOOD.com -- We Love Servers
    Phoenix, AZ Dedicated Servers in under an hour
    ★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
    Contact Us: sales@ioflood.com

  3. #3
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Our primary market is Low End Boxes. I think our 96MB plans make up over 75% of our clients. Our OpenVZ boxes have no issues maxing out the 100Mbps port we provide which is strange. Oh well, we have a backup plan in the event this isn't straightened out.
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  4. #4
    Quote Originally Posted by ZKuJoe View Post
    Our primary market is Low End Boxes. I think our 96MB plans make up over 75% of our clients. Our OpenVZ boxes have no issues maxing out the 100Mbps port we provide which is strange. Oh well, we have a backup plan in the event this isn't straightened out.
    Ah, ok. For what it's worth, openvz provides glorified chroot jails. Because it's not a fully virtualized environment, some of these things aren't issues that are on Xen. For a really low end situation, openvz makes more sense than Xen.
    IOFLOOD.com -- We Love Servers
    Phoenix, AZ Dedicated Servers in under an hour
    ★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
    Contact Us: sales@ioflood.com

  5. #5
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Yeah, we've learned that the hard way (I say that because I used to dislike OpenVZ sooo much but now that I am a provider I am appreciating how much better they are than Xen in many aspects). I just wish people didn't associate OpenVZ with overselling which is the whole reason we started offering Xen in the first place. (Sorry to get off topic. )
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  6. #6
    Join Date
    Apr 2010
    Location
    North Carolina
    Posts
    442
    My 128MB KVM VPS gets 20MB/s from a OpenVZ VPS in the same datacenter, so if it's the OS running out of ram why would KVM not be effected but Xen is effected?

  7. #7
    Join Date
    Jun 2011
    Location
    California
    Posts
    412
    I wanted to test your results, so I fired up one of our Debian templates and set it to 128MB of RAM (using XenServer).

    Code:
    root@debian60:~# free -m
                 total       used       free     shared    buffers     cached
    Mem:           119         52         66          0          2         34
    -/+ buffers/cache:         15        103
    Swap:            0          0          0
    root@debian60:~# wget http://speedtest.dal05.softlayer.com/downloads/test500.zip -O /dev/null
    --2011-08-27 01:02:34--  http://speedtest.dal05.softlayer.com/downloads/test500.zip
    Resolving speedtest.dal05.softlayer.com... 173.192.68.18
    Connecting to speedtest.dal05.softlayer.com|173.192.68.18|:80... connected.
    HTTP request sent, awaiting response... 200 OK
    Length: 524288000 (500M) [application/zip]
    Saving to: ?/dev/null?
    
    100%[======================================>] 524,288,000 68.8M/s   in 7.3s    
    
    2011-08-27 01:02:42 (68.9 MB/s) - ?/dev/null? saved [524288000/524288000]
    
    root@debian60:~#
    I ran the test a few times, seems to be stable at 550.4 megabits.

    I don't know what could be causing you to only get 4.9MB/s
    Last edited by Matt Connor; 08-27-2011 at 01:11 AM.
    |███| SSD Nodes
    |███| Build more for far less with the best value in cloud hosting.
    |███| RAID 10 SSDs | 10Gbps NICs | Tier-1 backbone | Scale without downtime | 9 years in business
    |███| https://www.ssdnodes.com

  8. #8
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Very weird! I just ran the same test on our internal network (like SSD Nodes did) and I was able to get over 60MB/s (after setting the port to 1Gbps)...
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  9. #9
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    @funkywizard:

    For a really low end situation, openvz makes more sense than Xen.
    Do you imply with this that Xen, because it is fully virtualised, is a way better choice then OpenVZ when in need of a high-end VPS?

    As for the problem of ZKuJoe the difference between OpenVZ and Xen could explain the cause. Although I'm not a Xen-expert, but I believe it divides all resources on the hardware-node (thats why it is full virtualised). The way the resources are divided can vary between different versions of Xen and you can even tweak a lot. It nowadays is even possible to oversell with Xen, although slightly more configuration is needed compared to OpenVZ.

    Could it be that the provider simply divides the amount of networkbandwith according to the package? So a small VPS that has 5% of the node only gets 5% of the 100Mbit. That would explain it. Is that a configuration option in Xen?

    OpenVZ can not do that. It gives all resources to everyone* (apart from some buffers that can impact network-performance) and only when there's a shortage it starts to throttle and divide things. This may be a pro or a con, but thats an other discussion. It could explain the weird differences between Xen servers.

  10. #10
    Quote Originally Posted by rimote View Post
    @funkywizard:



    Do you imply with this that Xen, because it is fully virtualised, is a way better choice then OpenVZ when in need of a high-end VPS? If so, please first learn some more about what OpenVZ is and how it works before you state such things. It simply is not true.

    A lot of people have invested time and effort in OpenVZ with great results that can max out all other forms of virtualisation. I think it is not fair to them to simply dishonor it in this manner.

    As for the problem of ZKuJoe the difference between OpenVZ and Xen could explain the cause. Although I'm not a Xen-expert, but I believe it divides all resources on the hardware-node (thats why it is full virtualised). The way the resources are divided can vary between different versions of Xen and you can even tweak a lot. It nowadays is even possible to oversell with Xen, although slightly more configuration is needed compared to OpenVZ.

    Could it be that the provider simply divides the amount of networkbandwith according to the package? So a small VPS that has 5% of the node only gets 5% of the 100Mbit. That would explain it. Is that a configuration option in Xen?

    OpenVZ can not do that. It gives all resources to everyone* (apart from some buffers that can impact network-performance) and only when there's a shortage it starts to throttle and divide things. This may be a pro or a con, but thats an other discussion. It could explain the weird differences between Xen servers.
    I'm implying that openvz might be more efficient when you don't have a lot of resources available.
    IOFLOOD.com -- We Love Servers
    Phoenix, AZ Dedicated Servers in under an hour
    ★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
    Contact Us: sales@ioflood.com

  11. #11
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    OK, you're right about that. I reread your post (after posting... not the best way, i know) and figured this out an erased a part of my post.

  12. #12
    Quote Originally Posted by rimote View Post
    OK, you're right about that. I reread your post (after posting... not the best way, i know) and figured this out an erased a part of my post.
    No worries.
    IOFLOOD.com -- We Love Servers
    Phoenix, AZ Dedicated Servers in under an hour
    ★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
    Contact Us: sales@ioflood.com

  13. #13
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Quote Originally Posted by rimote View Post
    Could it be that the provider simply divides the amount of networkbandwith according to the package? So a small VPS that has 5% of the node only gets 5% of the 100Mbit. That would explain it. Is that a configuration option in Xen?
    Why would I be able to get over 550MBps internally when I set the port speed to 1Gbps (but still only get less than 30Mbps for external connections)?
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  14. #14
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    maybe because internal connections go over a loopback adapter and not a real networkcard?

  15. #15
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    It would have to hit the router though so it would have to go over the network card. I meant internal to the network, not the server. Sorry for the confusion, I literally just woke up.
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  16. #16
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    OK, so you have two networks. A external (100 Mbit?) and an internal (1 Gbit?)

  17. #17
    Join Date
    Jul 2010
    Location
    ~/
    Posts
    1,382
    Hi ZKuJoe,

    Had exactly the same issue and got it resolved in the last week or so, I am still doing comparisons right now but the bottom line (Not meant to be an attack) The Xen templates from SSM have this issue the ones from stacklet dont, the difference being on a 96mb plan (Our micro package) after a massive amount of tweaking I could get it no faster than 12 MB/s after switching to the stacklet templates which are just XenSource conversions (I think) It shout up to around 80 MB/s

    Would be interested in working with you on a template comparison to nail this down if you like?
    -> INCEPTION HOSTING LIMITED Since 2010!
    -> I am most active on the lowendspirit hosting forum Come join us!
    -> PHOENIX USA & THE NETHERLANDS & UK EU

  18. #18
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Quote Originally Posted by rimote View Post
    OK, so you have two networks. A external (100 Mbit?) and an internal (1 Gbit?)
    No, we have 1 network (i.e. all of our NICs have public IPs, no private VLANs or anything). I configured the VPS port to 1Gbps for testing but normally it is 100Mbps (i.e. 11.3MB/s which I cannot achieve with any connection outside of the data center unless I have 256MB of RAM or more).
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  19. #19
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Quote Originally Posted by backtogeek View Post
    Hi ZKuJoe,

    Had exactly the same issue and got it resolved in the last week or so, I am still doing comparisons right now but the bottom line (Not meant to be an attack) The Xen templates from SSM have this issue the ones from stacklet dont, the difference being on a 96mb plan (Our micro package) after a massive amount of tweaking I could get it no faster than 12 MB/s after switching to the stacklet templates which are just XenSource conversions (I think) It shout up to around 80 MB/s

    Would be interested in working with you on a template comparison to nail this down if you like?
    We are using Stacklet for our Xen templates already (and we did try multiple templates for testing with no change in speeds).
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  20. #20
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    thats strange then. Hope backtogeek has offered the solution. Good luck

  21. #21
    Join Date
    Jul 2010
    Location
    ~/
    Posts
    1,382
    Thats odd, do you want a test VPS on aour 96mb plan so you can pull it apart and have something offsite to compare it to?
    -> INCEPTION HOSTING LIMITED Since 2010!
    -> I am most active on the lowendspirit hosting forum Come join us!
    -> PHOENIX USA & THE NETHERLANDS & UK EU

  22. #22
    Join Date
    Jan 2011
    Location
    Groningen, Netherlands
    Posts
    23
    funkywizard's first posts seems still relevant. Less RAM can have an effect on networkperformance. But why are other people getting better results is still strange. Are you sure it has nothing to do with configuration of Xen? Xen can use dynamic RAM allocation. Maybe you have this setting off and other on. That could maybe explain the issue. The best way to compare is to have a good look at the versioning and configuration of Xen and also the kernel running on the VPS. Else it is not a good comparison. Maybe this way you can find the difference between your install and other.

  23. #23
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Which network card are you using in SolusVM for your 96MB VPSs?
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

  24. #24
    Join Date
    Jul 2010
    Location
    ~/
    Posts
    1,382
    Just the default Realtek 8169.

    Don't use cachefly as a baseline BTW its never gives constant speeds, when this issue still hit it was driving me mad and it turned out cachefly was only spitting out at 7 MB/s at the time.

    Example:
    Old Template from SSM:

    root@96mb:~# wget http://files.ecatel.nl/speedtest/100mb.bin
    --2011-08-27 08:14:58-- http://files.ecatel.nl/speedtest/100mb.bin
    Resolving files.ecatel.nl... 89.248.167.16
    Connecting to files.ecatel.nl|89.248.167.16|:80... connected.
    HTTP request sent, awaiting response... 200 OK
    Length: 104857600 (100M) [application/octet-stream]
    Saving to: `100mb.bin.2'

    100%[======================================>] 104,857,600 9.04M/s in 11s


    NEW Template from Stacklet:


    root@96mb:~# wget http://files.ecatel.nl/speedtest/100mb.bin
    --2011-08-27 08:24:38-- http://files.ecatel.nl/speedtest/100mb.bin
    Resolving files.ecatel.nl... 89.248.167.16
    Connecting to files.ecatel.nl|89.248.167.16|:80... connected.
    HTTP request sent, awaiting response... 200 OK
    Length: 104857600 (100M) [application/octet-stream]
    Saving to: `100mb.bin.2'

    100%[======================================>] 104,857,600 27.8M/s in 3.6s
    -> INCEPTION HOSTING LIMITED Since 2010!
    -> I am most active on the lowendspirit hosting forum Come join us!
    -> PHOENIX USA & THE NETHERLANDS & UK EU

  25. #25
    Join Date
    Feb 2007
    Location
    Florida
    Posts
    1,932
    Yeah, we don't use Cachefly because they are to slow for testing but for some reason people love to use them as the benchmark (which is funny because even if we show them faster speeds from other data centers they don't believe them because it's not Cachefly).

    I've been able to double the speed (40Mbps) by using a Xen Template without PyGrub but considering the majority of our clients prefer one of the Templates with PyGrub (i.e. Debian), this doesn't help us any.
    -Joe @ Secure Dragon LLC.
    + OpenVZ Powered by Wyvern | KVM | cPanel Hosting | Backup VPSs | LowEndBoxes | DDOS Protection
    + Florida | Colorado | Illinois | California | Oregon | Georgia | New Jersey | Arizona | Texas

Page 1 of 2 12 LastLast

Similar Threads

  1. Replies: 1
    Last Post: 03-30-2011, 09:25 AM
  2. Same DC & Network, different speeds
    By Xandrios in forum Other Reviews
    Replies: 8
    Last Post: 12-25-2006, 09:13 PM
  3. Need with with testing network speeds.
    By Lpal-Patrick in forum Dedicated Server
    Replies: 6
    Last Post: 06-12-2006, 05:37 PM
  4. Javascript Browser Dependant?
    By stripeyteapot in forum Web Design and Content
    Replies: 2
    Last Post: 09-02-2004, 01:23 AM

Tags for this Thread

Posting Permissions

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