Results 1 to 30 of 30
  1. #1

    Any VPS hosts with less than 10 accounts per server?

    Are there any VPS hosts with less than 10 accounts per server? If so, please tell me everyone you know.

    I've noticed that some VPS providers dump 50+ accounts on one server and I find that unacceptable because the current shared host server I'm on has 71 accounts and the computer is of the same quality if not better than some VPS ones.

  2. #2
    Join Date
    Aug 2006
    Posts
    103
    It really depends on the hardware specifications, if you have a server with 32GB RAM it isnt going to only have 10 accounts on it.
    || Josh Anderson - CVO A&M Hosting
    || A&M Hosting | Virtual Private Servers
    || US | UK
    || No overselling guaranteed

  3. #3
    Join Date
    Aug 2006
    Posts
    103
    Some hosts will state the amount they put on the server, it may or may not be true. The only real way to tell would be to go with a Xen VPS provider.
    || Josh Anderson - CVO A&M Hosting
    || A&M Hosting | Virtual Private Servers
    || US | UK
    || No overselling guaranteed

  4. #4
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    Xen providers can definitely lie about this if they want (ie. to present CPU guarantees or contention ratios based on 8GB, but really use 16GB of RAM -- it will half the CPU and double the contention).
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  5. #5
    Join Date
    Oct 2006
    Location
    Everywhere!
    Posts
    32
    Quote Originally Posted by ezXEN
    You arent allowed to advertise your own services
    Sorry, i'm new to this forum. I was just trying to help.

    Quote Originally Posted by ezXEN
    and that sounds like a low quality host node if it is only 1 CPU.
    With quality do you mean speed? Even in this situation a box with a 3.8Ghz CPU and 10 VPS is much more faster that a "quality" one with dual XEON serving 50 VPS

  6. #6
    Join Date
    Nov 2005
    Location
    London UK
    Posts
    683
    It doesn't take a genius to figure out how many VPS nodes are on a server...most if not all oversell just like all other hosting products.

    The word "cheap" is more important than "quality" for most people.
    UK SSD VPS without compromise - upto 250k IOPS !
    www.clustered.net

  7. #7
    Join Date
    Oct 2006
    Location
    Everywhere!
    Posts
    32
    Quote Originally Posted by reiteration
    It doesn't take a genius to figure out how many VPS nodes are on a server...most if not all oversell just like all other hosting products.

    The word "cheap" is more important than "quality" for most people.
    i agree with the above statement

  8. #8
    Join Date
    Aug 2006
    Posts
    103
    Quote Originally Posted by stoidis
    Sorry, i'm new to this forum. I was just trying to help.



    With quality do you mean speed? Even in this situation a box with a 3.8Ghz CPU and 10 VPS is much more faster that a "quality" one with dual XEON serving 50 VPS
    There are many doing the same with dual core woodcrest boxes, so the fact that you only put a few on a lower quality node doesnt essentailly mean it will be faster. Especially seeing with 10 VPS on the server that is only 380MHz guaranteed CPU.
    Last edited by ezXEN; 10-28-2006 at 11:39 AM.
    || Josh Anderson - CVO A&M Hosting
    || A&M Hosting | Virtual Private Servers
    || US | UK
    || No overselling guaranteed

  9. #9
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    In all honesty...what size of VPS's atre we talking? 64MB RAM each X 10? There's no way you would ever see this..they could not pay the server rent.

    See...the number is not important at all...sadly...as I have said before..most hosts will not share resource allocations and CPUunit alocations in regards to virtuozzo and open VPS's.

    If we consider...say...one thing..cpuunits..I have only two numbers to share:

    P4 2.8Ghz 1Gb RAM has 141033 CPUunits
    Dual Core Dell 850 2Gb RAM has 300749 CPUunits

    Now...if you wish to compare...there's something to consider. Base VPs eats at least 1,000 CPU units from the getgo...so that gives you an idea on total available CPUunits on the server.

    Sadly..another thing most hosts don't talk about is connection speed. I'm not talking about NIC card speed (not even relevant) but true actual capped port speed. Some hosts will tell you if you ask..but they never publish these numbers...or I've seen any real cap speeds..just NIC card speed..which has nothing to do with how fast you or your customers can upload and download to the server.

    Now...if we consider memory...it works like this: "real memory" X3 (double swap) =
    allocatable memory. Still..if you consider a 64K server (real memory) and the server has 1MB RAM..well..the math is easy. However...you would not believe the level of virtualization that goes on...of a host keeps an eye on actual utilizations vs allocation...certainly..more VPS's can be put on one server...but at what cost?

    Speed. Stability. These two suffer...and..ohhh..so many of you have to have Cpanel...and..well...bye-bye unused CPU on that.

    So there you go...number means nothing. Hardware counts..port cap speed counts...and how well these are managed by the host count as well.

    Oh..and don't even get me started on the folly of people selling Cpanel hosting on VPS's. Simple test:

    1) Put a 10K..30K...60K file on your vps server w/cpanel
    2) Get a decent shared account...on a "normal" server and put the same files there

    3) Go to alertra and do the tests..and learn how horrible web servers vps's make when they have cpanel on them.

    Bryon
    Last edited by bryonhost1; 10-28-2006 at 01:17 PM.
    Bryon L Harvey
    Soil Relocation Engineer

  10. #10
    So basically if someone manages to offer a decent performance VPS the price of it is more than that of a similar performance dedicated server. (Same level of management.)

    Min 384 MB dedicated Ram and Min burstable to 768.
    Last edited by Tertsi; 10-28-2006 at 01:22 PM.

  11. #11
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    Basically..yes. Even though the performance of any vps solution is impressive...there is a certain amount of server CPU lost to management of the VPS's. I believe that a "upgrade" from shared to vps..is no upgrade at all. What's more..I have numbers that back up that...it's the truth. I think that more vps hosts need to focus on the market segment of people who are outgrowing shared..and are looking into dedicated servers. Big jump there...even if you are considering a so called "managed" server.

    Some hosts brag about server specs...but who cares? They charge you for 1/10 of the server resources..and you would be lucky if you get 1/20th of those resources.

    I mean...look at the cost differences between the two servers I currently have:
    141033CPUunits@$79.00/month
    [email protected]/month

    The sad thing is..the more expensive server has a slower connection than the less expensive one. Guess which one I'm going to keep and get even more when needed? Doesn't take a brainiac to figure that one out

    I would also watch out...some hosts have had users abuse the burst...so they took it away...for everybody. The sad thing is...these resources are allocated on a per user basis. In other words...they could have removed the burst option on the abusers...and left it on the ones that were not abusing it...it just was easier to get rid of it altogether..less to do.

    It all boils to down to one thing: Hosts that treat their customers fairly..and those who don't. The only way to learn is by trying...and nobody has tried more than me.
    In last three years..I would say I have had 30 VPS's with 24 different hosts.
    That's why I make such a big point: You really can't be a good host..until you have been a hosting customer...with all sorts of hosts.

    My experience with a so-called managed server was that this was a outright lie...and if I could not do certain things myself...they did not get done. There is quite a bit to be done...more even for a good vps host.

    There are tons of different opportunities...in any market you look at. The key is:
    1) Find the need
    2) Make sure this need is one they will actually pay for <cough-made this one last year>
    3) Find a way to reach your customers in a cost-effective way
    4) Test everything before offering new things to customers
    5) Talk to your customers...start building relationships
    6) Nurture and care for these..your first few customers can be the difference between making it..or not making it in the long run.

    As many accounts as I have had..how many do you think thought to just send a email to make sure everything was going ok? Do you need anything? Do you have any questions?

    One. It happened once. One time in three years..pathetic, ain't it?

    Oh..I've did it again. I've said *too* much. Will they ever learn? I doubt it. One can always hope.

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  12. #12
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    Your understanding of CPU units is quite skewed. The CPU units are a proportional share system and are dynamically calculated based on how many units have been allocated. Comparing your "300749" units to say the 1000 units assigned to a VPS a VPS has no bearing on performance of that VPS. What counts is how many units the other VPS's on the host node have and what their usage is.

    If you have 2 VPS's both with 1000 units both trying to use 100% of the CPU each will receive 50%, start up another one with 1000 units trying use all CPU and then all 3 each get 33%. Alternatively start up one with 2000 units and the 2 1000's drop to 25% each and the 2000 unit VPS receives 50% of the CPU.

    The units can be anything as it's a proportional system, going with my above example 1000 and 2000 can be substituted with 1 and 2, 100 and 200, 10000, and 20000, etc... what matters is the proportion to other VPS's, not the total units.
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  13. #13
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    The 1000 I was talking about was consumed by the host node...not a vps at all.
    My understanding of that is not wrong...nor skewed.

    I did learn something from this, though. It is a minimum amount of the CPU that will be available to the user. I was not aware that if cpulimit is not set...a vps can consume more than that amount if no other VPS is competing for the CPU.

    CPUunits do not change...it's set number. That is a fact...unless you add another processor...then..it would change.

    You are correct...this would work if cpulimit has not been set.

    I don't see how anything I said was skewed, though.

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  14. #14
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    I think the confusion was..that I have never set cpulimit myself..but had never seen any VPS take more than it was allocated.

    Example (On betsy)
    [root@ ~]# vzcpucheck
    Current CPU utilization: 206000
    Power of the node: 300746
    [root@~]# vzcpucheck -v
    vpsid units
    -----------------------
    0 1000 <<<This was what I was talking about-overhead-no vps
    192 5000
    867 100000
    426 100000
    Current CPU utilization: 206000
    Power of the node: 300746

    There are only three vps's running on betsy right now..the 1000 is overhead.
    Utilization is high...because I wanted it to be so.

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  15. #15
    Join Date
    Jun 2002
    Location
    Waco, TX
    Posts
    5,291
    Hello Byron,

    Just as a quick not with virtuozzo say you only have 2 VPSes(for simplicity sake only)
    1 is allocated 1000 CPU units
    1 is allocated 1200 CPU units

    The one with 1200 cpu units will have "priority" over the one with 1000 if the cpu is needed. If the VPS with 1000 units needs the cpu and the one with 1200 is not using it, then it can use the full amount of CPU. This is what SWsoft tech support has told me anyway. The unit measure is more cpu priority then a mhz rating as many think.

    This is why most hosts do equal share cpu across all plans, being that each one is allocated the same amount of units. I have experimented with adding one higher cpu units and just one VPS higher in cpu units can in fact slow down other VPSes dramatically. I was testing with [email protected] fromStanford to test 100% cpu usage on a VPS compared to [email protected] on a VPS with less cpu units, the power of the node went to the higher unit VPS, nearly choking the one with lesser to a halt.

    Now, that being said no one should be running [email protected] on a VPS, it will severely heat up the server using 100% and is abusive to other users on the node. It was ONLY a test.

  16. #16
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    I don't know [email protected].the worst case CPU/memory killer I could think of was IBM Websphere Application Server...and Sun's Application server as well..not only bringing Tomcat to the party..but I ton of other stuff I have not even had time to look at. I had no problems with servers allocated 2000 cpuunits and the Java servers 10 times that...everything worked..and worked fine.

    I could not get the server to swap..we had quite a debate on here..and I had to know the truth. It worked fine..of course..there's nothing like having a non-production server to learn on!

    Thanks

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  17. #17
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    Well...I'm at a loss here. I thought it was just a matter of figuring out a fomula..and then running with that. It simply is not so. I'm off to the openvz board to search..and perhaps ask the experts on there for help.

    I found this...and I thought..ok..I now have solid numbers..numbers they recommend..but they do not mesh.. And..of course..vzsplit has the sole purpose of adding even more confusion. This is what I'm up against..let's take one example.

    Ok..2GB RAM 8 vps's. Alright...easy enough..openvz gives me these numbers:

    kmemsize 16,384,000 18,022,400

    As you may know..I have two vps servers...one has 1 gb and the other has 2.
    I figured..ok..if I run vzsplit in a full spread (2,4,8,16,32 VPSs) on both servers...surely..the numbers will be different of each..after all...one has twice the physical RAm of the other. Wrongo!

    server1-1GB RAM kmem: 23052902:25358192 for 8 [email protected] RAM each
    server2-2GB RAM kmem: 22825164:25107680 for 8 [email protected] RAM each

    which makes no sense at all either. Oy!

    What this page did confirm what I have long suspected: many hosts are cheating their users on resources to allow the host to put more people on the server. See for yourself:

    http://wiki.openvz.org/UBC_configuration_examples_table

    Aye! As long as the people don't know they're being cheated...



    Bryon
    Last edited by bryonhost1; 10-29-2006 at 04:41 PM.
    Bryon L Harvey
    Soil Relocation Engineer

  18. #18
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    Quote Originally Posted by bryonhost1

    server1-1GB RAM kmem: 23052902:25358192 for 8 [email protected] RAM each
    server2-2GB RAM kmem: 22825164:25107680 for 8 [email protected] RAM each

    which makes no sense at all either. Oy!
    256MB VPS is 256MB VPS.. The slight difference may be due to small descrepencies with the hosts RAM/Swap allocations. To minimize that and ensure good working config files always run vzsplit with the "-s 0" option (assuming you don't want to overcommit).
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  19. #19
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    Uhhh...look at it again, Matt. One server has 1GB RAM..but 256x8=2GB. That's uh..a problem in my book..as total kmem allocations should never be over the Physical RAM size..or at least that's what the manual says. I don't want crashing servers, I don't. Anyways...I'm still confused...but as long as it works..who cares?

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  20. #20
    Join Date
    Aug 2006
    Posts
    42
    We gave up on letting things be calculated for us and just run manual calculations on total available memory vs memory used by vpses on the node. If we upgrade a VPS and suddenly the VPSes on a node are using more than the total available, we balance the node out with a node that has more available memory.

    It's not the prettiest method, but it does keep vpses running and it keeps nodes from dying.

    If you guys want to try a neat memory-starvation test, write a quick program to nicely consume all available memory and run it on 3 or 4 VPSes. Give them a nice burst size. When the memory has all been consumed they'll start frantically writing to swap (you can watch this through vzstat).

    Watch as the swap approaches 0mb, and then kablam! Your node is dead.

    Overallocating memory and burst is a very bad idea.

  21. #21
    Join Date
    Jun 2002
    Location
    Waco, TX
    Posts
    5,291
    The nice thing about the commercial Virtuozzo is that if you have a VPS that say is 20% undersold but then you get 5 people that upgrade, you can within minutes move them to another host node that is more balanced. It is a really nice feature and helps with keeping a balanced host node.

  22. #22
    Join Date
    Dec 2004
    Location
    Butler,TN
    Posts
    2,413
    Hi!
    DavidP-I figured as much..I'll take your word for it. No..I don't have to see that first hand. Still..it does seem like they would have made it a *little* easier. I did ask the folks at Openvz about this..and this is what I got:

    My advise is to keep summary of oomguarpages (which is physpages + used-swap) to be < 1/2 - 2/3 of your RAM. This will allow kernel to use caches effeciently and thus will provide a good performance. This is due to the fact that when caches are not working (e.g. node is overloaded) then performance starts to degrade rapidly

    No...I would be the last one to overload or overallocate a server. The above comment was more of a joke...I do that sometimes. No..I'll just take the examples they provided...and go from there...since there seems to be no basic formula to use to calculate memory allocations.

    Bryon
    Bryon L Harvey
    Soil Relocation Engineer

  23. #23
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    If you're really concerned about giving your customers/servers proper memory allocation you'd do well to purchase Virtuozzo and use SLM. No need to worry about UBC parameters at all... simply assign them a MB value and they are good to go. The type of trouble you are having right now is no doubt one of the reasons SLM was developed.
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  24. #24
    Join Date
    Aug 2006
    Posts
    42
    If you do use SLM it is very very important to be 100% sure you have properly configured it for the node. Failure to do so can result in frequent kernel panics and node lockups.

    SLM is a powerful tool, and takes virtualization one step further, but due to the potential problems it can cause, I would recommend becoming very familiar with it before running it.

  25. #25
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    What do you mean by 'properly configure' ? I don't see how setting SLM=yes, SLMMODE={all|slm}, and then "vzctl set veid --slmmemory limit avg:max --save" could possibly get messed up. If you're referring to the OOM policy (SLMPATTERN), the default is perfectly usable and I don't see how that could cause kernel panics. From the changelogs I've seen there were some SLM bugs fixed, but we have not seen them on our servers.
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  26. #26
    Join Date
    Aug 2006
    Posts
    42
    We've seen SLM cause kernel panic when run with the default setup on our nodes on multiple occasions.

    It's just a word of advice, you don't have to believe me, but we've been through hundreds of nodes, have thousands of customers, and I did see it happen on a regular basis until we made configuration changes within the kernel and SLM itself.

    I generally avoid touching SLM and leave that to the people it's assigned to, so I'm not privvy to the specific changes that were made, but I do know that it can cause problems.

  27. #27
    Join Date
    Nov 2001
    Location
    Atlanta, GA
    Posts
    632
    But there's no much to be changed. There's the ability to turn it on or of, set the MB value. If you wanto really fiddle with it, there's the /etc/vzslm.d/ files, but those are undocumented and not meant to be touched. There's very little you can do to actually affect how SLM works.

    We've had absolutely no panics relating to SLM. There has been some oddness in the 076 kernels, but it's very much stabilized in 078 and up.
    Former owner of A Small Orange
    New owner of <COMING SOON>

  28. #28
    Join Date
    Jan 2004
    Location
    Pennsylvania
    Posts
    939
    Same here Tim... we run a few thousand VPS's with all nodes having SLM enabled and only experienced some accounting inconsistencies early on -- no stability issues. I'll believe my own stats and experience and choose not to ignore the one that is not 'privvy to the specifics'.
    Matt Ayres - togglebox.com
    Linux and Windows Cloud Virtual Datacenters powered by Onapp / Xen
    Instant Setup, Instant Scalability, Full Lifecycle Hosting Solutions

    www.togglebox.com

  29. #29
    Join Date
    Aug 2006
    Posts
    42
    lol, fair enough. my information comes second hand anyway. I just deal with template issues/backups/load balancing/management. I will defer to you guys since you obviously know more on this subject than I do.

  30. #30
    Join Date
    Nov 2005
    Posts
    139

    Try http://www.myvpshost.com

    Try http://www.myvpshost.com
    Under 10 vpses per node. Super fast
    HD input/output on my benchmarks.
    Outstandig support, great connectivity.
    Have been with them for over 1 year and no 1 word of complaints.

    bob

Posting Permissions

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