Page 6 of 9 FirstFirst ... 3456789 LastLast
Results 126 to 150 of 218
  1. #126
    Join Date
    Apr 2009
    Location
    UK
    Posts
    824
    Quote Originally Posted by nibb View Post
    Thanks for sharing. I did go to the website some weeks ago and I could not find too much documentation. I guess that is coming soon.

    Maybe I have to give it a shoot in my lab testing. But things like this are better tested under load actually and I would need to generate some fake loading, users, traffic, etc. Maybe next month...

    I know Cloudlinux points out as marketing to put more customers into servers which you also mentioned. In my case and I think allot of other possible customers its not more customers per server, because honestly hardware has fixed limits and no software can change that, but its about the quality of users. What you do not want is one user using 30% of the server while the rest use almost none.

    I know for a fact that some users must probably hate this things, because they are now tracked on the spot when they are using just too much but for providers I think its great, as you indeed be sure when a client is not designed for shared anymore. Its a hard line currently as nobody has a real standard on what is enough or what is too little. Its a nightmare actually and this is why the marketing is so heavy. More controlled resources means better use of idle power, which of course means more profit per servers. So yes of course this products make sense.

    It's nice to have alternatives as well and what's even more interesting is that all 3 of this systems do the things completely different which is nice as they are truly not just copying each other. Maybe we can run all 3 together
    I have only experienced CL as a user and I find that the server is really crowded, over 1000 sites, and mysql pages are having hard time.

    As a customer with a small commerce, few clients a day, I see my page loading speed like yo-yo throughout the day.

    I have compared with othet hosts without CL and my site was consistently loading fast. I am looking for host right now that has no CL!

    cheers

  2. #127
    Quote Originally Posted by Losvre View Post
    I have only experienced CL as a user and I find that the server is really crowded, over 1000 sites, and mysql pages are having hard time.

    As a customer with a small commerce, few clients a day, I see my page loading speed like yo-yo throughout the day.

    I have compared with othet hosts without CL and my site was consistently loading fast. I am looking for host right now that has no CL!

    cheers
    It isn't really fair to say that cloudlinux is the problem here. We use CloudLinux on all our shared servers, and it does an excellent job of *preventing* the yo-yo effect.

    The hosts you compared to (without cloud linux), might simply not have yet overloaded their servers. For example if your test account was on a new server that isn't yet filled up with 1000 customers...

    Any server with a lot of customers is going to have a yo-yo effect, regardless of what kind of cpu/memory/disk io limiting it has on a per-account basis.

    For example, if Cloudlinux limits are set to allow each customer to use a maximum of 1 cpu core.. and the server has 8 cores.. then it will only take 8 customers using their maximum to crash the entire server.

    Neither cloudlinux nor betterlinux is ever going to "fix" that kind of situation.

    What CloudLinux (and betterlinux) does is reduce the chances of ONE (or a few) customers crashing the entire server.

    But a server can still crash and be "yo yo" up and down with both of these systems. It simply depends on how many customers there are, and what kind of websites they have, etc...

    In my opinion, the odds of having a yoyo server are higher without cloudlinux/betterlinux.

    I hope that makes sense?

    The only way you will ever truly get a server that doesn't suffer from possible "yo yo" effect is if you have a Virtual Private Server (VPS). This way, you are guaranteed a minimum amount of CPU/RAM/DISK IO, regardless of what other customers on the same physical server are doing. As long as you're on a shared hosting server, it is still possible for other customers to use up all the resources of your shared server, no matter if it's using cloudlinux/betterlinux/nothing.
    We are eNom PLATINUM PLUS resellers!
    Sign up today for an eNom.com reseller account with lowest possible pricing.
    * We provide support and service to over 4275 happy eNom domain name and SSL certificate resellers!

  3. #128
    Join Date
    Mar 2012
    Location
    Tampa, FL =)
    Posts
    1,954
    Quote Originally Posted by Losvre View Post
    I have only experienced CL as a user and I find that the server is really crowded, over 1000 sites, and mysql pages are having hard time.

    As a customer with a small commerce, few clients a day, I see my page loading speed like yo-yo throughout the day.

    I have compared with othet hosts without CL and my site was consistently loading fast. I am looking for host right now that has no CL!

    cheers
    The host is probably on CL6 which has I/O issues which causes spikes and then very slow loading times. We run sites on CL no problem and all we have ever had was praise from clients.

  4. #129
    Join Date
    Jun 2005
    Posts
    3,455
    Quote Originally Posted by mrzippy View Post
    It isn't really fair to say that cloudlinux is the problem here. We use CloudLinux on all our shared servers, and it does an excellent job of *preventing* the yo-yo effect.

    The hosts you compared to (without cloud linux), might simply not have yet overloaded their servers. For example if your test account was on a new server that isn't yet filled up with 1000 customers...

    Any server with a lot of customers is going to have a yo-yo effect, regardless of what kind of cpu/memory/disk io limiting it has on a per-account basis.

    For example, if Cloudlinux limits are set to allow each customer to use a maximum of 1 cpu core.. and the server has 8 cores.. then it will only take 8 customers using their maximum to crash the entire server.

    Neither cloudlinux nor betterlinux is ever going to "fix" that kind of situation.

    What CloudLinux (and betterlinux) does is reduce the chances of ONE (or a few) customers crashing the entire server.

    But a server can still crash and be "yo yo" up and down with both of these systems. It simply depends on how many customers there are, and what kind of websites they have, etc...

    In my opinion, the odds of having a yoyo server are higher without cloudlinux/betterlinux.

    I hope that makes sense?

    The only way you will ever truly get a server that doesn't suffer from possible "yo yo" effect is if you have a Virtual Private Server (VPS). This way, you are guaranteed a minimum amount of CPU/RAM/DISK IO, regardless of what other customers on the same physical server are doing. As long as you're on a shared hosting server, it is still possible for other customers to use up all the resources of your shared server, no matter if it's using cloudlinux/betterlinux/nothing.
    You are right, but also he is.

    From a provider perspective CloudLinux is wonderful, but from a client perspective, maybe not.

    Lets assume your are in a hosting server that does not use CL, and makes a good job at monitoring load and keeping things working nice. One customer maybe only use 1% or 5% most of the time, and then sometimes a week, or at some peak hours a day he uses 20%. His website will work and load fast.

    Now lets assume the same provider has Cloud Linux with a limit of 5% CPU, his websites will slow down or show errors at specific times of the day.

    So this is why some people are actually looking for providers without it, why? Because lets be honest, some customers have spikes, and some just like to use more or abuse servers a while at some times of the day. Of course this is not normal, an account paying 5$ canīt possible use half core for itself, otherwise you would go bankrupt.

    The problem with CloudLinux and this systems is that they can be abused, providers will now put more accounts per server, and lower limits even lower for customers. I cannot possible imagine what providers would 1000 accounts in a server, if the do, even with 16 cores, they would need to assign less than 1% CPU for each customer, so like anything else, CloudLinux can be also oversold. And this is not the fault of Cloud Linux or BetterLinux like you said. Its the provider.

    Now from a customer side, your website actually performs worst now, it slows down time at some hours and now you are prompt to attacks as well. If a website has 10% CPU limit with CloudLinux, all I need is reload it a few times to show an error. This means its more easily to actually DOS the website and bring it down, as you only require it to hit a very small amount of CPU or fill the process and it will display an error.

    For a provider this is ok, because this is exactly what it has to do, cage every customer. But from a customer side, this is just terrible.

    Now lets be fair. A provider that oversells, and provides a crappy service, will do the same with or without CloudLinux, nothing is going to change that. So this is not really the product fault.

    Now from a provider side, customers are abusing shared hosting for at least the past 10 years, most people use allot more CPU, RAM and process which they should be allowed for what they pay in shared hosting, so Cloud Linux is actually just exposing the problem the website already had for ever. Without Cloud Linux he may find a host that has some new servers, without websites, eventually the host will need to kick this client out when the server gets filled and he will again be some other months in a new provider until he hits the same problem again. This is why some people keep bouncing from one provider to the other, this is are mainly the abusers that are kicked from everywhere.

    Being caged just puts them the limits right away, which is the correct thing to do from a provider side. Otherwise you need to raise prices to provide more limits or send the customers on his way to a VPS or a plan with more resources, which of course will cost more.

    So both are right here and nobody is wrong either. Customers will look to have more for less, this means abusing servers for as less money as possible, or use as much as they can and providers will look exactly the other way around. Be sure everyone gets a fair piece of resources, someone that pays 10$ should no be able to use more than someone that pays 20$ and this is the idea behind the product.

    Its not for everyone, sure, this includes customers or providers. Some will see it perfectly for their services, while others not. So it really depends what type or services and market you serve.

  5. #130
    Quote Originally Posted by nibb View Post
    You are right, but also he is.

    From a provider perspective CloudLinux is wonderful, but from a client perspective, maybe not.
    its just not reasonable to blame software for a providers business model

    cloudlinux has NOTHING to do with the overall speed of an individual account - nothing at all...

    what IS a fact is that if you compare a like to like shared hosting provider - 2 that operate the exact same way, same model and have the same # of accounts per server, etc - the provider that uses cloudlinux will have more stable servers, will have less outages and will provide a better experience on average for the users on those servers vs the provider not using cloudlinux

    this is the same old argument that has always gone on in the hosting industry. there are good providers and there are not so good providers. there are also providers who provide value and real resources for your dollar, while there are providers selling lots of space and bandwidth for a low cost - and everything in between..

    blaming cloudlinux because a provider overloaded their servers and cant provide all users their allotted resources at all times is not cloudlinux's fault..

    its like saying OpenVZ is not any good because so many providers have abused it and overloaded their servers..

    dont blame the software - cloudlinux and openvz do not make servers slow - overloaded servers make servers slow

    if you arent happy with your cloudlinux provider, find a different one - as no matter how you cut it, you are better off with it vs without it...
    www.cartika.com
    www.clusterlogics.com - You simply cannot run a hosting company without this software. Backups, Disaster Recovery, Big Data, Virtualization. 20 years of building software that solves your problems

  6. #131
    Join Date
    Jun 2005
    Posts
    3,455
    Quote Originally Posted by cartika-andrew View Post
    its just not reasonable to blame software for a providers business model

    cloudlinux has NOTHING to do with the overall speed of an individual account - nothing at all...

    what IS a fact is that if you compare a like to like shared hosting provider - 2 that operate the exact same way, same model and have the same # of accounts per server, etc - the provider that uses cloudlinux will have more stable servers, will have less outages and will provide a better experience on average for the users on those servers vs the provider not using cloudlinux

    this is the same old argument that has always gone on in the hosting industry. there are good providers and there are not so good providers. there are also providers who provide value and real resources for your dollar, while there are providers selling lots of space and bandwidth for a low cost - and everything in between..

    blaming cloudlinux because a provider overloaded their servers and cant provide all users their allotted resources at all times is not cloudlinux's fault..

    its like saying OpenVZ is not any good because so many providers have abused it and overloaded their servers..

    dont blame the software - cloudlinux and openvz do not make servers slow - overloaded servers make servers slow

    if you arent happy with your cloudlinux provider, find a different one - as no matter how you cut it, you are better off with it vs without it...
    That was not what I said at all Andrew. Of course CloudLinux has absolutely nothing to do, I think you read my comments wrong , because I was saying exactly what you just posted. Its the provider that decides how to provide the services and the end quality.

    What I said, is that from a customer perspective having a 10% CPU limit is worst than not having any and being able to use 100%.

    That was my point. From a providers logic, this is not impossible and just crazy to offer, but this is how shared works more or less without Cloud Linux. Is that good? Is that bad? Lets not go into that as its all based on personal opinions and how a providers decides what is best for him or his customers.

    Also, assuming a provider without CloudLinux is more unstable and has more outages is also wrong, it depends again on the provider and not on the software which more or less contradicts your own comments. CloudLinux will not solve the problem of overselling and terrible providers, some will use it wisely as a tool, some as a way to oversell their severs even more, or a better word would be overloaded.

  7. #132
    Quote Originally Posted by nibb View Post
    That was not what I said at all Andrew. Of course CloudLinux has absolutely nothing to do, I think you read my comments wrong , because I was saying exactly what you just posted. Its the provider that decides how to provide the services and the end quality.
    Fair enough and sorry if I misunderstood


    Also, assuming a provider without CloudLinux is more unstable and has more outages is also wrong, it depends again on the provider and not on the software which more or less contradicts your own comments.
    see, this is the problem I have...

    what cloudlinux does is prevent a single user from using up all resources on a server - and many times this is out of the control of the provider or the customers. A customer could upload some bad code and cause an apache loop. A customers script could get compromised and eat up all of the server resources and take down all other users. A customer could upload a bad 3rd party wordpress module and eat up 100% of server resources and take down service for all other customers.. these were all things that all shared providers had to deal with before things like cloudlinux/betterlinux

    these tools provide more stable hosting platforms then is possible without them.. its just a fact with shared hosting..
    www.cartika.com
    www.clusterlogics.com - You simply cannot run a hosting company without this software. Backups, Disaster Recovery, Big Data, Virtualization. 20 years of building software that solves your problems

  8. #133
    Join Date
    Oct 2012
    Location
    Canada
    Posts
    3,064
    I use CloudLinux their support is great and their willing to help at anytime.... it also seems to be more popular so why not I have never heard of "Better Linux" so I cannot really speak on that but if you ask me +1 for CloudLinux.
    TrentaHost INC. || Fully Managed DDoS Protected Services Globally (NA - EU - Asia)
    Reseller Hosting- Pure SSD | Litespeed | Imunify360 | CloudLinux | 24x7 Support | Mailchannels
    Linux & Windows DDoS Protected SSD VPS - cPanel / WHM | DDoS Protection | Let's Encrypt | Pure-SSD
    DDoS Protected Locations : Portland, OR (North America) | Amsterdam, NL (Europe) | Singapore (Asia)

  9. #134
    Join Date
    Dec 2009
    Posts
    139
    Quote Originally Posted by SolidShellSecurity View Post
    The host is probably on CL6 which has I/O issues which causes spikes and then very slow loading times. We run sites on CL no problem and all we have ever had was praise from clients.
    Are you using the last version of CL and its components ?

  10. #135
    Join Date
    Apr 2009
    Location
    UK
    Posts
    824
    Quote Originally Posted by mrzippy View Post
    It isn't really fair to say that cloudlinux is the problem here. We use CloudLinux on all our shared servers, and it does an excellent job of *preventing* the yo-yo effect.

    The hosts you compared to (without cloud linux), might simply not have yet overloaded their servers. For example if your test account was on a new server that isn't yet filled up with 1000 customers...

    Any server with a lot of customers is going to have a yo-yo effect, regardless of what kind of cpu/memory/disk io limiting it has on a per-account basis.

    For example, if Cloudlinux limits are set to allow each customer to use a maximum of 1 cpu core.. and the server has 8 cores.. then it will only take 8 customers using their maximum to crash the entire server.

    Neither cloudlinux nor betterlinux is ever going to "fix" that kind of situation.

    What CloudLinux (and betterlinux) does is reduce the chances of ONE (or a few) customers crashing the entire server.

    But a server can still crash and be "yo yo" up and down with both of these systems. It simply depends on how many customers there are, and what kind of websites they have, etc...

    In my opinion, the odds of having a yoyo server are higher without cloudlinux/betterlinux.

    I hope that makes sense?

    The only way you will ever truly get a server that doesn't suffer from possible "yo yo" effect is if you have a Virtual Private Server (VPS). This way, you are guaranteed a minimum amount of CPU/RAM/DISK IO, regardless of what other customers on the same physical server are doing. As long as you're on a shared hosting server, it is still possible for other customers to use up all the resources of your shared server, no matter if it's using cloudlinux/betterlinux/nothing.
    Hi zippy,

    probably you are right, but being with few hosts for almost a year each I have found that servers with litespeed/CL tend to put 2-3 times more sites on the same server specs.

    I would understand a 30% more clients but not twice! I might have been unlucky or I am very demanding but my sites are really very low traffic MBs per month.

    Thanks

  11. #136
    Join Date
    Oct 2007
    Posts
    4,332
    Quote Originally Posted by Losvre View Post
    Hi zippy,

    probably you are right, but being with few hosts for almost a year each I have found that servers with litespeed/CL tend to put 2-3 times more sites on the same server specs.

    I would understand a 30% more clients but not twice! I might have been unlucky or I am very demanding but my sites are really very low traffic MBs per month.

    Thanks
    You may want to check with the hosting provider on their average number of accounts per server, CloudLinux CPU limit, etc before you sign up.

    There is no reason for a provider to not provide these information to you unless they have something to hide.
    [ James Lee - Cloud & Web Hosting Specialist • 10+ Years WHT Veteran]

    [ Magento Performance Consultation by Magento Master ]

  12. #137
    Quote Originally Posted by Losvre View Post
    Hi zippy,

    probably you are right, but being with few hosts for almost a year each I have found that servers with litespeed/CL tend to put 2-3 times more sites on the same server specs.

    I would understand a 30% more clients but not twice! I might have been unlucky or I am very demanding but my sites are really very low traffic MBs per month.

    Thanks
    Hi Losvre

    I understand your comments, but, things need to be put into perspective here. number of accounts doesnt mean anything - I mean the hardware we are using today is 100 times more powerful then the hardware we were using 5-7 years ago. Before CloudLinux - We used to specifically use lower end servers just to keep the number of clients per server down and reduce the impact of one client having a problem and impacting everyone else..

    What you should really be looking at is how much cpu and ram you get and if the provider can actually provide that to you at all times..

    thats the thing with shared hosting though - is its now also switching to a hard resource model.. a $25 plan with a guaranteed 256MB RAM and 1GB disk and 20GB Bandwidth is probably worth more to most consumers vs a $3 plan with unlimited space and bandwidth and 26MB RAM.

    just understand what you are buying and from whom and you should be fine - dont blame an OS, rather, get a better understanding of your actual hard server resource requirements and look for a provider who caters to that (more so then massive space/bandwidth plans)

    try a provider like medialayer for example - their shared hosting is going to cost a little more - but, you will likely never need to worry about performance
    www.cartika.com
    www.clusterlogics.com - You simply cannot run a hosting company without this software. Backups, Disaster Recovery, Big Data, Virtualization. 20 years of building software that solves your problems

  13. #138
    Join Date
    Mar 2012
    Location
    Tampa, FL =)
    Posts
    1,954
    Quote Originally Posted by bdx33 View Post
    Are you using the last version of CL and its components ?
    Yes. For now. We have had to do a lot of tweaks to keep the I/O from spiking the servers but if they can't fix this soon we will switch over to BL because we are tired of this glitch for the past several months now. We just can't afford a 100.00 spike due to the clients that need us stable 24/7.

  14. #139
    Join Date
    Apr 2009
    Location
    UK
    Posts
    824
    @JLHC, thanks for the tips. I didn't know all these tricks till I started seeing slow loading pages.

    @Andrew, I am looking to a reseller account as I do have several accounts. Saying that, ML hosting account in cluster or what? couldn't figure out from their site.

    Many thanks to everyone

  15. #140

    Days late, but here...

    Quote Originally Posted by bdx33 View Post
    Where are the tests ?
    BetterLinux > Cloudlinux

    Days late, but here are the tests that I promised that show you are crazy if you run CL and expect good I/O limiting, and you are crazy to not be running BetterLinux. Perhaps I'm a little biased so take that into consideration - biased, but not wrong . I'm really sorry. I've had the data for almost 2 weeks and never posted it.

    Anyway, here are the tests that I ran from several posts back with graphs that showed how i/o limiting is pretty broken on CL.

    I/O test details
    ----------------

    WARNING: before running any I/O test, be sure to shutdown crond / anacron or any potential background I/O activity that would interfere with the measurements (Including backups).

    Configuration
    ~~~~~~~~~~~~~

    Create a simple "container" with a 4MB/s I/O limit.

    - CloudLinux:

    [root@cl ~]# cat /etc/container/ve.cfg
    <?xml version="1.0" ?>
    <lveconfig>
    <defaults>
    <cpu limit="0"/>
    <io limit="0"/>
    <ncpu limit="0"/>
    <pmem limit="0"/>
    <mem limit="0"/>
    <nproc limit="0"/>
    </defaults>
    <lve id="500">
    <io limit="4096"/>
    </lve>
    </lveconfig>

    Q1: How do we set I/O limits per device? Ouch! Apparently thats not possible with CL (if we're not wrong), so we're going to skip all the per-device tests. By the way, this is INCREDIBLY bad. This means you are setting the SAME limits for a machine that has both SSDs and rotational disks on the same machine. This method can't possibly feasible in a production environment?!?! Please for the sake of your customers turn this off in CL if you have drives of varying speed!

    Q2: How do you configure iops-based limits (I/O operations per second)? Ouch, not possible again with CL(if we're not wrong), so we're also going to skip the iops-based tests. Again, same spiel as above. This is nuts when you have drives that aren't the same speed!

    - BetterLinux:

    [root@bl ~]# cat /etc/betterlinux/iothrottled.conf.d/user_limits.conf
    ...
    bl_users bw max inc="^/",exc="^/media" 4000000

    [root@bl ~]# cat /etc/betterlinux/iothrottled.conf.d/user_limits.conf
    ...
    uid bl_users 501-1000

    [root@bl ~]# grep "^auto_scale\|seek_balancing " /etc/betterlinux/iothrottled.conf.d/scale.conf
    auto_scale 0 # [Default = 1]
    seek_balancing 0 # [Default = 1]

    NOTE: auto_scale and seek_balancing have been disabled to better measure the "fixed" limit enforcements of drive. We do it this way for testing again because CL daoesn't support autoscaling of drive resources depending on demand, and CL also doesn't support any seek overload by individual programs or users (so we are "hobbling" BetterLinux to function in only fixed-limits mode for these tests to compare against what CL offers today).

    Generic I/O tests
    ~~~~~~~~~~~~~~~~~

    Script to perform the complex I/O tests and collect system statistics:

    ---
    #!/bin/sh

    do_test() {
    iotype=$1
    tasks=$2
    bs=$3
    size=$4
    ioengine=$5

    dstat --cpu --sys --load --proc --io --disk --disk-util -D vda --output dstat-$tasks-$iotype-$bs-$size.log >/dev/null 2>&1 &
    pid=$!
    fio --rw=$iotype --bs=$bs --size=$size --numjobs=$tasks --ioengine=$ioengine --name=seeker --output=fio-$tasks-$iotype-$bs-$size.log
    kill $pid
    }

    for iotype in read randread; do
    for ioengine in sync libaio mmap splice; do
    for bs in 32K 32M; do
    for tasks in 1 2 4 8; do
    do_test $iotype $tasks $bs 128M $ioengine
    done
    done
    done
    done
    ---

    This script generates different I/O load and collects many different types of system statistics that can be used with any graphing software of your choice: both dstat and fio give lots of relevant data. Check out the load average for example, the disk utilization percentage, page faults, etc. We basically use these tests (among many others) and used these available metrics to compare the results.

    Specific metrics used to generate our graphs:
    - Aggregate I/O bandwidth: plot read/write columns from the output of dstat
    - Disk utilization %: "util" column from the output of dstat
    - Cache locality: majf + minf values, taken from the output of fio minf = minor page faults: page present in memory but not mapped by the MMU majf = major page faults: page not present in memory (a major fault requires an I/O operation)
    - Page cache rewrite latency: this is a specific test (use rewrite.c below and measure the time to complete each write/re-write: the first write should be limited - however, the next re-write(s) that takes place only in page cache should not, because they're not going to generate any I/O activity). This is a *HUGE* problem with CL's i/o "solution". How can you be justified in limiting writes to cache that are not even the first write? I'm totally on board with limiting the first write, but each additional one??? Come on - thats just broken.

    ---
    /*
    * rewrite.c
    */
    #include <errno.h>
    #include <stdio.h>
    #include <stdlib.h>
    #include <unistd.h>
    #include <time.h>
    #include <fcntl.h>
    #include <sys/types.h>

    static char buf[32 * 1024 * 1024];

    int main(int argc, char **argv)
    {
    int fd, i;

    fd = open("junk", O_WRONLY | O_CREAT, 0600);
    if (fd < 0) {
    perror("open");
    exit(1);
    }
    for (i = 0; i < 10; i++) {
    time_t start, end;

    lseek(fd, 0, SEEK_SET);
    start = time(NULL);
    if (write(fd, buf, sizeof(buf)) < 0) {
    perror("write");
    exit(1);
    }
    end = time(NULL);

    printf("%d: %zus\n", i, end - start);
    fflush(stdout);
    }
    unlink("junk");
    return 0;
    }
    ---

    Again, sorry for the delay in the stats. Please feel free to ask questions, find bugs in either BetterLinux or CL, or to generally improve some of the tests below to expose problems in either solution.

    Thanks,
    Matt Heaton / CEO BetterLinux.com
    Last edited by mheaton; 12-06-2012 at 01:55 AM.

  16. #141
    Join Date
    Mar 2012
    Location
    Tampa, FL =)
    Posts
    1,954
    Quote Originally Posted by mheaton View Post
    BetterLinux > Cloudlinux

    Days late, but here are the tests that I promised that show you are crazy if you run CL and expect good I/O limiting, and you are crazy to not be running BetterLinux. Perhaps I'm a little biased so take that into consideration - biased, but not wrong . I'm really sorry. I've had the data for almost 2 weeks and never posted it.

    Anyway, here are the tests that I ran from several posts back with graphs that showed how i/o limiting is pretty broken on CL.

    I/O test details
    ----------------

    WARNING: before running any I/O test, be sure to shutdown crond / anacron or any potential background I/O activity that would interfere with the measurements (Including backups).

    Configuration
    ~~~~~~~~~~~~~

    Create a simple "container" with a 4MB/s I/O limit.

    - CloudLinux:

    [root@cl ~]# cat /etc/container/ve.cfg
    <?xml version="1.0" ?>
    <lveconfig>
    <defaults>
    <cpu limit="0"/>
    <io limit="0"/>
    <ncpu limit="0"/>
    <pmem limit="0"/>
    <mem limit="0"/>
    <nproc limit="0"/>
    </defaults>
    <lve id="500">
    <io limit="4096"/>
    </lve>
    </lveconfig>

    Q1: How do we set I/O limits per device? Ouch! Apparently thats not possible with CL (if we're not wrong), so we're going to skip all the per-device tests. By the way, this is INCREDIBLY bad. This means you are setting the SAME limits for a machine that has both SSDs and rotational disks on the same machine. This method can't possibly feasible in a production environment?!?! Please for the sake of your customers turn this off in CL if you have drives of varying speed!

    Q2: How do you configure iops-based limits (I/O operations per second)? Ouch, not possible again with CL(if we're not wrong), so we're also going to skip the iops-based tests. Again, same spiel as above. This is nuts when you have drives that aren't the same speed!

    - BetterLinux:

    [root@bl ~]# cat /etc/betterlinux/iothrottled.conf.d/user_limits.conf
    ...
    bl_users bw max inc="^/",exc="^/media" 4000000

    [root@bl ~]# cat /etc/betterlinux/iothrottled.conf.d/user_limits.conf
    ...
    uid bl_users 501-1000

    [root@bl ~]# grep "^auto_scale\|seek_balancing " /etc/betterlinux/iothrottled.conf.d/scale.conf
    auto_scale 0 # [Default = 1]
    seek_balancing 0 # [Default = 1]

    NOTE: auto_scale and seek_balancing have been disabled to better measure the "fixed" limit enforcements of drive. We do it this way for testing again because CL daoesn't support autoscaling of drive resources depending on demand, and CL also doesn't support any seek overload by individual programs or users (so we are "hobbling" BetterLinux to function in only fixed-limits mode for these tests to compare against what CL offers today).

    Generic I/O tests
    ~~~~~~~~~~~~~~~~~

    Script to perform the complex I/O tests and collect system statistics:

    ---
    #!/bin/sh

    do_test() {
    iotype=$1
    tasks=$2
    bs=$3
    size=$4
    ioengine=$5

    dstat --cpu --sys --load --proc --io --disk --disk-util -D vda --output dstat-$tasks-$iotype-$bs-$size.log >/dev/null 2>&1 &
    pid=$!
    fio --rw=$iotype --bs=$bs --size=$size --numjobs=$tasks --ioengine=$ioengine --name=seeker --output=fio-$tasks-$iotype-$bs-$size.log
    kill $pid
    }

    for iotype in read randread; do
    for ioengine in sync libaio mmap splice; do
    for bs in 32K 32M; do
    for tasks in 1 2 4 8; do
    do_test $iotype $tasks $bs 128M $ioengine
    done
    done
    done
    done
    ---

    This script generates different I/O load and collects many different types of system statistics that can be used with any graphing software of your choice: both dstat and fio give lots of relevant data. Check out the load average for example, the disk utilization percentage, page faults, etc. We basically use these tests (among many others) and used these available metrics to compare the results.

    Specific metrics used to generate our graphs:
    - Aggregate I/O bandwidth: plot read/write columns from the output of dstat
    - Disk utilization %: "util" column from the output of dstat
    - Cache locality: majf + minf values, taken from the output of fio minf = minor page faults: page present in memory but not mapped by the MMU majf = major page faults: page not present in memory (a major fault requires an I/O operation)
    - Page cache rewrite latency: this is a specific test (use rewrite.c below and measure the time to complete each write/re-write: the first write should be limited - however, the next re-write(s) that takes place only in page cache should not, because they're not going to generate any I/O activity). This is a *HUGE* problem with CL's i/o "solution". How can you be justified in limiting writes to cache that are not even the first write? I'm totally on board with limiting the first write, but each additional one??? Come on - thats just broken.

    ---
    /*
    * rewrite.c
    */
    #include <errno.h>
    #include <stdio.h>
    #include <stdlib.h>
    #include <unistd.h>
    #include <time.h>
    #include <fcntl.h>
    #include <sys/types.h>

    static char buf[32 * 1024 * 1024];

    int main(int argc, char **argv)
    {
    int fd, i;

    fd = open("junk", O_WRONLY | O_CREAT, 0600);
    if (fd < 0) {
    perror("open");
    exit(1);
    }
    for (i = 0; i < 10; i++) {
    time_t start, end;

    lseek(fd, 0, SEEK_SET);
    start = time(NULL);
    if (write(fd, buf, sizeof(buf)) < 0) {
    perror("write");
    exit(1);
    }
    end = time(NULL);

    printf("%d: %zus\n", i, end - start);
    fflush(stdout);
    }
    unlink("junk");
    return 0;
    }
    ---

    Again, sorry for the delay in the stats. Please feel free to ask questions, find bugs in either BetterLinux or CL, or to generally improve some of the tests below to expose problems in either solution.

    Thanks,
    Matt Heaton / CEO BetterLinux.com
    Thanks for the info Matt Heaton. We are planning to start transforming all our CL servers to BL for better performance.

    Really looking forward to working with you Matt.

  17. #142
    Join Date
    Feb 2002
    Location
    Australia
    Posts
    24,027
    Wow, just spent the last hour reading through this thread. Takes me back to the old days of WHT!!
    • WLVPN.com • NetProtect owned White Label VPN provider •
    • Increase your hosting profits by adding VPN to your product line up •

  18. #143
    Quote Originally Posted by Aussie Bob View Post
    Wow, just spent the last hour reading through this thread. Takes me back to the old days of WHT!!
    Yup. Not to mention all the old folks who pop by!
    We are eNom PLATINUM PLUS resellers!
    Sign up today for an eNom.com reseller account with lowest possible pricing.
    * We provide support and service to over 4275 happy eNom domain name and SSL certificate resellers!

  19. #144
    Join Date
    Jun 2001
    Location
    Princeton
    Posts
    1,029
    Just in case, if you still have any IO issues with CL6 kernel -- try this one, memory reclamation was redone, and that solves quite a few issues:
    http://www.cloudlinux.com/blog/clnew...ages-fixes.php
    Igor Seletskiy
    CEO @ Cloud Linux Inc
    http://www.cloudlinux.com
    CloudLinux -- The OS that can make your Shared Hosting stable

  20. #145
    Join Date
    Feb 2002
    Location
    Australia
    Posts
    24,027
    I love that we hosts have access to CL and BL but it makes me wonder why wasn't resource isolation software like this baked into cPanel long ago?

    I think CL is the best thing since sliced bread. Would have loved this years back, when the servers weren't so powerful and whatnot. Would have saved a lot of drama I reckon.
    • WLVPN.com • NetProtect owned White Label VPN provider •
    • Increase your hosting profits by adding VPN to your product line up •

  21. #146
    Join Date
    Apr 2009
    Location
    UK
    Posts
    824
    Hi guys,

    Since the heat is getting up...may I ask you what do you think about the hive server optimisation? Not sure if can post links but everybody does

    http://www.1h.com/products/hive

  22. #147
    Join Date
    Oct 2004
    Posts
    632
    Quote Originally Posted by Steven View Post
    Igor,
    There's lots of name calling,etc right now.
    But.. to be fair, it really seems like you guys didn't care about actual problems until betterlinux started pushing hard.. (think: cloudlinux 6 IO issue).
    It was documented a few times by me on this board, I had customers who contacted your support and your support told them to optimize mysql.. yet we removed cloudlinux completely and to this day many months later are more stable than when cloudlinux 6 was on them.

    Just saying.. your company was pretty lazy up until recently.. You put all of your focus into cagefs and forgot about the rest of your software. There was a point where I was getting ~5 to 10 tickets a week with the same complaint 'my cloudlinux 6 server is spiking to loads of 50.0 to 100.0', and they would want their server optimized... I am glad betterlinux is giving grief... otherwise I am pretty sure that IO issue would still exist. I still don't think IO is up to par with where it should be.. but whatever. We shall see how it all pans out.

    Both products do what they are meant to do, however based on experience with CL and how they deal with service impacting issues (Cloudlinux 6 IO issue).. BL is winning.
    Igor is however very helpful via email when I have a question.
    I have internal testing going on with both of these products, I hope to have some results to show after the holidays when I am not so busy.
    have to agree with these and this is why we don;t really use CL (nor BL) so far...

    we don;t want to deal with any sudden load bust on system that is designed to resolve sudden load burst...

    would love to see test from Steven and see how BL works with cpanel servers

  23. #148
    Join Date
    Oct 2003
    Posts
    9,264
    Hopefully, neither will ever be on my systems.
    What this industry needs is a consortium heavily involved in open sourcing creations for each other's improvement -- rather than yet-another-repo of pay2play software.

    With that said I've heard great things about both products..

  24. #149

    Mostly agree...

    Quote Originally Posted by David View Post
    Hopefully, neither will ever be on my systems.
    What this industry needs is a consortium heavily involved in open sourcing creations for each other's improvement -- rather than yet-another-repo of pay2play software.

    With that said I've heard great things about both products..
    Actually, I pretty much agree. I know this seems weird coming from someone selling a commercial product. However, let me explain - BetterLinux is a complete labor of love for me. I will almost certainly not break even on the time and money spent to create the product. So why create it then? I spent so many years at Bluehost with such a high frustration level with cpu/io/bandwidth/process overages that I just felt compelled to create a product to solve these issues to the best of my ability. I have spent millions of dollars in developer salaries creating this product. I sell the product for $9 a month. Not too lucrative However, I am enjoying myself much more than when I ran Bluehost!

    The second someone comes along and has a free product that will do what our product does I will smile real big pack my bags, and go work on another project with not an ounce of remorse, and with a whole lot of satisfaction that the problem is finally solved - no matter who it was that accomplished it. I just want to see the problems gone once and for all!

    Thanks,
    << Signature to be setup in Profile >>
    Last edited by Alex; 12-20-2012 at 10:15 PM.

  25. #150
    Quote Originally Posted by mheaton View Post
    Actually, I pretty much agree. I know this seems weird coming from someone selling a commercial product. However, let me explain - BetterLinux is a complete labor of love for me. I will almost certainly not break even on the time and money spent to create the product. So why create it then? I spent so many years at Bluehost with such a high frustration level with cpu/io/bandwidth/process overages that I just felt compelled to create a product to solve these issues to the best of my ability. I have spent millions of dollars in developer salaries creating this product. I sell the product for $9 a month. Not too lucrative However, I am enjoying myself much more than when I ran Bluehost!

    The second someone comes along and has a free product that will do what our product does I will smile real big pack my bags, and go work on another project with not an ounce of remorse, and with a whole lot of satisfaction that the problem is finally solved - no matter who it was that accomplished it. I just want to see the problems gone once and for all!

    Thanks,
    << signature to be setup in profile >>
    Matt, it is interesting to see your position on this. I'm guessing you made decent $$ selling your previous ventures, and this now allows you to spend time/effort/$ on the things you enjoy.

    That's nice to see. Hopefully you are at least coding from a beach somewhere.

    It's a catch-22, however, because as a consumer reading what you wrote... the first thing that pops into my head is that you really have no "real" reason (ie: financial incentive, the need to make $ to pay the bills) to create a super-amazing product that will do exactly what it says. The only motivation is your personal drive and determination. (Which, given your history, seem to be quite high and sincere.)

    I guess, what I'm trying to say is that, generally, the company or person who *needs* to create a high-quality product to survive is more likely to do so.. compared to the company or person that is just doing it for fun or as an interesting challenge.

    Although of course, there are plenty of open-source (free) examples that prove this statement incorrect.

    So it's important that when people read your post.. they understand you are not really like a normal person. You are a machine who is one-track focused and determined to produce a high-quality product, even though "BetterLinux is a complete labor of love" and not necessarily a real business venture that you expect to get rich from. You put *everything* into your efforts, because it's just how you roll. You aren't happy with a "it's ok" solution. You strive for "it's the best there is!" perfection and don't rest until that happens.

    I think once people understand how you work, then it will help to relieve the "he doesn't really need to make a good product since he's already rich" auto-response to your post.

    So that being said, I'm looking forward to seeing the cpanel UI for BetterLinux. As I've said before, the technical side of things is just one side. Just as important (and possibly more so, from a business perspective) is the customer-facing side of things, the UI, the "simpleness" of the system, etc... The vast majority of server owners who need/want Cloudlinux or BetterLinux will never use a command line or manually edit a config file. Sad. But such is the way the hosting industry works these days.

    Last edited by Alex; 12-20-2012 at 10:15 PM. Reason: Quoted Edited Post
    We are eNom PLATINUM PLUS resellers!
    Sign up today for an eNom.com reseller account with lowest possible pricing.
    * We provide support and service to over 4275 happy eNom domain name and SSL certificate resellers!

Page 6 of 9 FirstFirst ... 3456789 LastLast

Similar Threads

  1. Replies: 0
    Last Post: 12-04-2011, 06:56 PM
  2. Replies: 0
    Last Post: 11-06-2011, 07:28 PM
  3. Replies: 0
    Last Post: 09-18-2011, 05:10 PM
  4. Replies: 0
    Last Post: 07-15-2011, 11:19 PM
  5. Replies: 0
    Last Post: 06-27-2011, 01:20 AM

Posting Permissions

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