Page 4 of 17 FirstFirst 123456714 ... LastLast
Results 76 to 100 of 416
  1. #76
    Join Date
    Dec 2000
    Location
    Atlanta
    Posts
    576
    Does anyone know how many snapshots you can keep on hand? Does this part work well?
      0 Not allowed!

  2. #77
    Join Date
    Jun 2006
    Location
    United Kingdom
    Posts
    152
    Is TGL basically like a VPS sort of device, if so, has anyone tried UnixBench'ing a TGL environment to see what sort of performance it provides?
    Chris Imrie
    Freelance Consultant
      0 Not allowed!

  3. #78
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by drewnick
    Does anyone know how many snapshots you can keep on hand? Does this part work well?
    LT will have to answer this one - there is no technical limit to this - it is just the amount of storage involved and how they account/charge for it.
      0 Not allowed!

  4. #79
    Join Date
    Dec 2004
    Location
    New York, NY
    Posts
    10,710
    Quote Originally Posted by RazorBlue - Chris
    Is TGL basically like a VPS sort of device, if so, has anyone tried UnixBench'ing a TGL environment to see what sort of performance it provides?
    Yes, they actually use Xen so it is very much like a VPS.

    I got about ~ 107 (IIRC) on unixbench with the lowest package.
    MediaLayer, LLC - www.medialayer.com Learn how we can make your website load faster, translating to better conversion rates for your business!
    The pioneers of optimized web hosting, featuring LiteSpeed Web Server & SSD Storage - Celebrating 10 Years in Business
      0 Not allowed!

  5. #80
    Join Date
    Feb 2004
    Location
    Atlanta, GA
    Posts
    5,662
    Quote Originally Posted by layer0
    Yes, they actually use Xen so it is very much like a VPS.

    I got about ~ 107 (IIRC) on unixbench with the lowest package.

    thats also all dependent upon what hardware is behind the node....

    You could have a

    Semprron 2800, 1gb, 1 x 500GB SATA as each node
    or
    Dual Quad-Core Xeon, 16GB RAM, 6 x 146GB SAS10k per node


    the benchmarks still go back to the core hardware
      0 Not allowed!

  6. #81
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by WireSix
    thats also all dependent upon what hardware is behind the node....

    You could have a

    Semprron 2800, 1gb, 1 x 500GB SATA as each node
    or
    Dual Quad-Core Xeon, 16GB RAM, 6 x 146GB SAS10k per node


    the benchmarks still go back to the core hardware
    Guys, forgive me the software guy for the stupid question, but does it still matter? You get a grid server, you set it up with the software you need. It is not longer tied to a given physical server, and even to a given grid. You can move it around, and you can migrate it from one grid to another. So the only thing the benchmark tells you is how fast it runs on this grid today. If the performance of a given grid is good enough, you keep it there. If not, you move it to another grid...
      0 Not allowed!

  7. #82
    Join Date
    Dec 2004
    Location
    New York, NY
    Posts
    10,710
    Quote Originally Posted by WireSix
    thats also all dependent upon what hardware is behind the node....

    You could have a

    Semprron 2800, 1gb, 1 x 500GB SATA as each node
    or
    Dual Quad-Core Xeon, 16GB RAM, 6 x 146GB SAS10k per node


    the benchmarks still go back to the core hardware
    I am aware...the current grid is based on Semprons....
    MediaLayer, LLC - www.medialayer.com Learn how we can make your website load faster, translating to better conversion rates for your business!
    The pioneers of optimized web hosting, featuring LiteSpeed Web Server & SSD Storage - Celebrating 10 Years in Business
      0 Not allowed!

  8. #83
    Join Date
    Feb 2004
    Location
    Atlanta, GA
    Posts
    5,662
    Quote Originally Posted by VladMiloushev
    Guys, forgive me the software guy for the stupid question, but does it still matter? You get a grid server, you set it up with the software you need. It is not longer tied to a given physical server, and even to a given grid. You can move it around, and you can migrate it from one grid to another. So the only thing the benchmark tells you is how fast it runs on this grid today. If the performance of a given grid is good enough, you keep it there. If not, you move it to another grid...


    Yeah I fully understand. What I was trying to stress is that people are not "benchmarking" Applogic but rather the strength of the node behind their single instance.
      0 Not allowed!

  9. #84
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by WireSix
    Yeah I fully understand. What I was trying to stress is that people are not "benchmarking" Applogic but rather the strength of the node behind their single instance.
    I see your point. Which, BTW, brings another issue to the picture: we need a hardware-independent way to express CPU power. When configuring an app in AppLogic, I can set RAM, disk space and bandwidth in a way that is not dependent on the particular grid it is running on. But I don't have a convenient way to do this when it comes to CPU...

    Unfortunately, over the years everyone has gotten addicted to measuring CPU performance in GHz. Today, with the clock frequency measure broken, we are left with nothing at all to help us compare CPUs objectively.
      0 Not allowed!

  10. #85
    I send a ticket asking some things to layeredtech and still waiting their response, everyone it's doing the same and for this they are delaying?
    Honesting.es honest european provider.
      0 Not allowed!

  11. #86
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by pueblosnet
    I send a ticket asking some things to layeredtech and still waiting their response, everyone it's doing the same and for this they are delaying?
    I don't know what the problem might be, but I spoke with Jeremy earlier today and they are working on getting more people trained on supporting the grid.
      0 Not allowed!

  12. #87
    Join Date
    Jun 2004
    Location
    Bay Area
    Posts
    1,320
    Quote Originally Posted by RazorBlue - Chris
    Is TGL basically like a VPS sort of device, if so, has anyone tried UnixBench'ing a TGL environment to see what sort of performance it provides?
    Check the UnixBench thread
      0 Not allowed!

  13. #88
    Join Date
    Jun 2006
    Location
    United Kingdom
    Posts
    152
    Okay i'll do that shortly, did fire some Q's over to TGL (LT) but have no reply, mainly due to me sending the Q's late USA time.
    Chris Imrie
    Freelance Consultant
      0 Not allowed!

  14. #89
    Join Date
    Dec 2000
    Location
    Atlanta
    Posts
    576
    I called this morning. They are still very much learning themselves about this product - as I would have expected. Still a lot of variables in the air.
      0 Not allowed!

  15. #90
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by Xandrios
    Check the UnixBench thread
    The problem for a vendor like 3Tera is that as soon as we decide to standardize on a given benchmark, it makes all others (and their supporters) feel bad, and they let you know in no uncertain terms. So, until the community self-organizes around a single CPU performance benchmark, vendors will have to stick with GHz and percent of CPU.
      0 Not allowed!

  16. #91
    Join Date
    Aug 2005
    Location
    Canada
    Posts
    862
    The Benchmark post:
    http://www.webhostingtalk.com/showpo...&postcount=650

    This is very interesting.

    Good overall score for the small RAM.

    Low process creation can be coming from the overhead of virtualization and process control/monitoring.
    And shell script can be affected by slow process creation.

    And the File copy results is amusing.
    It shows higher figure for smaller buffer size.
    Maybe due to the way file system is implemented in the AppLogic.


    I'm looking for a comparative result from plain Xen with similar RAM/CPU, now.
      0 Not allowed!

  17. #92
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    This is the reply to a posting in the MT thread - the poster suggested we move to the GridLogic thread where this belongs anyway.

    Quote Originally Posted by extras
    Thank you, again, for the information.

    Are you willing to provide us more information about AppLogic OS?

    I would like to know about the virtualization technology, for example.
    There are several known virtualization systems, like UML, XEN, VMware, Virtuozo, ....
    Are you using one of them (or something derived from them)?
    I think it requires a lot of work just to make that part.
    AppLogic uses a modified version of Xen for server virtualization. The storage and network virtualization engines are proprietary and developed in 3Tera.

    Quote Originally Posted by extras
    Then, there is a question about the file system.

    I've read that you have an advisor who is an expert of file systems.
    And again, there are several known file system that would allow us to use inexpensive IDE/SATA disks found directly on the motherborad (or on the storage server).
    But these system may experience problem with replication and increased number of node, as the grid grows.

    The good news is Gigabit NIC is getting cheaper and cheaper, and maybe you can use separate NIC for file system, or do bonding and make the net redundant.

    Still, I'm not so sure if the HDD on each physical node can be utilized well for fully redundant way to be accessed from multiple node, simultaneously, without sacrifycing the performance.
    AppLogic does not use a common file system (none of them scales well) Instead, it takes over the HDD on each physical node and combines those disks into a shared pool of storage accessible from every node on the system. On this system you can create virtual volumes. Each volume is completely distributed. It consists of one or more mirrors, each located on a different server. As you can see from the postings here, performance is fine.

    Quote Originally Posted by extras
    Finally, the question fo redundancy (or High availability).

    You said you can put cluster with load-balancer withing the Grid.
    But I thought the Grid is already offering both redundancy and performance.

    Does the grid sit on Physically FULLY redundant system?
    If so, what's the merit of constructing load-balanced cluster withing the grid?
    The main reason to have a load balancer is not for high availability - it is, ahem, to balance the load. The load balancer allows you to aggregate the performance of several web servers and make them appear as a single server on the net. For example, in our cPanel cluster, we use 8 load-balanced Apache servers to serve customer account sites. The result is that the system is very difficult to overload - you will need to saturate all 8 Apaches before the cluster becomes slow to respond.

    Load balancing and performance clustering are always somewhat application-specific: the way you cluster JBoss servers is different from the way you cluster MySQL, is different from the way you cluster Oracle, is different from the way you cluster Apache, etc.

    To try and implement all this at the core grid services would be a failure, since it would force the grid to be dependent on the specific set of database engines, app servers, web servers, etc. Instead, AppLogic deals with the things that can be made common (storage, network, scheduling, etc.) in the core services, and makes it easy to assemble application-specific clusters from a library of virtual appliances.
    [/quote]
      0 Not allowed!

  18. #93
    Join Date
    Aug 2005
    Location
    Canada
    Posts
    862
      0 Not allowed!

  19. #94
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by extras
    The Benchmark post:
    http://www.webhostingtalk.com/showpo...&postcount=650

    This is very interesting.

    Good overall score for the small RAM.

    Low process creation can be coming from the overhead of virtualization and process control/monitoring.
    And shell script can be affected by slow process creation.

    And the File copy results is amusing.
    It shows higher figure for smaller buffer size.
    Maybe due to the way file system is implemented in the AppLogic.


    I'm looking for a comparative result from plain Xen with similar RAM/CPU, now.
    I think that the low process creation comes from the fact that the system commits memory pages with a "copy on write" algorithm. This results in more activity switching between virtual machines during process creation.

    I would agree on the shell script results interpretation above.

    The file result is a useful side effect of the way AppLogic handles storage. My guess is that the small files end up being cached in memory on the servers where their mirrors reside, so they end up being served from cache. Large files hit disk.
      0 Not allowed!

  20. #95
    Join Date
    Aug 2005
    Location
    Canada
    Posts
    862
    Thank you Vlad, for your "high availability".
    Quote Originally Posted by VladMiloushev
    AppLogic uses a modified version of Xen for server virtualization. The storage and network virtualization engines are proprietary and developed in 3Tera.
    That's good to know.

    Actually, I was dreaming of Junk yard grid data center, the other night.
    What is the best way to put bunch of junk PCs into a grid, cluster, whatever
    to obtain scalability and high availability?

    I was thinking of Xen (or VMwareserver) to create homogeneous VM to abosorb hardware issues, and also allow the VM to be shooted down from centralized server(s) and allow each physical server to be splitted into multiple separated role/servers.

    AppLogic does not use a common file system (none of them scales well) Instead, it takes over the HDD on each physical node and combines those disks into a shared pool of storage accessible from every node on the system. On this system you can create virtual volumes. Each volume is completely distributed. It consists of one or more mirrors, each located on a different server. As you can see from the postings here, performance is fine.
    That's what I was thinking.

    And I was looking for existing file system like that, already available when I was thinking of the junk grid, too.
    So, I was reading about gfs, xfs, coda, and so on.
    http://www.linux-ha.org/RelatedTechnologies/Filesystems


    The main reason to have a load balancer is not for high availability - it is, ahem, to balance the load.
    Yes, I know that. My question was more like "Why do we need load balancer if AppLogic is already offering scalability?"
    I mean, if you can make one webserver bigger, you don't have to loadbalance many of them.

    Load balancing and performance clustering are always somewhat application-specific: the way you cluster JBoss servers is different from the way you cluster MySQL, is different from the way you cluster Oracle, is different from the way you cluster Apache, etc.

    To try and implement all this at the core grid services would be a failure, since it would force the grid to be dependent on the specific set of database engines, app servers, web servers, etc. Instead, AppLogic deals with the things that can be made common (storage, network, scheduling, etc.) in the core services, and makes it easy to assemble application-specific clusters from a library of virtual appliances.
    I also saw the demo-flash, and I think I have a better image, now.

    If I resume AppLogic as: Custormized Xen + distributed redundant file system + VM-appliance manager, would you like to add/modify anything?

    Another question I have is somewhat related to the loadbalancing.
    Can we set 300% MAX CPU, for example?
    Can AppLogic "bond" multiple physical servers to create huge VM image?

    And the most important question of all.
    Do you have full redundancy at the hardware level?
    I mean, do you use multiple switches, NICS, for example?
      0 Not allowed!

  21. #96
    Join Date
    Oct 2006
    Posts
    68
    Yeah they look pretty busy, I'm still waiting for a response since friday afternoon.
      0 Not allowed!

  22. #97
    Join Date
    Dec 2000
    Location
    Atlanta
    Posts
    576
    By the time you read this, http://www.thegridlayer.com/ may be back up, but it has been down all morning. :-(

    Ironic, anyone? Happy Holidays!
      0 Not allowed!

  23. #98
    Join Date
    Nov 2006
    Posts
    64
    must not be too redundant.. I can't get there either.

    I figured it wasn't as impressive as it sounds, and the more I read about it the less it is.
      0 Not allowed!

  24. #99
    Join Date
    Aug 2005
    Location
    Canada
    Posts
    862
    DNS problem?
    Code:
    --09:38:22--  http://www.thegridlayer.com/
               => `-'
    Resolving www.thegridlayer.com... failed: Host not found.
    command failed with code 1
      0 Not allowed!

  25. #100
    Join Date
    Nov 2006
    Location
    Southern California
    Posts
    173
    Quote Originally Posted by drewnick
    By the time you read this, http://www.thegridlayer.com/ may be back up, but it has been down all morning. :-(

    Ironic, anyone? Happy Holidays!
    Guys, www.thegridlayer.com (the web site) is not running on a grid. I love dealing with naysayers :-)
      0 Not allowed!

Page 4 of 17 FirstFirst 123456714 ... LastLast

Posting Permissions

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