New server infrastructure

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • mIDO
    Junior Member
    • Mar 2004
    • 28

    New server infrastructure

    Hi!
    Today we've received good news about the new server infrastructure. Andrew works to give us the best and this is the reason that makes Dathorn great. You can't imagine what kind of nightmare is to work with hosting comanies in Spain for example!! The don't take care of servers, hardware nor software ;(

    But I have a few questions about the new infrastructure:
    - The price will be the same for us?
    - Will we have any upgrade? Disk space, bandwidth?
    - Do you have estimation times for servers? Can we see a schedule to know when will we be moved?
    - New servers are bigger, and I suppose there will be more customers per server. Will it affect to performance or with CloudLinux this is not a problem?
    - How long can take the migration? Downtime for us?

    Regards,
    Miguel
  • ph23man
    Junior Member
    • Mar 2004
    • 13

    #2
    Any information regarding the datacenter and network connectivity?

    Comment

    • AndrewT
      Administrator
      • Mar 2004
      • 3653

      #3
      mIDO, I'm just going to run through your questions in order:

      - Your pricing will remain the same.

      - We do plan on offering clients a special to increase their resource limits, especially disk space, once all servers have been migrated.

      - We hope to complete all of the migrations within 3-4 months. The first server is already being migrated on Friday. Beyond that we do not have a specific schedule. As a general rule we'll probably just go down the line oldest to newest with cpanel59 being the oldest and cpanel74 being the newest.

      - There will indeed be more clients on each server. We would be crazy not to do this given the power that each of these servers will have. Performance will not be impacted and will just get even better if anything. The exception would be those abusing system resources in the first place, as now they would get limited by CloudLinux. We don't overload / fill our servers to the brim like many other hosts. An overloaded server makes clients unhappy and causes more work for us, nobody wins there. To start we're just migrating servers 1:1 to see how it goes and we'll work from there.

      - There is zero downtime involved with the migrations and as long as you are using any of our gzo.com name servers or are using custom name servers with our new DNS IP addresses, the migration will be completely seamless with no changes required by you. Right now we're planning on each server migration taking approximately 24 hours. We'll be able to better judge this once we've done a few.

      ph23man, we'll be located in CoreXchange's new facility. Initially we'll be using a combination of IP Transit (a blend of Level 3, Global Crossing, and TATA) and the data center's blend (ATT, AboveNet, Verio, Level 3, and Global Crossing) in addition to peering. This may change as clients are migrated over and we can better evaluate real world routes.

      Comment

      • mIDO
        Junior Member
        • Mar 2004
        • 28

        #4
        In the migration progress I think we should stablish a "no-changes" time where we should block databases and avoid to do any changes on our websites, is that correct?
        Thanks for your detailed information, as always, so transparent

        Comment

        • AndrewT
          Administrator
          • Mar 2004
          • 3653

          #5
          It certainly would be best if no changes were made during the migration period as they may or may not make it to the new server. Once an individual cPanel account is migrated to the new server the DNS cluster is automatically updated with the new IP address. So overall it should only be a couple of hours between the time that a domain is transferred and when it begins to fully resolve to the new server.

          Comment

          • ZYV
            Senior Member
            • Sep 2005
            • 315

            #6
            Hi!

            I am very happy about this announcement. To my mind, the step from renting / co-locating the servers to having a facility of one's own is a big one and can probably be compared to the step when you move on from re-selling accounts to co-locating the hardware. I hope I'll live up to the moment when Andrew will finish building his first DC :-) Congratulations to Andrew and the Dathorn team!

            I have a couple of technical comments / questions below:

            1) Re. Western Digital RE4 SATA Disks --- just be sure to check the serial numbers; this year I had to ship a large quantity of these drives back, because they were having head parking issues (read parked like every 15 minutes) with RHEL5 due to a firmware bug and basically in a year of rather light usage (as compared to a shared hosting) parked themselves to death. I have to say that I had no trouble getting a replacement from WD with no questions asked, but still I could have saved myself a week worth of rebuilding RAIDs.

            Also just as Seagate these days, WD seems to have a tendency to assign some cryptic proprietary meaning to the SMART parameters which makes them difficult interpret. I reckon Samsung and Hitachi are among the few that are still exposing standard compliant interface making SMART a bit more useful than just a PASS / FAIL indicator.

            Somehow I have got disappointed in this whole enterprise SATA concept, but apparently what is left is to not go above RAID10 and keep the fingers crossed...

            2) Really cool idea to use SSD for databases! I wonder whether their lifetime will be comparable to the SATA drives in real life. I am a bit anxious about using them even in RAID1, because I have no statistics on lifetime and the internets as full of scary stories.

            3) Does CloudLinux mean that we're back to Apache or mod_hostinglimits can be loaded by LiteSpeed as well?

            4) Why CloudLinux, actually :-) ? RHEL 6 was finally released not long ago and it includes full support for cgroups which look like a very simple, convenient and reliable way to enforce memory, CPU and I/O limits to me. Also, they do not depend upon custom Apache modules etc.; whatever runs as a particular user can be limited.

            Is CloudLinux able to do something that cgroups can't or it's just because it's more integrated, better supported etc.?

            5) Re. KSplice: are you going to produce patches yourself?! Since they've been bought by Oracle they don't seem to provide services for RHEL to new customers anymore

            Comment

            • AndrewT
              Administrator
              • Mar 2004
              • 3653

              #7
              Thanks ZYV!

              1. Though these servers are obviously quite new we've not had any problems with the RE4 drives. These are also very widely used in the hosting industry, etc. and have been around for a while. Regardless, we'll be monitoring arrays closely as always, and now we have hot spares in addition to colds.

              2. SSDs have really come down in price and their performance is nothing short of amazing compared to a rotational disk, which is often the biggest bottleneck in a shared hosting server. I've been using them in my own PCs for a while now and this was something that I really wanted to implement in the new servers. Fortunately we were able to get it done. Next time around maybe we'll be able to move to all SSDs. Only time will tell what their lifetime will be in an environment like this.

              3. We're still using LiteSpeed which offers its own support for the LVEs.

              4. Mainly we just weren't comfortable moving to RHEL 6 so soon. Granted it has been out for ~9 months and is at 6.1 now, we didn't want to possibly end up with a mess of issues between cPanel and everything else. cPanel only released stable support for RHEL 6 a little over a month ago. They actually plan to use cgroups in CloundLinux 6 so we'll end up there eventually one way or another.

              5. Patches are still being provided to existing clients. We've had an account with them for quite a while now so this wasn't really an issue. I'm sure Oracle will continue to find ways to mess it up though. Meanwhile we'll continue to use it.

              Comment

              • Pedja
                Senior Member
                • Mar 2004
                • 329

                #8
                The onlz thing that concernes me is CloutLinux licencing. As I understood, you have to pay CloudLinux licence on monthly basis and if you do not pay or something else happens, CloudLinux stops working. That means that you and we become dependant on CloudLinux...

                Althrough I trust Dathorn 100% am not that sure about CloudLinux, if for nothing else, then just for licencing method they use which leads to blackmailing (I noticed that monthly fee for CloundLinux raised several times and now is doubled that initial).

                Comment

                • AndrewT
                  Administrator
                  • Mar 2004
                  • 3653

                  #9
                  That really isn't an issue at all. Many providers are using CloudLinux now and cPanel even officially supports and licenses it. It is highly unlikely that it is going away. Regardless, CloudLinux installs on top of RHEL / CentOS. Reverting to a vanilla RHEL / CentOS is fairly trivial. Their licensing system can't just "break" your system.

                  Comment

                  • powvex
                    Member
                    • Mar 2004
                    • 67

                    #10
                    new infrastructure

                    I've seen new servers are in somewhere called CoreExchange. Is it reliable?? I've never heard of them before. also cant find much about them or their network...

                    Comment

                    • AndrewT
                      Administrator
                      • Mar 2004
                      • 3653

                      #11
                      powvex, I've merged your new thread into this existing, related one. Rest assured that we wouldn't be putting our hardware there if we didn't think that it was reliable. We've had some gear in there for a few months now with zero down time as measured by external monitoring. We've also been monitoring their network from the outside since they even became an option, again with zero issues.

                      Comment

                      • mIDO
                        Junior Member
                        • Mar 2004
                        • 28

                        #12
                        One question. I think you wrote the SSD disk would be for MySQL only. This means that our MySQL databases will be now limited? What if MySQL disk space exceeds the hard disk drive space?

                        Comment

                        • AndrewT
                          Administrator
                          • Mar 2004
                          • 3653

                          #13
                          The SSDs are used for MySQL but they are also used for many other things. As always, you don't have to worry about the server running out of disk space. That is our responsibility and we've planned accordingly. Our new servers that are "full" (as in no new migrations to them) are only cracking 33% SSD utilization at most and this is even with the SSDs being under partitioned for wear purposes.

                          You will notice that MySQL disk usage is now counted towards a cPanel user's disk usage. This is a fairly new feature that cPanel implemented and it simply was not enabled on our older servers. It's only fair that this be enabled really. Disk space is disk space regardless of where it is used and scripts are storing more and more data in MySQL databases.

                          Comment

                          Working...