1. Please take a little time for this simple survey! Thank you for participating!
    Dismiss Notice
  2. Dear Pleskians, please read this carefully! New attachments and other rules Thank you!
    Dismiss Notice
  3. Dear Pleskians, I really hope that you will share your opinion in this Special topic for chatter about Plesk in the Clouds. Thank you!
    Dismiss Notice

plesk & nas (& lb/ha)

Discussion in 'Plesk for Linux - 8.x and Older' started by cvgeldern, Oct 13, 2005.

  1. cvgeldern

    cvgeldern Guest

    0
     
    As our hosting business grows and our cutomers are mostly very demanding companies, we want to take our services to the next level of reliability.
    So we are researching technics like Load Balancing and High Availability using NAS and syncing multiple geographical locations.

    As with some dedicated custom installed servers (and our DNS and Mail) we already succeeded in some manners, but now we want to do so with our Plesk servers.
    Has anyone tried to install Plesk using a NAS? Or synchronizing data? And what were the experiences?

    We are especially interested in a NAS solution to Plesk as it has some other benefits like scalability of storage and processor power.
     
  2. atomicturtle

    atomicturtle Golden Pleskian

    29
     
    Joined:
    Nov 20, 2002
    Messages:
    2,110
    Likes Received:
    7
    Location:
    Washington, DC
    Sure, I've done it several ways. It depends on what you're trying to do really, but a good rule of thumb is the "better, faster, cheaper: Pick any two" rule.

    First way is to use plesk expand, which breaks pieces of PSA up into multiple servers.

    Second, would be a distributed cluster, where each service lives on 1+n servers. These are very custom configurations, and tend to have the high availability/max performance. Faster and Better, but not cheaper

    Third would be the server virtualization route, where you combine multiple physical servers into one virtual server (viruozzo in reverse). above average performance, maximum availability. Better and faster, but not cheaper

    The "diet coke" configurations, would be the hot-spare model, where you've got a duplicate of the live running box. Better, and Cheaper, but not faster.

    All of these have their shortcomings, so one thing you'll have to plan for are the things that just wont work in clustered environments, and the additional costs you pick up in human resources. That being said, you can do some things with the clustered environments that you cant on a regular PSA box, like geographic load balancing, or packing in 10,000+ domains into a single cluster with five 9 availability.
     
  3. cvgeldern

    cvgeldern Guest

    0
     
    Thank you for your reply, we already used your mail gateway/backup setup, so I already assumed you know what your talking about ;)

    As I know what you mean in your scenarios (I think), I would like to ask you for a topology example/image (or whatever) for each of them to have some visualistion of them.
    Off course we already did some research, but with Plesk we need to know all downsides and impossibilities of the mentioned scenarios.

    Our own preference seems to go to the Second or Third option. About costs it's simple: initial larger investments are no issue, costs in the long term (TCO) are. :D

    Which scenarios do you use for Plesk? And are there any other people here with practical experiences in this matter?
     
  4. atomicturtle

    atomicturtle Golden Pleskian

    29
     
    Joined:
    Nov 20, 2002
    Messages:
    2,110
    Likes Received:
    7
    Location:
    Washington, DC
    Some examples for option #2 was one hosting operation where it was a mass "one-click" environment, minimal flexibility on the CP side so PSA was more of an automation framework. The design goal was to host 10's of thousands of domains, with minimal downtime. That system has been going for about 3 years now (and no downtime I might add, not in 3 years). Another environment used the same model for some very high load email hosting, and a few high volume websites.


    Option #3 is a fairly new design that I've only done 2 of so far, both in an application hosting environment (CRM, etc). Its very good on the middleware layer for processing (so great for bloaty J2EE apps). One thing you cant do with these is use more than 1 IP, so when you get into SSL certs youve got to add in an additional proxy server layer. The good news is that PSA will (mostly) install without modification into it.
     
  5. cvgeldern

    cvgeldern Guest

    0
     
    ok, we did some testing and are now ready to brainstorm of any logic mistakes we could make now.

    one issue is from where in plesk to mount to the storage cluster:
    /usr?
    /usr/local/psa?
    /usr/local/psa/home?
    off course we want all data from plesk like mail, mysql, html on the shared storage, and we could make some links

    the second issue is how to update plesk, depending on from where to mount to the shared storage?
    usually plesk also updates the homedirs etc, so that could break other nodes in the (frontend) cluster

    there will be related issues, but these are ones we come up with now...
     
  6. atomicturtle

    atomicturtle Golden Pleskian

    29
     
    Joined:
    Nov 20, 2002
    Messages:
    2,110
    Likes Received:
    7
    Location:
    Washington, DC
    You'll probably have to take the existing PSA packages apart for updates in a clustered environment and make your own custom ones. Thats how we did it on the option #2 type systems. The only systems where you can basically drop the packages in as-is are the option #3 type designs (inverted vserver).
     
  7. incipio

    incipio Guest

    0
     
    Hi ART, it seems to me that option #3 (inverted vserver) is a great overall solution for "ha psa", but have you actually implemented a cluster that ran psa using this method or just "other" applications like J2EE apps? More importantly, can you expand on any issues you've had, specifically the problems with more than 1 IP address and SSL. What's involved in adding the "additional proxy server" layer? Thank you.
     
  8. atomicturtle

    atomicturtle Golden Pleskian

    29
     
    Joined:
    Nov 20, 2002
    Messages:
    2,110
    Likes Received:
    7
    Location:
    Washington, DC
    Sure I've built a lot of clusters using all of these different designs. I could probably write a book on the topic (I'll have to talk to my editor) as they are very complex. The rule is: better, faster, cheaper... pick any two. It really depends on what you are trying to do.

    Some general observations for better/faster, you need a lot of systems, and everything needs to be redundant. In general, 8 nodes is good place to start. You'll need a very robust out of band backbone (again redundant), paired gigabit switches for example. You'll also need a solid load balancing layer, I use linux-ha, but there are pleanty of commercial apps out there, like bigIP. I like GFS for the file system layer, which ideally youd dedicate another rundundant gigabit network. If you do opt for massively large file systems, say over 8TB+, then definitely go with a storage array from EMC. You'll save a lot of time and money in the long run.
     
  9. superbock

    superbock Guest

    0
     
    Anyone ever used m/cluster for MySQL, integrated with Plesk? I'm starting to look at it, and seems like a good product. Trying to find some success stories about it, preferably with control panel integration.
     
  10. superbock

    superbock Guest

    0
     
    Ok, got a quote from Emic / Continuent, and didn't like it.

    Still looking for a nice MySQL Load-Balancing solution that can be integrated with Plesk in any way.

    Ideas? (scott?)

    br
     
  11. sieb@

    sieb@ Guest

    0
     
    We've completed the majority of our virtualization migration. We've managed to take a dozen seperate servers and services and combined them into three on one big machine running VMWare GSX Server.

    We didn't set out for HA, just reduced downtime and reduced chance of hardware failure. Going all out HA/Clustering would have taken too much time, resources, and money to setup. I talked a bit with Scott about it and did alot of research but it would have been too much for our sized operation. But now with Virtualization, if I make changes/updates to a server, I can take a snapshot before I make the changes so when they blow up, I just revert to the snapshot and all is well again. Only about a 5 minute process.

    If the main server itself dies for some reason (which is also redundant hardware wise), we can quickly load our base images onto a new server and restore from the latest backup. About the fraction of the time from loading from bare metal.

    Virtualization is a more cost effective solution and a nicer way to start/manage resources. You can even cluster within the virtual environment if you had the horsepower. With ESX server, you can abstract the entire Virtual layer from the Physical layer and seemlessly move virtual servers around (given you have the fiber infrastructure for highspeed inter-communication). I highly recommend it.

    Scott, I'd buy that book. :D
     
  12. atomicturtle

    atomicturtle Golden Pleskian

    29
     
    Joined:
    Nov 20, 2002
    Messages:
    2,110
    Likes Received:
    7
    Location:
    Washington, DC
    Yeah this is a really good digression here, if you're going for High Availability, vmware is definitely a really good option. If you store the images on an NAS or other cluster file system, then you've esentially abstracted away your hardware layer for serving your OS's. You will take a performance hit with vmware, but your whole system is about as portable as it can get. And snapshots make backups completely moot. In fact I do all of my testing in vmware for this specific reason.

    Now for maximum performance, and availability imagine setting up a virtuozzo or linux vserver cluster. You basically get the same thing as above, but with higher performance at the cost of OS flexibility and snapshots (you can still do this with rsync however).
     
  13. DotnetKris

    DotnetKris Guest

    0
     
    How about storing the data on a SAN rather than the clustered file system?
     
Loading...