Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Looking for reliable VPS in Atlanta
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Looking for reliable VPS in Atlanta

Hey All,

I have a 3-node MariaDB Galera cluster that's running on a Delimiter dedicated server and two 6GB RAM HostUS VPSes (on different physical servers). The cluster relies on consensus voting/quorum, so for the cluster to work, at least two nodes must be alive and well. To avoid split-brain, the cluster refuses to function as soon as two of the three nodes are down.

Anyway, to keep the story short, while HostUS is pretty good overall, I just hit this problem now, and my apps died as a result. My two VPSes are still down as we speak.

Before you ask "did you file a ticket?", I'd like to first say that this isn't a HostUS complaint. This is me wanting to become more robust, and have another VPS in the mix to help me maintain quorum.

In short, if you have any good high-RAM offers in Atlanta, I'm all ears. I really would like low latency to both Delimiter and HostUS; Galera is sensitive to that. If anyone could match the HostUS offer, that'd be ideal, but not required.

Comments

  • matthewvzmatthewvz Member, Host Rep

    RamNode. They are a little pricier than HostUS but if you are looking for good uptime and awesome performance RamNode is the way to go. I've had nothing but near perfect uptime with RamNode.

    Thanked by 1Nick_A
  • MarkTurnerMarkTurner Member
    edited September 2014

    What about Delimiter's dual X5150, 16GB, 500GB disk $20/month (paid quarterly), you could setup three of them, private VLAN if need-be.

    If its an emergency then we can try to get them setup immediately - PM me if you need this.

  • emreemre Member, LIR

    @amhoab I've got one delimiter server doing nothing,

    virtualmin/webmin installed on it.

    I will share it with you if you teach me MariaDB Galera cluster.

    pm me if interested.

  • Nick_ANick_A Member, Top Host, Host Rep

    Check us out! Not the cheapest, but voted the best here for a year straight if that matters.

  • VPSDime

  • matthewvz said: RamNode

    I hear a lot of good about them, but their LE[BT] offers are pretty low in resources. The prices on their site for the specs I'm looking for are a bit too up there.

    MarkTurner said: Delimiter

    I certainly considered this, but fear that I might be in a predicament similar to the one I'm in now. If there's ever wide maintenance (or an outage) at Delimiter, I want to ensure that everything still remains online. I have a 48GB RAM server from you all through a deal I got about a year ago, and it's been working pretty well, although there was a maintenance or two. That's what concerns me. If you have separate availability zones, I'm all ears.

    emre said: I've got one delimiter server doing nothing

    If you truly don't mind me squatting on it, I could expand into a 5-node cluster. That would help mitigate the risks I described in response to @MarkTurner. I'd be happy to show you the ropes around Galera.

  • matthewvzmatthewvz Member, Host Rep

    @0xdragon said:
    VPSDime

    Not in Atlanta.

  • @amhoab - I've PM'ed you

  • ramhost.us

  • GreenHostBoxGreenHostBox Member
    edited September 2014

    How's New Jersey? It ain't too far from Atlanta and our server is receiving awesome low ping from HostUS as you can see below. We can make a special offer for your request.


    [root@server]# ping -c 4 162.245.216.243

    PING 162.245.216.243 (162.245.216.243) 56(84) bytes of data.

    64 bytes from 162.245.216.243: icmp_seq=1 ttl=50 time=30.9 ms

    64 bytes from 162.245.216.243: icmp_seq=2 ttl=50 time=31.1 ms

    64 bytes from 162.245.216.243: icmp_seq=3 ttl=50 time=31.0 ms

    64 bytes from 162.245.216.243: icmp_seq=4 ttl=50 time=30.9 ms



    --- 162.245.216.243 ping statistics ---

    4 packets transmitted, 4 received, 0% packet loss, time 3033ms

    rtt min/avg/max/mdev = 30.919/30.997/31.105/0.069 ms

  • @rajprakash said:
    ramhost.us

    Now that brings back memories...

  • RamNode!

    Thanked by 1Nick_A
  • aglodekaglodek Member
    edited September 2014

    @amhoab said: If you truly don't mind me squatting on it, I could expand into a 5-node cluster. That would help mitigate the risks I described in response to MarkTurner. I'd be happy to show you the ropes around Galera.

    I'm working on a similar setup, coincidentally also in Atlanta, with one node of the cluster on a Delimiter dedi ;) However, given how Galera works, I'm toying with the idea of a 3-city distributed setup (e.g. Atlanta + Phoenix + NYC) - in my case, this is to serve as back end for apps on multiple boxes across the US, Europe and Asia.

    With respect to your comment above: do keep in mind that adding 2 additional nodes (for a total of 5) will slow the cluster down considerably (by 20-25% maybe?), with little upside in terms of redundancy.

    EDIT: this said, additional inactive, but ready to go nodes may be a good idea in case of failure of one of the primary 3 nodes though. Maybe even spin them up and update DB content once in a while? ;)

  • @aglodek - if you use Delimiter for those three cities we can do them all on-net rather than over public Internet. For things like Galera/MongoDB/Couch you need predictable latency.

    Thanked by 1aglodek
  • AlexanderMAlexanderM Member, Top Host, Host Rep

    @amhoab said:
    Hey All,

    I have a 3-node MariaDB Galera cluster that's running on a Delimiter dedicated server and two 6GB RAM HostUS VPSes (on different physical servers). The cluster relies on consensus voting/quorum, so for the cluster to work, at least two nodes must be alive and well. To avoid split-brain, the cluster refuses to function as soon as two of the three nodes are down.

    Anyway, to keep the story short, while HostUS is pretty good overall, I just hit this problem now, and my apps died as a result. My two VPSes are still down as we speak.

    Before you ask "did you file a ticket?", I'd like to first say that this isn't a HostUS complaint. This is me wanting to become more robust, and have another VPS in the mix to help me maintain quorum.

    In short, if you have any good high-RAM offers in Atlanta, I'm all ears. I really would like low latency to both Delimiter and HostUS; Galera is sensitive to that. If anyone could match the HostUS offer, that'd be ideal, but not required.

    Unfortunately we received a 40Gbps+ DDOS this morning and our network provider shut down or BGP Session resulting in around an hour of downtime at 4AM EST.

    Thanks,
    Alexander

  • Go with Ramnode, no complaints.. ever.. about anything - worth the cash 1000%. You're already with HostUs - just picked up a cheap yearly with them and all good so far (I haven't put anything serious on the server or setup monitoring yet).

    Thanked by 1Nick_A
  • @MarkTurner said: aglodek - if you use Delimiter for those three cities we can do them all on-net rather than over public Internet. For things like Galera/MongoDB/Couch you need predictable latency.

    I expect there will be about 15% writes to 85% reads, hence the latency between the 3 Galera nodes is not that critical. With expected latency on the order of 40-70ms between nodes, writes to the cluster will be very slow, but the setup is to serve mostly as high availability and 100% redundant backend for web apps on 20+ servers, located all over, which will be configured to write/read to/from the Galera node in closest proximity first, before failing over to other nodes.

    This said, your offer is of interest to me. Please feel free to PM me with your offer re other cities and on-net connectivity including latency (for comparison to open Internet connectivity between same locations). I am interested in Phoenix, Las Vegas, Seattle, Chicago, NYC and Miami.

  • @AlexanderM said:

    No biggie; I understand that these thing happen and are to be expected. That's another driver for me trying to diversify; having 2 of 3 nodes with one provider in the same place was living on the edge. I'll likely still keep one of the VPSes for this, and the other I'll use for something else, so no worries; I'll be around still :) .

    Thanked by 1AlexanderM
  • GreenHostBox said: How's New Jersey

    I unfortunately need really low ping times (within single digit ms) for the cluster to work to expectations. Could you PM me your offer anyway? I have other needs in that area.

  • aglodek said: I'm toying with the idea of a 3-city distributed setup (e.g. Atlanta + Phoenix + NYC) - in my case, this is to serve as back end for apps on multiple boxes across the US, Europe and Asia.

    With respect to your comment above: do keep in mind that adding 2 additional nodes (for a total of 5) will slow the cluster down considerably (by 20-25% maybe?), with little upside in terms of redundancy.

    That's definitely an interesting setup, and I'm curious to know how it performs. Galera's certification-based replication applies the transaction on all nodes (before commit) to ensure it's successful, and with that type of geo spread, it probably won't be the most performance (write) beast.

    If you can do split reads/writes in your application(s), using standard MySQL replication could give you better write performance (but perhaps some slave lag for reads). You can also use both, with a reliable Galera cluster in one region, and slaves elsewhere.

  • aglodekaglodek Member
    edited September 2014

    @amhoab said: That's definitely an interesting setup, and I'm curious to know how it performs. Galera's certification-based replication applies the transaction on all nodes (before commit) to ensure it's successful, and with that type of geo spread, it probably won't be the most performance (write) beast.

    Yes, I'm aware. However, I'm anticipating very high read to write ratios. This said, I have to test how Drupal handles the anticipated high time lag. Furthermore @MarkTurner raised some valid cluster stability concerns at high latency between nodes.

    If you can do split reads/writes in your application(s), using standard MySQL replication could give you better write performance (but perhaps some slave lag for reads). You can also use both, with a reliable Galera cluster in one region, and slaves elsewhere.

    Yes, I have considered such a hybrid setup, but it adds another layer of complexity to an already complex setup (MariaDB/Galera). I want to see how Galera (and Drupal) handle this first, before complicating things further ;)

    I expect to have 2 test Galera clusters up and running by end of October. As described before, one of them will be a distributed cluster (say, Atlanta + Phoenix + NYC). The second one - call it the "control" cluster - will consist of 3 nodes in close proximity in Atlanta. This will give me the opportunity to compare their benchmarks and work out all the distance-related kinks (i.e. latency and network). I have close to 1,000 gTLDs and ccTLDs of my own to use as a pretty realistic test base, together with a corresponding number of mini websites I expect to have running by December. As you can see, a whole flock of birds with one stone here ;)

  • aglodek said: I'm working on a similar setup, coincidentally also in Atlanta, with one node of the cluster on a Delimiter dedi ;) However, given how Galera works, I'm toying with the idea of a 3-city distributed setup (e.g. Atlanta + Phoenix + NYC) - in my case, this is to serve as back end for apps on multiple boxes across the US, Europe and Asia.
    EDIT: this said, additional inactive, but ready to go nodes may be a good idea in case of failure of one of the primary 3 nodes though. Maybe even spin them up and update DB content once in a while? ;)

    Can I ask what is your Galera config for a WAN setup?
    I've tried to setup such a thing in Amsterdam with 3 vps from 3 different DC and from time to time (3-4 days max of up) the cluster just goes down.

  • @a_chris - This is usually caused by latency. I haven't done much scaling with Galera/MySQL but what I did notice is its very latency sensitive. So if someone is flooding the interface with packets, or there is cross-datacentre latency then you'll find things will fall over.

  • aglodekaglodek Member
    edited September 2014

    @a_chris: still working on this, so can't share config details at this point.

    However, from your description of the whole cluster going down, seems to indicate "lack of quorum" (i.e. 2 out of 3 nodes "failing", maybe due to networking issues or node downtime or a mix of both). Were I to encounter this kind of issue, I would scale back to a single data center to eliminate external factors and see how Galera performs in such "controlled" environment.

Sign In or Register to comment.