So I have a cloud virtual machine on google compute, does this mean by nature that it is highly available? If the VM is running on a single piece of hardware on GCE, if the piece of hardware breaks then the VM could go down. Is the VM running on some kind of RAID, but for servers? So if one of the machines goes down another machine will pick up and continue running the vm? Thanks.
The machine itself is not highly available. However, Google takes several steps to increase reliability:
Storage is replicated and independent of the physical machine the VM is running on (obviously not for local SSD). This means that even if the physical machine catches on fire, only the "runtime" state is lost but the attached disks are fine.
VMs can live-migrate. This is a setting you can control. If enabled, the VM will be migrated to a different physical machine on maintenance events. Live-migration can lead to brief performance degradation while memory etc. is synced to the other host but the machine is not shut down / restarted.
Even when the physical host suddenly dies, you can set your instance to restart automatically on a new machine. If you plan to use this mode, make sure your instance is able to cleanly boot to serving state without manual intervention.
If you need high availability, the best approach is to spread your instances among zones of the same region and using a network or HTTP(S) loadbalancer. These will automatically stop sending traffic to a machine in case it becomes unhealthy. Also see this short youtube video on Google's network architecture for more info.
For high availability of your application data, there are highly available options like Datastore for database-like usage and Cloud Storage for file-oriented data. Keep in mind that Cloud SQL also runs on a single instance/physical machine which means that you have to setup slaves/replicas to get high availability. However, you can also do that with your favorite DB system on plain Compute Engine instances if you are willing to maintain them yourself.
Related
I was studying VMWare's VSphere suite, cloud computing virtualization platform.
I could not figure out whether there's any relation between VMotion and VMFS in the suite?
VMotion enables the live migration of running virtual machines from
one physical server to another with zero down time.
VMFS is a clustered file system that leverages shared storage to allow multiple physical hosts to read and write to the same storage simultaneously.
Is there any relation between them?
No.
As you mention, VMFS is the file system we use by default on "block" shared storage (i.e. LUNs). This allows us to have the same LUN mounted for read/write on multiple ESXi hosts which is not allowed with many file systems.
vMotion is when we move a running VM from one ESXi host to another. We do this by copying the running memory state from one host to another. When then "stun" the VM for a short period of time and quickly move it's virtual NIC to the new server. The VM "starts" on the far side in the same state, thus it appears like the VM has always been running. That is to say we "move" the running VM even though we are actually just creating a new VM with exactly the same memory state and disk.
The only relationship is that if you have a VM whose VMDKs live in a Datastore which is shared across multiple ESXi hosts, the vMotion process doesn't have to copy the VMDK which makes the process much simpler and faster. Since VMFS is one way we can support shared storage, it is common to have VMDK's on VMFS based datastores (in this case 1 datastore = one VMFS formatted LUN). Since VMFS is our oldest shared storage technology, it's the most common and usually be best understood by our customers.
However, any shared storage will work just fine for vMotion, including VSAN, VVOL and NFS based shared storage.
i'm working on an application that needs to be tested in a HPC cluster.
i'm thinking about using xcat as a resource manager.
i don't have much hardware resources, i have one HP desktop and MacBook laptop.
the question: is it possible to set up a virtual cluster (using virtualBox or KVM) on one hardware resource
thanks,
The short answer here is yes, depending on how much memory and disk you have available on your one machine. I've done this numerous times on a MacBook Pro with 8 GB of RAM.
The long answer is that there is absolutely nothing magical about an HPC cluster. All you need to test basic parallel applications in a simulated cluster environment are two or more VMs which meet these criteria:
Same OS, as identical as possible.
Passwordless authentication (ssh key based auth).
Same software stack in same location on all nodes (See #4 or use rsync).
At least one shared filesystem, e.g. NFS mounted $HOME
Shared network with name resolution configured (correct /etc/hosts on all nodes)
None of this requires job schedulers, provisioning tools or any complex networking. You can find many NFS setup howtos to help get one node set up to share $HOME to the others, this might be the most complicated part. VirtualBox does a good job of setting up local networking.
On top of this you can layer setting up a job scheduler like SLURM (highly recommended), provisioning tools like Warewulf or xCat, parallel filesystems across the VMs (BeeGFS is easy to set up and a great introduction), etc. I have had a full featured stateless cluster simulated on my Macbook Pro a number of times using tools from this list and VirtualBox VMs. It's a great way to learn about setting up an HPC cluster.
As a very beginner, I only know how to create VMs and install OS on these using Oracle VirtualBox. All the VMs created are dependent on the hardware resources (CPU, RAM etc.) of a single machine. If the machine goes down the VMs will go down. Need to know how VMs can be created using taking resources from different physical machines (manually or dynamically) to avoid failure of any VMs.
For example: There are 4 physical machines having 8 core and 16GB RAM each. Now, I want to create three VM having having 8 core and 16GB RAM taking from different physical machines. If one physical machine goes down, no VM will be down.
You can look up clustering solutions (e.g. VMware clusters, or Hyper-V failover clusters). In this model, if a physical host goes down, then the virtualization platform will power up the VMs on other hosts.
If you're looking for zero downtime, then VMware has something called Fault Tolerance in which a shadow copy of a VM is running on a different host and is continuously synchronized with the primary copy. If the primary host goes down, the shadow copy can take over with zero downtime (e.g. you don't have to boot from the shadow copy because it's already running). This feature, while cool, has a lot of real-world limitations in how it inter-operates with other features of VMware. For example, as of vSphere 6.0, you cannot do various kinds of migrations for such VMs, etc. I believe it also requires a more expensive license.
These solutions generally require some shared resources between the physical hosts (most notably storage). Otherwise they will not work (or at the very least, performance will greatly suffer).
Docker is an abstraction of OS (kernal) and below, VM is abstraction of Hardware. What is the point of running a Docker on an VM (like Azure) (apart from app portability)? should they not be directly hosting docker on the hardware?
Docker doesn't provide effective isolation for kernel-level security exploits (there's only one ring 0, and it's shared across all containers). Thus, one could reasonably wish to have the additional isolation provided by a virtualization mechanism.
Keep in mind that much of Docker's value is not about security, but about containerization -- building and distributing portable applications in such a way as to ensure that coupling between layers occurs only where and how intended.
The advantage of a cloud system like Azure is that you can go online with your credit card and get a machine up and running in a few minutes. This is enabled by that machine being virtual. Also VMs let you share hardware across multiple users with hardware-level isolation.
If everything else was equal, i.e. you didn't need any of the features of a VM, then you would be correct that a physical machine should be used, as it will run more efficiently.
Simplified, I have an application where data is intended to flow over the internet between two servers. Ideally, I'd like to test at what point the software ceases to function. At what lowerbound limit (bandwidth, latency, dropped packets) do things stop working to test the reliability of the software.
What I thought I would do was the following:
Setup up 3 machines (VMware instances)
Install the 2 applications on two of the servers.
Setup up the 3rd server to sit between the two machines by doing some sort of magic with Routing and Remote Access on Windows 2003
Install either Traffic Shaper XP or NetLimiter to limit the bandwidth
Run something like TMnetSim Network Simulator to simulate a bad connection.
Does this sound like a good idea or are there easier/better ways of doing this? I'm not that comfortable on Linux and my team mates are even less so.
WANem does exactly this. We have used it both in a virtual machine on the desktop and on a dedicated old pc and it worked great. It can simulate all sorts of broken connectivity.
FreeBSDs ipfw has provisions to simulate links with a given bandwith, latency or error rate. You could use that FreeBSD machine as your machine "in the middle" in your above setup.
You probably can also run at least one of the endpoints on the same machine if you want to reduce the amount of servers involved.
Someone actually packaged up the settings and whatnot necessary for the FreeBSD solution to this problem and they call it DUMMYNET.
It simulates/enforces queue and bandwidth limitations, delays, packet losses, and multipath effects. It also implements a variant of Weighted Fair Queueing called WF2Q+. It can be used on user's workstations, or on FreeBSD machines acting as routers or bridges.
It can simulate exactly what you want, and its free and will boot onto commodity hardware. They even have a canned install of it that is small enough to put on a floppy disk (!) that you can download at that link.
Maybe it is time to learn a bit about Linux because adding a 50ms delay on every outgoing packet can be done in typing just one line:
tc qdisc add dev eth0 root netem delay 50ms
For more see the Linux Traffic Control HOWTO
We had a similar requirement some ten years ago - I'll see if I can recall how we managed it.
If I remember, we wrote a socket proxy program which was controlled by inetd on a UNIX box. This socket would accept connections from a client and open equivalent sessions through to the server. It would then loop, passing messages in both directions.
The way we achieved WAN characteristics was to introduce random delays (with upper and lower limits) in both the connection establishment and the passing of data once the link was up.
It also had the feature to drop the link occasionally as WAN links were less reliable for us than local traffic.
I recall we had to make it threaded to stop the delays from affecting reverse traffic on the link.
There is a very good (and free) Microsoft solution for that, we use it for quite some time and it works great, it can very easily simulate every thing(packet loss, low bandwidth, disconnection, latency....)
This is the best solution i found for a windows environment
More information and a download link can be found here: MARCO blog post
this product has gone some evolution and it is now integrated into visual studio as part of the automation testing, but i found the use of the standalone(that is quite hard to find, so keep a local copy) to work much better. keep in mind that you need at least two computers(or VMs) since you need to pass through a network adapter in order for the application to work its magic.