I am working on a project that involves a peer to peer network. Someone raised concerns that we may be expecting a larger bandwidth than is reasonable.
Suppose we had a large number of registered nodes (in the thousands, possibly 10,000), and these nodes constantly are receiving data which they wish to propagate around the network. The data doesn't have to get to every node, but we would like it to get to most of them.
In general, how much data creation and transmission could be handled reasonably as the number of nodes increases? I am hoping that, in my case, the answer is more than 50 MB/minute (as this is the maximum amount of data my system is expected to create), but I don't have a basis for comparison.
Related
I am currently reading up on some distributed systems design patterns. One of the designs patterns when you have to deal with a lot of data (billions of entires or multiple peta bytes) would be to spread it out across multiple servers or storage units.
One of the solutions for this is to use a Consistent hash. This should result in an even spread across all servers in the hash.
The concept is rather simple: we can just add new servers and only the servers in the range would be affected, and if you loose servers the remaining servers in the consistent hash would take over. This is when all servers in the hash have the same data (in memory, disk or database).
My question is how do we handle adding and removing servers from a consistent hash where there are so much data that it can't be stored on a single host. How do they figure out what data to store and what not too?
Example:
Let say that we have 2 machines running, "0" and "1". They are starting to reach 60% of their maximum capacity, so we decide to add an additional machine "2". Now a large part the data on machine 0 has to be migrated to machine 2.
How would we automate so this will happen without downtime and while being reliable.
My own suggested approach would be that the service hosing consistent hash and the machines would have be aware of how to transfer data between each other. When a new machine is added, will the consistent hash service calculate the affected hash ranges. Then inform the affect machine
of the affected hash range and that they need to transfer affected data to machine 2. Once the affected machines are done transferring their data, they would ACK back to the consistent hash service. Once all affected services are done transferring data, the consistent hash service would start sending data to machine 2, and inform the affected machine that they can remove their transferred data now. If we have peta bytes on each server can this process take a long time. We there for need to keep track of what entires where changes during the transfer so we can ensure to sync them after, or we can submit the write/updates to both machine 0 and 2 during the transfer.
My approach would work, but i feel it is a little risky with all the backs and forth, so i would like to hear if there is a better way.
How would we automate so this will happen without downtime and while being reliable?
It depends on the technology used to store your data, but for example in Cassandra, there is no "central" entity that governs the process and it is done like almost everything else; by having nodes gossiping with each other. There is no downtime when a new node joins the cluster (performance might be slightly impacted though).
The process is as follow:
The new node joining the cluster is defined as an empty node without system tables or data.
When a new node joins the cluster using the auto bootstrap feature, it will perform the following operations
- Contact the seed nodes to learn about gossip state.
- Transition to Up and Joining state (to indicate it is joining the cluster; represented by UJ in the nodetool status).
- Contact the seed nodes to ensure schema agreement.
- Calculate the tokens that it will become responsible for.
- Stream replica data associated with the tokens it is responsible for from the former owners.
- Transition to Up and Normal state once streaming is complete (to indicate it is now part of the cluster; represented by UN in the nodetool status).
Taken from https://thelastpickle.com/blog/2017/05/23/auto-bootstrapping-part1.html
So when the joining node is in the Joining State, it is receiving data from other nodes but not ready for reads until the process is complete (Up status).
DataStax also has some material on this https://academy.datastax.com/units/2017-ring-dse-foundations-apache-cassandra?path=developer&resource=ds201-datastax-enterprise-6-foundations-of-apache-cassandra
I am writing a simple homogenous cluster application using Akka 2.2.3 and Scala; a particle filtering algorithm in which each node shares data with other cluster members at random points in time. It's currently a research application, not a business critical system.
At present, every node sends a fixed size message to a randomly selected node each second. This works, but I have concerns regarding performance when scaling (e.g. cloud versus local)
Nodes may get overloaded sending data
Nodes may get overloaded with incoming messages from other cluster members
The network may become the bottleneck
I'd like to run the application with sized clusters on different networks and and achieve good performance without manual tuning/monitoring. What simple approaches could I take to tuning message size and frequency to mitigate the above concerns?
You can try to use Adaptive Load Balancing of Akka cluster-aware routers.
Or you can try to create your own mechanism for reduce nodes overload. For example it may contain a custom mailbox for your Receiver actor which may periodically add a message containing the current mailbox queue length to the head of the mailbox queue. Then your Receiver actor may relay such a message to all cluster nodes. In this case you will be able to maintain cluster-wide statistics of Receiver's mailboxes usage (just remember to include TTL in that statistics and treat no statistics message received as empty mailbox indicator).
If your application can tolerate partial data losses then you can simply drop extra messages directly from the Receiver's mailboxes (as well will require a custom mailbox implementation).
If it's possible with your algorithm another approach would be to turn the thing around and have nodes ask other random nodes for data at random points in time. Work pulling is usually easier to control than pushing.
I'm new to distributed systems, and I'm reading about "simple Paxos". It creates a lot of chatter and I'm thinking about performance implications.
Let's say you're building a globally-distributed database, with several small-ish clusters located in different locations. It seems important to minimize the amount of cross-site communication.
What are the decisions you definitely need to use consensus for? The only one I thought of for sure was deciding whether to add or remove a node (or set of nodes?) from the network. It seems like this is necessary for vector clocks to work. Another I was less sure about was deciding on an ordering for writes to the same location, but should this be done by a leader which is elected via Paxos?
It would be nice to avoid having all nodes in the system making decisions together. Could a few nodes at each local cluster participate in cross-cluster decisions, and all local nodes communicate using a local Paxos to determine local answers to cross-site questions? The latency would be the same assuming the network is not saturated, but the cross-site network traffic would be much lighter.
Let's say you can split your database's tables along rows, and assign each subset of rows to a subset of nodes. Is it normal to elect a set of nodes to contain each subset of the data using Paxos across all machines in the system, and then only run Paxos between those nodes for all operations dealing with that subset of data?
And a catch-all: are there any other design-related or algorithmic optimizations people are doing to address this?
Good questions, and good insights!
It creates a lot of chatter and I'm thinking about performance implications.
Let's say you're building a globally-distributed database, with several small-ish clusters located in different locations. It seems important to minimize the amount of cross-site communication.
What are the decisions you definitely need to use consensus for? The only one I thought of for sure was deciding whether to add or remove a node (or set of nodes?) from the network. It seems like this is necessary for vector clocks to work. Another I was less sure about was deciding on an ordering for writes to the same location, but should this be done by a leader which is elected via Paxos?
Yes, performance is a problem that my team had seen in practice as well. We maintain a consistent database & distributed lock manager; and orignally used Paxos for all writes, some reads and cluster membership updates.
Here are some of the optimizations we did:
As much as possible, nodes sent the transitions to a Distinguished Proposer/Learner (elected via Paxos), which
decided on write ordering, and
batched transitions while waiting for the response from the prior instance. (But batching too much also caused problems.)
We had considered using multi-paxos but we ended up doing something cooler (see below).
With these optimizations, we were still hurting for performance, so we split our server into three layers. The bottom layer is Paxos; it does what you suggest; viz. merely decides the node membership of the middle layer. The middle layer is a custom-in-house-high-speed chain consensus protocol, which does consensus & ordering for the DB. (BTW, chain-consensus can be viewed as Vertical Paxos.) The top layer now just maintains the database/locks & client connections. This design has lead to several orders of magnitude latency and throughput improvement.
It would be nice to avoid having all nodes in the system making decisions together. Could a few nodes at each local cluster participate in cross-cluster decisions, and all local nodes communicate using a local Paxos to determine local answers to cross-site questions? The latency would be the same assuming the network is not saturated, but the cross-site network traffic would be much lighter.
Let's say you can split your database's tables along rows, and assign each subset of rows to a subset of nodes. Is it normal to elect a set of nodes to contain each subset of the data using Paxos across all machines in the system, and then only run Paxos between those nodes for all operations dealing with that subset of data?
These two together remind me of the Google Spanner paper. If you skip over the parts about time, it's essentially doing 2PC globally and Paxos on the shards. (IIRC.)
I have been reading about a database named Starcounter. It makes a claim that it can handle loads that a "NoSql"-database only can handle without dropping consistency. As far as I understand the CAP-theorem, if you keep consistency, you lose availability or partition tolerance. So what trick makes StarCounter work?
I can imagine that StarCounter is fast, but the claim that NoSql needs to drop consistency to keep up seems a little bit strange to me. Can anyone please explain?
Thanks in advance
Roland
The short answer
The CAP theorem (aka Brewers theorem) cannot be beaten for a single piece of information (like a consistent database). If you have a horizontally scaled database, you won't get consistency and performance. This conclusion comes from the laws of physics and can be deducted from Brewers theorem and Einsteins theories of relativity. You need to scale-in/up, not out. Not very "cloudy", but as the enemies of Galileo would probably confess if they were alive today, nature does a poor job at honouring human fashion.
Scaling consistent data
I'm sure there are other approaches, but Starcounter works by hosting the database image in RAM. Instead of moving database data to the application code, parts of the application code is moved to the database. Only data in the final response gets moved from the original place in RAM memory (where the data was in the first place). This makes most of the data stay put even if there are millions of requests processed every second. The downside is that the database needs to know the programming language of your application logic. The upside, however, is obvious if you have ever tried to serve millions of HTTP requests/sec, each requiring extensive database access.
A more thourough answer
The question is a good one. It is no wonder you find it strange as it was only a few years back that CAP was proven (turned into a theorem). Many developers are as disappointed as a kid would be when theoretical physicist tells him to stop looking for the perpetual motion machine because it cannot work. We still want the scale-out consistent database, don't we?
The CAP theorem
The CAP theorem gives that any piece of information cannot have consistency (C), availability (A) and partition tolerance (P). It applies to a unit of information (such as a database). You can of course have independent pieces of information that operates differently. One piece could be AP, another could be CA and a third could be CP. You just cant have the same information being CAP.
The problem with the impossibility of the 'P' in a consistent and available database results in how a scaled-out database MUST do signalling between the nodes. The conclusion must be, that even in a hundred years from now, CAP gives that a single piece of consistent data will have to live on hardware interconnected using hard wires or light beams.
The problem with the P in CAP
The problem lies in performance if you apply horizontal scaling to an available consistent database. A good performance was the very reason to do horizontally scaling in the first place, this is a very bad thing. As every node needs communicate with the other nodes whenever there is database access to achieve consistency, and given the fact that signalling is ultimately limited by the speed of light, you are left with sad but true fact that database scientist (as well as CPU scientists) are not just being stubborn for failing to see scale-out as a a magical silver bullet. It will not happen because it cannot happen (however, parts of your database could be placed in a AP set, so remember, we are talking about consistent data here). Adding the theories of Einstein to the CAP theorem and the small box wins of the cloudy data-center for consistent data.
Perpetual machines and CAP
The state of things in the database community is a little bit like the state of perpetual motion machines when horse and carriage was the way to get to work. Without any theoretical evidence against it, the patent offices granted hundreds of patents for impossible perpetual machines. Today, we may laugh at this, but we have a similar situations in the database industry with consistent scale-out databases. When you hear somebody claim that they have a scale-out ACID database, be cautious. It was only after the dot com crash mathematicians at MIT proved Brewer right at the CAP theorem was officially born, so the hunt for the impossible has unfortunately not died off just yet. You can compare this, if you want, to the way laggards kept trying to invent the perpetual machine for years after modern theoretical physics should reasonably have put a stop to it. Old habits die hard (my apologies to anyone on Stack overflow still making drawings of bearings and arms moving ad finitum on their own accord - I don't mean to be offensive).
CAP and performance
All is not lost however. Not all pieces of information needs to be consistent. Not all pieces needs to scale-out. You just have the accept Brewers theorem and make the best out of it.
For applications such as Facebook, consistency is dropped. This is okay as data is entered once and then is manipulated by a single users. Still we can experience the side effects in everyday Facebook usage such as things popping in and out of existence for a while.
However, in most business applications, data needs to be correct. The sum of all accounts in your bookkeeping needs to amount to zero. Your stock inventory must equal to 8 if you sold 2 out of 10 items even if there are multiple users buying from the same stock.
The problem with scaling out available data is that you have to make do without partition tolerance. This fancy word simply means that you have to signal between the nodes in your cloud at all times. And as it takes light a few nanoseconds to travel a single meter, this becomes impossible without making your scale-out result in less performance rather than more performance. Of course, this is only true for consistent data. The implications of this has been known by the engineers of Intel, AMD, Oracle et. al for a long time. It is not their scientist haven't heard of scale-out. It is just that they have come to accept the world as Einstein described it.
Some comfort in the gloom
If you do the math, you find that a single PC has instructions to spare on each human being living on Earth for each second it is running (google on 'modern CPU' and 'MIPS'). If you do some more math, like taking the total turnover of Amazon.com (you can find it at wwww.nasdaq.com) divided by the price of an average book, you will find that the total number of sales transactions can fit in RAM of a single modern PC. The cool thing is that the number of items, customers, orders, products etc. occupies the same amount of space in 2012 as it did in 1950. Images, video and audio has increased in size, but numeric and textual information does not grow per item. Sure the number of transactions grows, but not in the same phase as computer power grows. So the logical solution is to scale out read-only and AP data and "scale-in/up" business data.
"Scale-in" instead of "scale-out"
Database engines and business logic running in a VM (like the Java VM or the .NET CLR) typically use fairly effective machine code. This means that moving memory is the overshadowing bottleneck of total throughput for a consistent database. This is often referred to as the memory wall (wikipedia has some useful information).
The trick is to transfer code to the database image instead of data from the database image to the code (if using a MVC or a MVVM pattern). This means that the consuming code executes in the same address space as the database image and that data is never moved (and the disk is merely securing transactions and images). Data can stay in the original database image and does not have to be copied into the memory of the application. Instead of treating the database as a RAM database, the database is treated as primary memory. Everything stays put.
Only data that is part of the final user response is moved out of the database image. For a large scale applications with hundreds of millions of simultaneous users this typically amounts to only a few million requests per second, something that a single PC has no problem with handling given that the HTTP packaging is done on gateway servers. Fortunately, such servers scales out beautifully as they don't need to share data.
As it turns out, the disk is fast at sequential writes so a raided disk can persist terabytes or changes every minute.
Horizontal scaling in Starcounter
Normally you do not scale a Starcounter node. It scales-in rather than out. This works well for a few million simultaneous users. To go above that, you need to add more Starcounter nodes. They can be used to partition data (but then you lose consistency and Starcounter is not designed for partitioning so it is less elegant than solutions such as Volt DB). So a better alternative is to use the additional Starcounter nodes as gateway servers. These servers simple accumulates all incoming HTTP requests for a millisecond at a time. This might sound like a short amount of time, but it is enough to accumulate thousands of request if you decided you need to scale Starcounter. The batch of requests are then sent to the ZLATAN node (Zero LATency Atomicity Node) a thousand times a second. Each such batch can contain thousands of requests. In this way, a few hundred million user sessions can be served by a single ZLATAN node. Although you can have several ZLATAN nodes, there is only one active ZLATAN node at a time. This is how the CAP theorem is honored. To go above that, you need to consider the same tradeoff as Facebook and others.
Another important note is that the ZLATAN node does not serve applications with data. Instead, the applications controller code is run by the ZLATAN node. The cost of serializing/deserializing and sending data to an application is far greater than to process the controller logic cycles. I.e. the code is sent to the database instead of the other way around (a traditional approach is that the applications asks for data or sends data).
Making the "shared-everything" node faster by doing less
The use of the database as a "heap" for the programming language instead of a remote system for serialization and deserialization is a trick that Starcounter calls VMDBMS. If the database is in RAM, you should not move data from one place in RAM to another place in RAM which is the case with most RAM databases.
There is no 'trick'. Starcounter is talking about speed, while CAP/NoSQL are talking about scalability. There is a trade-off between features+scalability vs speed.
Sometimes it's OK to ignore scalability if you can prove there are bottlenecks elsewhere. For instance, a new startup shouldn't worry about their website scaling to a million users, they should worry about getting their first hundred users. (Does anyone remember how often Twitter was down in the early days?) Starcounter can be useful if their transaction rate is much greater than your web page hit rate.
On the other hand, I don't trust anyone who lumps all "NoSQL" Databases together. The various NoSQL databases are more different than alike. They have radically different architectures and properties. Some of them scale to thousands of nodes, some of them don't scale beyond one node. Sometimes adding scalability slows you down. Sometimes removing features speeds you up.
http://strata.oreilly.com/2010/12/strata-gems-mysql-handlersocket.html
We have a data system in which writes and reads can be made in a couple of geographic locations which have high network latency between them (crossing a few continents, but not this slow). We can live with 'last write wins' conflict resolution, especially since edits can't be meaningfully merged.
I'd ideally like to use a distributed system that allows fast, local reads and writes, and copes with the replication and write propagation over the slow connection in the background. Do the datacenter-aware features in e.g. Voldemort or Cassandra deliver this?
It's either this, or we roll our own, probably based on collecting writes using something like
rsync and sorting out the conflict resolution ourselves.
You should be able to get the behavior you're looking for using Voldemort. (I can't speak to Cassandra, but imagine that it's similarly possible using it.)
The key settings in the configuration will be:
replication-factor — This is the total number of times the data is stored. Each put or delete operation must eventually hit this many nodes. A replication factor of n means it can be possible to tolerate up to n - 1 node failures without data loss.
required-reads — The least number of reads that can succeed without throwing an exception.
required-writes — The least number of writes that can succeed without the client getting back an exception.
So for your situation, the replication would be set to whatever number made sense for your redundancy requirements, while both required-reads and required-writes would be set to 1. Reads and writes would return quickly, with a concomitant risk of stale or lost data, and the data would only be replicated to the other nodes afterwards.
I have no experience with Voldemort, so I can only comment on Cassandra.
You can deploy Cassandra to multiple datacenters with an inter-DC latency higher than a few milliseconds (see http://spyced.blogspot.com/2010/04/cassandra-fact-vs-fiction.html).
To ensure fast local reads, you can configure the cluster to replicate your data to a certain number of nodes in each datacenter (see "Network Topology Strategy"). For example, you specify that there should always be two replica in each data center. So even when you lose a node in a data center, you will still be able to read your data locally.
Write requests can be sent to any node in a Cassandra cluster. So for fast writes, your clients would always speak to a local node. The node receiving the request (the "coordinator") will replicate the data to other nodes (in other datacenters) in the background. If nodes are down, the write request will still succeed and the coordinator will replicate the data to the failed nodes at a later time ("hinted handoff").
Conflict resolution is based on a client-supplied timestamp.
If you need more than eventual consistency, Cassandra offers several consistency options (including datacenter-aware options).