All posts by Heidi-ann

Majority agreement is not necessary for consensus

Did you know that majority agreement isn’t required by Paxos?

In fact, most of the time, the sets of nodes required to participate in agreement (known as quorums) do not even need to intersect with each other.

This is the observation which is made, proven and realised in our latest paper draft, Flexible Paxos: Quorum Intersection Revisited. A draft of which is freely available on ArXiv.

The paper provides the theoretical foundation of the result, as well as the formal proof and a discussion of the wide reaching implications. However, in this post I hope to give a brief overview of the key finding from our work and how it relates to modern distributed systems. I’d also recommend taking a look at A More Flexible Paxos by Sugu Sougoumarane.

 

Background

Paxos is an algorithm for reaching distributed agreement. Paxos has been widely utilized in production systems and (arguably) forms the basis for many consensus systems such as Raft, Chubby and VRR. In this post, I will assume that the reader has at least some familiarity with at least one consensus algorithm (if not, I would suggest reading either Paxos made moderately complex or the Raft paper).  

For this discussion, we will consider Paxos in the context of committing commands into a log which is replicated across a system of nodes.  

The basic idea behind Paxos is that we need two phases to commit commands:

  1. Leader election – The phase where one node essentially takes charge of the system, we call this node the “leader”. When this node fails, then the system detects this and we choose another node to take the lead.
  2. Replication – The phase where the leader replicates commands onto other nodes and decides when it is safe to call them “committed”.

The purpose of the leader election phase is twofold. Firstly, we need to stop past leaders from changing the state of the system. Typically, this is done by getting nodes to promise to stop listening to old leaders (e.g. in Raft, a follower updates its term when it receives a RequestVotes for a higher term).  Secondly, the leader needs to learn all of the commands that have been committed in the past (e.g. in Raft, the leader election mechanism ensures that only nodes with all committed entries can be elected).

The purpose of the replication phase is to copy commands onto other nodes. When sufficient copies of a command have been made, the leader considers the request to be committed and notifies the interested parties (e.g. in Raft, the leader applies the command locally and updates its commit index to notify the followers in the next AppendEntries).

We refer to the nodes who are required to participate in each phase as the “quorum”. Paxos (and thus Raft) uses majorities for both leader election and replication phases.

 

The key observation

The guarantee that we make is that once a command is committed, it will never be overwritten by another command. To satisfy this, we must require that the quorum used by the leader election phase will overlap with the quorums used by previous replication phase(s). This is important as it ensures that the leader will learn all past commands and past leaders will not be able to commit new ones.

Paxos uses majorities but there are many other ways to form quorums for these two phase and still meet this requirement. Previously, it was believed that all quorums (regardless of which phase they are from) needed to intersect to guarantee safety. Now we know that this need not be the case. It is sufficient to ensure that a leader election quorum will overlap with replication phase quorums.

In the rest of this post, we will explore a few of the implication of this observation. We will focus on two dimensions: (1) how can we improve the steady state performance of Paxos? (2) how can we improve the availability of Paxos?

 

Improving the performance of Paxos

We can now safely tradeoff quorum size in the leader election phase for the quorum size in the replication phase. For example, in a system of 6 nodes, it is sufficient to get agreement from only 3 nodes in the replication phase when using majorities for leader election. Likewise, it is sufficient to get agreement from only 2 nodes in the replication phase if you require that 5 nodes participate in leader election.

Reducing the number of nodes required to participate in the replication phase will improve performance in the steady state as we have fewer nodes to wait upon and to communicate with.

But wait a second, isn’t that less fault tolerant?

Firstly, we never compromise safety, this is a question of availability. It is here that things start to get really interesting.

Let’s start by splitting availability into two types: The ability to learn committed commands and elect a new leader (leader election phase) and the ability for the current leader to replicate commands (replication phase).

Why split them? Because both of these are useful in their own right. If the current leader can commit commands but we cannot elect a new leader then the system is available until the current leader fails. If we can elect a new leader but not commit new commands then we can still safely learn all previously committed values and then use reconfiguration to get the system up and running again.

In the first example, we used replication quorums of size n/2 for a system of n nodes when n was even. This is actually more fault-tolerant than Paxos. If exactly n/2 failures occur, we can now continue to make progress in the replication phase until the current leader fails.

For different quorum sizes, we have a trade off to make. By decreasing the number of nodes in the replication phase, we are making it more likely that a quorum for the replication phase will be available. However, if the current leader fails then it is less likely that we will be able to elect a new one.

The story does not end here however.

We can be more specific about which nodes can form replication quorums so that it is easier to intersect with them. For example, if we have 12 nodes we can split them into 4 groups of 3. We could then say that a replication quorum must have one node from each group. Then when electing a new leader, we need only require any one group to agree. This is shown in the picture below, on the left we simply count the number of nodes in a quorum and on the right we use the groups as described.

Untitled drawing-2

It is the case in both examples that 4 failures could be sufficient to make the system unavailable if the leader also fails. However, with groups is not the case that any 4 failures would suffice, now only some combinations of node failures are sufficient (e.g. one failure per group).

There is a host of variants on this idea. There are also many other possible constructions. The key idea is that if we have more information about which nodes have participated in the replication phase then it is easier for the leader election quorums to intersect with replication quorums.

We can take this idea of being more specific about which nodes participate in replication quorums even further. We could extend the consensus protocol to have the leader notify the system of its choice of replication quorum(s). Then, if the leader fails, the new leader need only get agreement from one node in each possible replication quorum from the last leader to continue.

No matter which scheme we use for constructing our quorums and even if we extend our protocol to recall the leader’s choice of quorum(s), we always have a fundamental limit on availability. If all nodes in the replication quorum fail and so does the leader then the system will be unavailable (until a node recovers) as no one will know for sure what the committed command was.

Improving the availability of Paxos

So far, we have focused on using the observation to reduce the number of nodes required to participate in the replication phase. This might be desirable as it improves performance in the steady state and makes Paxos more scalable across a larger number of nodes.

However, it is often the case that availability in the face of failure is more important than steady state performance. It may also be the case that a deployment only has a few nodes to utilise.

We can apply this observation about Paxos the other way around. We can increase the size of the replication quorum and reduce the size of the quorum for leader election.  In the previous example, we could use 2 nodes per group for replication then only requiring 2 nodes for any group for leader election.

Returning to the trade off from before. By increasing the number of nodes in the replication phase, we are making it more likely that we will be able recover when failures occurs. However, we increase the chance of a situation where we have enough nodes to elect a leader but we do not have enough nodes to replicate. This is still useful as if we can elect a leader then we can reconfigure to remove/replace the nodes.

At the extreme end, we can require that all nodes participate in replication and then only one node needs to participate in leader election. Assuming we can handle the reconfiguration,  we can now handle F failures using only F+1 nodes.

 

Conclusion

Paxos is a single point on a broad spectrum of possibilities. Paxos and its majority quorums is not the only way to safely reach consensus.  Furthermore, the tradeoff between availability and performance provided by Paxos might not be optimal for most practical systems.

 

Caveats

There are as always many caveats and we refer to the paper for a more formal and detailed discussion.

  1. If this result is applied to Raft consensus, we do still need leader election quorums to intersect. This is because Raft uses leader election intersect to ensure that each term is used by at most one leader. This is specific to Raft and is not true more generally.
  2. This is far from the first time that it has been observed that Paxos can operate without majority. However, previously it was believed that all quorums (regardless of which of the phases they were from) needed to intersect. This fundamentally limits the performance and availability of any scheme for choosing quorums and is probably why we do not really see them used in practice.

Acknowledgments

During the preparation of our paper, Sugu Sougoumarane released a blog post, which explains for a broad audience some of the observations on which this work is based. 

 

`The most useful piece of learning for the uses of life is to unlearn what is untrue. ”  Antisthenes

OSCON & Code Mesh

I’m pleased to announce that I’ll be speaking at OSCON and Code Mesh in London this winter. Having heard great things about both events and looking at the lineup, I am very much looking forward to it. Drop me a line if your around in London during either event.

Title: Distributed Consensus: Making Impossible Possible

Abstract:

In this talk, we explore how to construct resilient distributed systems on top of unreliable components.

Starting, almost two decades ago, with Leslie Lamport’s work on organising parliament for a Greek island. We will take a journey to today’s datacenters and the systems powering companies like Google, Amazon and Microsoft. Along the way, we will face interesting impossibility results, machines acting maliciously and the complexity of today’s networks.

Ultimately, we will discover how to reach agreement between many parties and from this, how to construct new fault-tolerance systems that we can depend upon everyday.

Paper notes on S-Paxos [SRDS’12]

The following is a paper notes for “S-Paxos: Offloading the Leader for High Throughput State Machine Replication”. This paper was recommended to me as a example of high-throughput consensus, achieved by offloading responsibilities from the leader.

The paper starts by demonstrates that JPaxos is throughput limited by leader CPU, peaking at 70 kreqs/sec, where as the throughput of S-Paxos can reach 500 kreqs/sec.

Algorithm

The (normal case) algorithms works as follows:
  • Any node is able to receive a request for a client, lets call this coordinator
  • Coordinator sends request and its ID to all nodes
  • All nodes send ack with ID to all other nodes
  • When leader receives f+1 asks then sends phase 2a for ID
  • When leader receives f+1 successful phase 2B for ID then send commit for ID to all
  • When coordinator receives commit for ID, then execute request and reply to client
Path of request:
client -> all -> all -> all -> leader -> all -> client
1 + n + n^2 + n + n + n + 1 = n^2 + 4 + 1
or
client -> all -> all -> all -> all -> client
1 + n + n^2 + n + n^2 + 1 = 2n^2 + 2 + 1
depending on if message 2b is sent to all nodes or just leader
for comparison multi-paxos is:
client -> leader -> all -> leader -> client
1 + n + n + 1 = 2n + 2
or
client -> leader -> all -> all -> client
1 + n + n^2 + 1
depending on if message 2b is sent to all nodes or just leader
The paper proposes various optimizations such as batching and pipelining and message piggybacking to reduce network load

Evaluation

The evaluation demonstrated that under the right conditions S-Paxos can achieve 5x the throughput of JPaxos. Throughput the evaluation, graph x-axis shows number of closed loop clients, which are client who send the next request when the previous response is received. Without any indication of client latency, this did not tell us much about the rate of incoming requests. For example, the graphs in Figure 4 do need seem to be a fair comparison as # of client for Paxos and S-Paxos represents quite at different workloads.

Conclusion

I like the basic idea of this paper and its is interesting to see that latency was increased only by 1/3. However, the system places a substantially load on the network when compared to Paxos and the leader is still required to execute phase 2 paxos for each client request.

Paper Notes on PBFT [OSDI’99]

Practical Byzantine Fault Tolerance (PBFT) is the foundational paper in Byzantine consensus algorithm. Typically, distributed consensus algorithms assume nodes may fail (and possible recovery), but BFT extends this to tolerate nodes acting arbitrarily.

Assumptions

BPFT makes strong assumption about these arbitrary failures. Notability that at most f failures will occur in a system of 3f+1 nodes, this bound applies to both safety and progress guarantees. In contrast, in typical paxos style consensus algorithms the bound of f failures in 2f+1 node only applies to progress guarantee, safety will always be guaranteed even if the system is not able to make progress. BPFT makes the usual asynchronous assumption, common to many consensus algorithms.

Algorithm

Section 4.2 describes the algorithm in details, the core idea is that once the nodes have been made aware of the request for a given log index, they under go a two step process where each node notifies every one of their agreement and proceed when at least f+1 nodes in the 2f+1 responses also agree. At end of the second stage, the client can be notified. Likewise, the view change protocol is a similarly enhanced version of view change from VRR.

Performance

BPFT takes 1.5 RRTs to commit a request in the common case, all requests must reach the system via a master node and after this time, all nodes have knowledge that the request was committed. In contract, typical multi-paxos algorithms (e.g., Raft and VRR) take 1 RTT to commit a request in the common case, again all requests must go via a master node but after this time only the master node has knowledge that the request was committed. Typical leaderless consensus algorithm (e.g., single-valued paxos) take at least 2 RTT to commit a request but requests can be submitted to any node. All cases we exclude the RTT from the client to the system node and back.

BPFT differs from typical consensus algorithms in the number of messages exchanged. Whilst there exists many optimizations to piggyback messages and reduce message load, we can approximate estimate of number of messages for algorithms (this time we are including client communication). A typical multi-paxos algorithms, might send 2n+2 messages for a commit in a system of n nodes, whereas BPFT might send 1 + 2n + 2n^2, a significant difference.

Conclusions

PBFT in an interesting read and an excellent contribution to consensus. However, I am not sure how common it is that a modern datacenter might meet assumptions laid out by the paper, in particular that at most f nodes might be faulty in a system 3f+1 nodes. In this environment, nodes are usually running a same OS, configurations, software stacks and consensus implementation. Thus if an adversary is able compromise a machine then it’s likely they can compromise many, thus failures are very much dependent on each other. Lets see if the 17 years of research since has been able to improve on this.

Do you want a shed or a castle?

I have seen the error of my (programming) ways. Let me explain…

To me, programming in OCaml is like trying to build a house from just breeze blocks. It takes a long time to build even a simple shed. However. when its done, its really quite solid.

To me, programming in Go is like building a house from an array of complex pre-build components. In the blink of an eye, you have an amazing castle, complete with turrets and ornate window frames.

You open the door to your beautiful new castle and it all fails down. Each time you rebuild one part, another falls down.

You are full of regrets as you sleep in the wreckage of your fallen castle and wish for a solid shed.

Another fallen castle – rod collier [CC BY-SA 2.0 (http://creativecommons.org/licenses/by-sa/2.0)], via Wikimedia Commons

Yours truly,

Someone fighting to hold up a fallen castle

EDIT: here’s  some more example of what a falling castle looks like

Screen Shot 2016-04-28 at 14.46.46

Screen Shot 2016-04-28 at 14.52.34

Speaking at QCon London 2016

I am pleased to announce that I’ll be speaking at this year’s QCon London. I’ll be speaking in the “Modern CS in the real world” track, hosted by none other than Adrian Colyer, from the morning paper.  The abstract for my talk,  Making the Impossible Possible is as follows:

In this talk, we explore how to construct resilient distributed systems on top of unreliable components.

Starting, almost two decades ago, with Leslie Lamport’s work on organising parliament for a Greek island. We will take a journey to today’s datacenters and the systems powering companies like Google, Amazon and Microsoft. Along the way, we will face interesting impossibility results, machines acting maliciously and the complexity of today’s networks.

Ultimately, we will discover how to reach agreement between many parties and from this, how to construct new fault-tolerance systems that we can depend upon everyday.

The talk will be based upon the material from my master lecture, Reaching reliable agreement in an unreliable world. The slides for which are online and below.

Paper Notes on “CORFU: A distributed shared log” [TOCS’13]

Last Monday we looked at Tango, a system for replicating a data structure to provide linearizable semantics and fault-tolerance. Tango is built up on CORFU, a replicated log, built over storage nodes. This paper notes article covers “CORFU: A distributed shared log” also by Balakrishnan et al. from TOCS December 2013. I believe that this paper elaborates on “CORFU: A Shared Log Design for Flash Clusters” by Balakrishnan et al from NSDI 2012 and that the NSDI paper is not a prerequisite for the TOCS paper.

What we already know?

From the Tango paper, we learned that CORFU provides a replicated log with support for the following operations: append, check, read, trim and fill.

Its main two components are: a sequencer, which hands out addresses in the log and storage nodes (such as SSDs) which store log entries. Storage nodes are divided into clusters and a variant of chain replication is used between them. Each cluster is responsible for a subset of log addresses.

Summary

This paper presents CORFU, a shared log, distributed over storage nodes. It main advantage over existing systems is its scalability as its not bottlenecked by the I/O of a single host.

Local addresses -> physical addresses

Each storage node exposes a infinite write-once logical address space with read, write, delete and fill operations for each address. Delete is used when the data at a particular logical address is not longer required, its physical address can be reused but its logical address cannot. Fill is used to mark that a logical address will not be used in the future. These are implemented using a hash table with various optimizations. A seal operations is also provided which locks a storage node to operates with equal or higher epoch number.

Global addresses -> local addresses

Each client resolves a global address into a set of nodes and a local address in two stages. Firstly, the client uses a local copy of a mapping from ranges in the global addresses space to disjoint subsets of hosts. For example, addresses 0-100K map to replica sets A-C and addresses 100K-200K map to replica sets D-F. A deterministic function (like mod and div) maps the specific global address to a local address a specific replica set like A or C.

Replication

Replica are written to using client-driven chain replication, this means the client writes to each replica in a deterministic order and waits for successful acknowledgment for each storage node before continuing. As a result, write latency scales linearly with the number of replicas. In contrast, majority-based protocol like Raft, Paxos and VRR can replicate a write in as little as 1 RTT, regardless of the number of nodes, in the right conditions. The downside of such protocols is that to tolerant f failure we need 2f+1 nodes instead of f+1.

If a client fails to complete this process then it may be filled in by the next client. Like Raft, this means that clients may be given false negatives, and thus the application utilizing the log must be able to handle this. For example, Raft uses operation id’s and caches to prevent multiple application of the same operation to a state machine in SMR.

A more important failure case is where a client fails to see a committed write since the replica it is querying was removed from the replica set due to network partition/failure and the client is not aware of this change. This is address by issuing leases to the storage units from the sequencer.

Changing projections

Corfu’s sequencer is in some ways analogous to the coordinator/master in tradition protocols. Likewise, Corfu’s changing projection has many parallels with VRR’s view changes or Raft’s term changes. In all cases, a monotonically increasing value (known as the epoch number/view number/term) separates different perspectives on system configuration. Example prospectives include a period of leadership in Raft or a set of projections in CORFU. Each node stores its prospective of this value and each message between nodes includes it. Projection change is initiated by a client, then agreed by storage nodes using a Paxos-like consensus protocol and then each (involved) storage node is sealed in the process. Clients learn of the projection change when they contact storage nodes (since it includes their outdated epoch) and they retrieve the new projection information from a networked storage drive.

My interpretation of changing projections is that each projection change can include a completely new configuration. This mechanism provides us with dynamic membership, in addition to a mechanism for dealing with network failures and partitions.

Evaluation

The authors’ experimental evaluation seems very promising. In some ways, it is difficult for me to determine which gains are from the hardware and which are from CORFU’s design. All of the experiments use two replicas per cluster, thus just two failures are capability of bringing the system down. The use of client-driven chain replication means that I would really like to see how the system scales (particularly its latency) with more replicas.

Conclusions

Corfu is a very interesting system and seems to be a novel point on the solution space of distributed log solutions. Next time, we will take a look at the open source implementation, CorfuDB.