Parallelization
Analysis of issues in parallel computing, especially parallelized database management. Related subjects include:
Some notes on new-era data management, March 31, 2013
Hmm. I probably should have broken this out as three posts rather than one after all. Sorry about that.
Performance confusion
Discussions of DBMS performance are always odd, for starters because:
- Workloads and use cases vary greatly.
- In particular, benchmarks such as the YCSB or TPC-H aren’t very helpful.
But in NoSQL/NewSQL short-request processing performance claims seem particularly confused. Reasons include but are not limited to:
- It’s common for databases or at least working sets to be entirely in RAM — but it’s not always required.
- Consistency and durability models vary. What’s more, in some systems — e.g. MongoDB — there’s considerable flexibility as to which model you use.
- In particular, there’s an increasingly common choice in which data is written synchronously to RAM on 2 or more servers, then asynchronously to disk on each of them. Performance in these cases can be quite different from when all writes need to be committed to disk. Of course, you need sufficient disk I/O to keep up, so SSDs (Solid-State Drives) can come in handy.
- Many workloads are inherently single node (replication aside). Others are not.
MongoDB and 10gen
I caught up with Ron Avnur at 10gen. Technical highlights included: Read more
Appliances, clusters and clouds
I believe:
- The trend to clustered computing is sustainable.
- The trend to appliances is also sustainable.
- The “single” enterprise cluster is almost as much of a pipe dream as the single enterprise database.
I shall explain.
Arguments for hosting applications on some kind of cluster include:
- If the workload requires more than one server — well, you’re in cluster territory!
- If the workload requires less than one server — throw it into the virtualization pool.
- If the workload is uneven — throw it into the virtualization pool.
Arguments specific to the public cloud include:
- A large fraction of new third-party applications are SaaS (Software as a Service). Those naturally live in the cloud.
- Cloud providers have efficiencies that you don’t.
That’s all pretty compelling. However, these are not persuasive reasons to put everything on a SINGLE cluster or cloud. They could as easily lead you to have your VMware cluster and your Exadata rack and your Hadoop cluster and your NoSQL cluster and your object storage OpenStack cluster — among others — all while participating in several different public clouds as well.
Why would you not move work into a cluster at all? First, if ain’t broken, you might not want to fix it. Some of the cluster options make it easy for you to consolidate existing workloads — that’s a central goal of VMware and Exadata — but others only make sense to adopt in connection with new application projects. Second, you might just want device locality. I have a gaming-class PC next to my desk; it drives a couple of monitors; I like that arrangement. Away from home I carry a laptop computer instead. Arguments can be made for small remote-office servers as well.
| Categories: Cloud computing, Clustering, Data warehouse appliances, Exadata, NoSQL, Software as a Service (SaaS) | 8 Comments |
Hadoop execution enhancements
Hadoop 2.0/YARN is the first big step in evolving Hadoop beyond a strict Map/Reduce paradigm, in that it at least allows for the possibility of non- or beyond-MapReduce processing engines. While YARN didn’t meet its target of general availability around year-end 2012, Arun Murthy of Hortonworks told me recently that:
- Yahoo is a big YARN user.
- There are other — paying — YARN users.
- YARN general availability is now targeted for well before the end of 2013.
Arun further told me about Tez, the next-generation Hadoop processing engine he’s working on, which he also discussed in a recent blog post:
With the emergence of Apache Hadoop YARN as the basis of next generation data-processing architectures, there is a strong need for an application which can execute a complex DAG [Directed Acyclic Graph] of tasks which can then be shared by Apache Pig, Apache Hive, Cascading and others. The constrained DAG expressible in MapReduce (one set of maps followed by one set of reduces) often results in multiple MapReduce jobs which harm latency for short queries (overhead of launching multiple jobs) and throughput for large-scale queries (too much overhead for materializing intermediate job outputs to the filesystem). With Tez, we introduce a more expressive DAG of tasks, within a single application or job, that is better aligned with the required processing task – thus, for e.g., any given SQL query can be expressed as a single job using Tez.
This is similar to the approach of BDAS Spark:
Rather than being restricted to Maps and Reduces, Spark has more numerous primitive operations, including map, reduce, sample, join, and group-by. You can do these more or less in any order.
although Tez won’t match Spark’s richer list of primitive operations.
More specifically, there will be six primitive Tez operations:
- HDFS (Hadoop Distributed File System) input and output.
- Sorting on input and output (I’m not sure why that’s two operations rather than one).
- Shuffling of input and output (ditto).
A Map step would compound HDFS input, output sorting, and output shuffling; a Reduce step compounds — you guessed it! — input sorting, input shuffling, and HDFS output.
I can’t think of much in the way of algorithms that would be logically impossible in MapReduce yet possible in Tez. Rather, the main point of Tez seems to be performance, performance consistency, response-time consistency, and all that good stuff. Specific advantages that Arun and I talked about included:
- The requirement for materializing (onto disk) intermediate results that you don’t want to is gone. (Yay!)
- Hadoop jobs will step on each other’s toes less. Instead of Maps and Reduces from unrelated jobs getting interleaved, all the operations from a single job will by default be executed in one chunk. (Even so, I see no reason to expect early releases of Tez to do a great job on highly concurrent mixed workload management.)
- Added granularity brings opportunities for additional performance enhancements, for example in the area of sorting. (Arun loves sorts.)
| Categories: Databricks, Spark and BDAS, Hadoop, Hortonworks, MapReduce, Workload management, Yahoo | 14 Comments |
Key questions when selecting an analytic RDBMS
I recently complained that the Gartner Magic Quadrant for Data Warehouse DBMS conflates many use cases into one set of rankings. So perhaps now would be a good time to offer some thoughts on how to tell use cases apart. Assuming you know that you really want to manage your analytic database with a relational DBMS, the first questions you ask yourself could be:
- How big is your database? How big is your budget?
- How do you feel about appliances?
- How do you feel about the cloud?
- What are the size and shape of your workload?
- How fresh does the data need to be?
Let’s drill down. Read more
NuoDB marketing mishegas
I must start by apologizing for giving a quote in a press release whose contents I deplore. Unlike occasions on which I’ve posted about inaccurate quotes, in this case the fault is mine. The quote is quite accurate. And NuoDB didn’t mislead me about the release’s contents; I just neglected to ask.
NuoDB evidently subscribes to the marketing fallacy:
- Big DBMS companies hit people repeatedly with marketing cudgels.
- We want to be a big DBMS company.
- Therefore we will hit people repeatedly with marketing cudgels too.
But to my taste, NuoDB’s worst travesty is not the deafening drumroll before launch (I asked off their mailing list months before), nor the claim that NuoDB’s launch would be a “big day” for the database industry (annoying but ordinary hype), nor the emergent flock of birds foofarah, nor even NuoDB’s overwrought benchmark marketing (distressingly many vendors do that).
Rather, I think NuoDB’s greatest marketing offense to date is its Codd-imitating “12 rules” for cloud database management. Read more
| Categories: Cloud computing, Clustering, NuoDB, Theory and architecture | 9 Comments |
Introduction to NuoDB
NuoDB has an interesting NewSQL story. NuoDB’s core design goals seem to be:
- SQL.
- Transactions.
- Very flexible topology, including:
- Local replicas.
- Remote replicas.
- Easy deployment and management.
| Categories: Cache, Cloud computing, Clustering, Database compression, NewSQL, NuoDB | 5 Comments |
Introduction to GenieDB
GenieDB is one of the newer and smaller NewSQL companies. GenieDB’s story is focused on wide-area replication and uptime, coupled to claims about ease and the associated low TCO (Total Cost of Ownership).
GenieDB is in my same family of clients as Cirro.
The GenieDB product is more interesting if we conflate the existing GenieDB Version 1 and a soon-forthcoming (mid-year or so) Version 2. On that basis:
- GenieDB has three tiers.
- GenieDB’s top tier is the usual MySQL front-end.
- GenieDB’s bottom tier is either Berkeley DB or a conventional MySQL storage engine.
- GenieDB’s bottom tier stores your entire database at every node.
- If you replicate locally, GenieDB’s middle tier operates a distributed cache.
- If you replicate wide-area, GenieDB’s middle tier allows active-active/multi-master replication.
The heart of the GenieDB story is probably wide-area replication. Specifics there include: Read more
| Categories: Cache, Cloud computing, Clustering, GenieDB, Market share and customer counts, MySQL, NewSQL | 4 Comments |
NewSQL thoughts
I plan to write about several NewSQL vendors soon, but first here’s an overview post. Like “NoSQL”, the term “NewSQL” has an identifiable, recent coiner — Matt Aslett in 2011 — yet a somewhat fluid meaning. Wikipedia suggests that NewSQL comprises three things:
- OLTP- (OnLine Transaction Processing)/short-request-oriented SQL DBMS that are newer than MySQL.
- Innovative MySQL engines.
- Transparent sharding systems that can be used with, for example, MySQL.
I think that’s a pretty good working definition, and will likely remain one unless or until:
- SQL-oriented and NoSQL-oriented systems blur indistinguishably.
- MySQL (or PostgreSQL) laps the field with innovative features.
To date, NewSQL adoption has been limited.
- NewSQL vendors I’ve written about in the past include Akiban, Tokutek, CodeFutures (dbShards), Clustrix, Schooner (Membrain), VoltDB, ScaleBase, and ScaleDB, with GenieDB and NuoDB coming soon.
- But I’m dubious whether, even taken together, all those vendors have as many customers or production references as any of 10gen, Couchbase, DataStax, or Cloudant.*
That said, the problem may lie more on the supply side than in demand. Developing a competitive SQL DBMS turns out to be harder than developing something in the NoSQL state of the art.
Data(base) virtualization — a terminological mess
Data/database virtualization seems to be a hot subject right now, and vendors of a broad variety of different technologies are all claiming to be in the space. A terminological mess has ensued, as Monash’s First and Third Laws of Commercial Semantics are borne out in spades.
If something is like “virtualization”, then it should resemble hypervisors such as VMware. To me:
- The core feature of a hypervisor is that it allows many somethings to run and coexist where ordinarily only one something would come into play. Here the “many somethings” are virtual machines and what’s going on inside them, and the “one something” is the ordinary operating system/hardware computing stack.
- A core feature of original VMware was that the “many somethings” could be quite different — for example, the operating environments of numerous different hardware systems you wanted to decommission, or of new systems that you didn’t want to buy quite yet.
- Important features of hypervisors include:
- The ability to have multiple virtual machines run side by side at once, safely.
- Flexible and powerful workload management if the virtual machines do contend for resources.
- Easy management.
- The negative feature of having sufficiently low overhead.
Anything that claims to be “like virtualization” should be viewed in that light. Read more
| Categories: Clustering, Data integration and middleware, ScaleDB, Theory and architecture, Transparent sharding | 5 Comments |
Spark, Shark, and RDDs — technology notes
Spark and Shark are interesting alternatives to MapReduce and Hive. At a high level:
- Rather than persisting data to disk after every step, as MapReduce does, Spark instead writes to something called RDDs (Resilient Distributed Datasets), which can live in memory.
- Rather than being restricted to maps and reduces, Spark has more numerous primitive operations, including map, reduce, sample, join, and group-by. You can do these more or less in any order. All the primitives are parallel with respect to the RDDs.
- Shark is a lot like Hive, only rewritten (in significant parts) and running over Spark.
- There’s an approach to launching tasks quickly — ~5 milliseconds or so — that I unfortunately didn’t grasp.
The key concept here seems to be the RDD. Any one RDD:
- Is a collection of Java objects, which should have the same or similar structure.
- Can be partitioned/distributed and shuffled/redistributed across the cluster.
- Doesn’t have to be entirely in memory at once.
Otherwise, there’s a lot of flexibility; an RDD can be a set of tuples, a collection of XML documents, or whatever other reasonable kind of dataset you want. And I gather that:
- At the moment, RDDs expire at the end of a job.
- This restriction will be lifted in a future release.