We all know what XML is right? Just in case not, no problem here is what it is all about.
<root> <node>5</node> </root>
Now, what the computer really needs is the number five and some context around it. In XML you (human and computer) can see how it represents context to five. Now lets say instead you have a business XML document like FPML
<FpML xmlns="http://www.fpml.org/2007/FpML-4-4" xmlns:fpml="http://www.fpml.org/2007/FpML-4-4" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" version="4-4" xsi:schemaLocation="http://www.fpml.org/2007/FpML-4-4 ../fpml-main-4-4.xsd http://www.w3.org/2000/09/xmldsig# ../xmldsig-core-schema.xsd" xsi:type="RequestTradeConfirmation"> <!-- start of distinct --> <strike> <strikePrice>32.00</strikePrice> </strike> <numberOfOptions>150000</numberOfOptions> <optionEntitlement>1.00</optionEntitlement> <equityPremium> <payerPartyReference href="party2"/> <receiverPartyReference href="party1"/> <paymentAmount> <currency>EUR</currency> <amount>405000</amount> </paymentAmount> <paymentDate> <unadjustedDate>2001-07-17Z</unadjustedDate> <dateAdjustments> <businessDayConvention>NONE</businessDayConvention> </dateAdjustments> </paymentDate> <pricePerOption> <currency>EUR</currency> <amount>2.70</amount> </pricePerOption> </equityPremium> </equityOption> <calculationAgent> <calculationAgentPartyReference href="party1"/> </calculationAgent> <documentation> <masterAgreement> <masterAgreementType>ISDA2002</masterAgreementType> </masterAgreement> <contractualDefinitions>ISDA2002Equity</contractualDefinitions> <!-- populate credit support document with correct value --> <creditSupportDocument>TODO</creditSupportDocument> </documentation> <governingLaw>GBEN</governingLaw> </trade> <party id="party1"> <partyId>Party A</partyId> </party> <party id="party2"> <partyId>Party B</partyId> </party> </FpML>
That is a lot of extra unnecessary data points. Now lets look at this using Apache Avro.
With Avro, the context and the values are separated. This means the schema/structure of what the information is does not get stored or streamed over and over and over and over (and over) again.
The Avro schema is hashed. So the data structure only holds the value and the computer understands the fingerprint (the hash) of the schema and can retrieve the schema using the fingerprint.
This type of implementation is pretty typical in the data space.
When you do this you can reduce your data between 20%-80%. When I tell folks this they immediately ask, “why such a large gap of unknowns”. The answer is because not every XML is created the same. But that is the problem because you are duplicating the information the computer needs to understand the data. XML is nice for humans to read, sure … but that is not optimized for the computer.
Here is a converter we are working on https://github.com/stealthly/xml-avro to help get folks off of XML and onto lower cost, open source systems. This allows you to keep parts of your systems (specifically the domain business code) using the XML and not having to be changed (risk mitigation) but store and stream the data with less overhead (optimize budget).
Founder, Principal Consultant
Big Data Open Source Security LLC
The origins of Impala can be found in F1 – The Fault-Tolerant Distributed RDBMS Supporting Google’s Ad Business.
One of many differences between MapReduce and Impala is in Impala the intermediate data moves from process to process directly instead of storing it on HDFS for processes to get at the data needed for processing. This provides a HUGE performance advantage and doing so while consuming few cluster resources. Less hardware to-do more!
There are many advantages to this approach over alternative approaches for querying Hadoop data, including::
- Thanks to local processing on data nodes, network bottlenecks are avoided.
- A single, open, and unified metadata store can be utilized.
- Costly data format conversion is unnecessary and thus no overhead is incurred.
- All data is immediately query-able, with no delays for ETL.
- All hardware is utilized for Impala queries as well as for MapReduce.
- Only a single machine pool is needed to scale.
We encourage you to read the documentation for further exploration!
There are still transformation steps required to optimize the queries but Impala can help to-do this for you with Parquet file format. Better compression and optimized runtime performance is realized using the ParquetFormat though many other file types are supported.