MarketUpdate 1

o

t

a

v

o

n

n I tisans AG Ar Striim data ftware Streamlio InfluxData So O SAS TIBC IBM tics platforms

n

o

i Streaming analy

p

m

a

h C essing platforms h ec tix onfluent C The highest scoring companies are nearest the companies The highest scoring EsperT Stream proc Analy Stream

xSystems r products are colour coded to ensure that readers do to coded products are colour K

SQL e ng Stream le Key: in thisvendors that all Note apples with pears. not compare best-of-breed. are consider Bullseye in each segment is calculated based on their combined based on their combined is calculated in each segment It is important that colour note to score. and overall innovation other products to relative scored been products have coded coding. with the same colour Most notably, this includes the three Hadoop distributors. have We elected not to include these vendors, primarily because their capabilities are discussed here anyway and it would be complex to compareConfluent (which provides commercial support for Kafka) with, say, , which also leverages Kafka. The exceptions to this rule are where vendors are unique supportingin a particular technology or where open source products have been leveraged as a part of a proprietary solution. Figure 1: for a score then defines a benchmark The analyst centre. ratings and all from their overall domain leading company Those that that are in the champions segment. those above segments, or Challenger in the Innovator remain are placed position The exact score. depending on their innovation l Cha “we offer a https://www. for more details); and it extends our

In effect, streaming analytics vendors tend to In that context, it important is to understand Thus streaming analytics platforms are often © 2018 Bloor various open source stream processing projects. this has still left us something in of a quandary because there a third is class of supplier, which have put together solution stacks based upon the coded these separately so that readers can compare compare that these can so readers separately coded apples with apples. As usual, there are exceptions and these are bi-coloured our in diagram. However, exclusively on analytics whereas stream processing suppliers target both data integration and analytics. In our Bullseye diagram we have therefore colour stream processing products are offering one or particularly, More solution. components a more of vendors the in former camp tend to focus platform. They are, effect, in solutions rather than platforms solutions. for offer solutions whereas the vendors supporting business intelligence and analytics tools, may have data preparation capabilities built into them, and you can often train machine learning models on the but we do not offer a streaming analytics platform per se.” integrated embedded or with tightly delivered source, mostly Apache projects. For example, as one vendor the in latter space explained to us supports that analytics platform processing streaming platforms are typically proprietary and usually product broader much a setencompass stream than processing products which are, generally speaking, more narrowly focused, and which consist of open discuss this element of stream processing our primary focus on is support for analytics. The other major difference that is streaming analytics difference is that many stream processing products stream processing thatdifference many is are targeted at data integration and data movement as well as or instead of analytics. While we will how stream processing differs from streaming analytics. The first and probably mostsignificant bloorresearch.com/technology/streaming-analytics- platforms/ reportprevious broadened our have because we platforms. stream processing include to coverage here go into the specific features that you might expect fromstreaming a analytics platform, as they were covered that in paper (see This Market Update both extends and references our Market 2016 Report on Streaming Analytics. It references it to the extent that we do not Introduction Streaming Analytics Streaming MarketUpdate 2 In our report 2016 we also included Cisco, Cisco, Microsoft and Time Compression have As far as Ancelus concerned,we is have replaced Asfar as proprietary vendors are concerned, we com (not necessarily the fount of all wisdom) you will see that InfluxData, thecompany behind InfluxDB, is generating the most interest as a time-series database and we have therefore opted to include InfluxDBin place of Ancelus. Again, this should not be considered as any aspersion on Ancelus. also leverages both Spark Streaming and Storm, as above. discussed Microsoft, Time Compression Strategies (Ancelus) and Informatica. In the case of Informatica it true is that the company has an event-process engine but this used is to support data integration rather than analytics and so has been omitted from this paper. The same consideration would apply to other data integration vendors, such as Ab Initio, that have capabilities. processing event all drawn short straws because of the number of new vendors that we wanted to include here. This should not be treated as any aspersion on their respective products and thesame true is for Oracle. It likely that is their relative positions compared to other streaming analytics platforms has remained relatively constant over the last two years and readers interested their in positioning should refer to our previous Market Report. this with InfluxDB. Thekey issue with database- oriented approaches to streaming analytics that is they support time-series.Ancelus, InfluxDB and SystemsKx all do this. However, if you look at www.db-engines. (Pulsar, Heron and Bookkeeper). would We have included Databricks had they responded to our requestsfor information but we do provide a brief discussion of relevant Spark features and, any in case, StreamAnalytix extensively leverages Spark Streaming well(as as Storm). have We excluded Samza and Storm because our research makes it clear that these products have lost traction within the marketplace. have We also omitted Gearpump as this remains only an incubator project. The one company that we might have included but have not, and which has perhaps drawn the short straw, is Lightbend, the company behind Akka and Akka Streams. have taken a similar approach as to open source projects the in sense that we have omitted products that are based on other offerings that we are Event Oracle Thus Complex neither already. covering because discussed, are Kinesis Amazon nor Processing they are both based on third-party technology (Esper and SQLStream respectively, where Esper an is open source – but not an Apacheproject – engine). have, We however, included StreamAnalytix, which built is on Esper, because it

Apache Beam an is abstraction layer for stream thereFinally, are various types of complementary As far as streaming platforms themselves are The most well-known Apache-based distributed distributed Apache-based well-known most The In terms of data flow management is this © 2018 Bloor Apache projects Thus we cover. the vendors concerned are Confluent (Kafka), data Artisans (Flink), andStreamlio are concerned our reviews are limited to the discussion of the vendors that are the main contributors to, and which provide commercial support the for, various The vendors and products and vendors The This intended is to be a representative report rather than a comprehensive one. In so far as open source products analytic processing) databases implemented via SQL engines. Hadoop on database that may be associated with stream processing. Most commonly, these are either event stores (for log and similar data) or OLAP (online processing that lets you write code against a standard API and then execute the result on any platform. (incubating), Kafka, Samza, Spark Streaming and Storm as well as the non-Apache projects Akka Streams and Gearpump (based on Akka). concerned, there are multiple choices. Offerings include the Apache projects: Apex, Flink, Heron Akka (whichopen is source but not an Apache project), Apache Flume and Apache Pulsar (which an is incubator project). messaging system Kafka. is However, while this was the space it originally started in, the technology has since has been expanded to incorporate processing and storage capabilities. Also this in category are to this Streamsets, is which also has overlapping messaging. distributed with capabilities together with Apache MiNiFi, which focuses on the collection of data at source, particularly on edge Competitive approach). agent-based an (using devices you might findin a dataintegration tool. For example, Apache NiFi typically is used to manage data flows within an (IoT) environment, Beam and assorted databases suchas Apacheand Bookkeeper. essentially similar to the sort of data flow management distributed messaging, stream processing and machine and deep learning libraries such as MLlib, MADlib and Associated Apache include technologies TensorFlow. – that can be used together to buildstreaming a analytics solution. The fundamental elements of this ecosystem comprise data flow management, Asnoted, streamprocessing platforms exist within an ecosystem – typically based on Apache projects The streamThe ecosystem processing MarketUpdate 3

“window” is available is and this provides you have problems with long running c) you cannot process individual events, a) As we have noted, we would like to have been In conjunctionwith the Kappa architecture, you have issues with events arriving out of time beyond the stream processing platform itself. For what it worth is we would comment that while Spark has both greater maturity and a much larger installed base than any of its Apache-based competitors, we believe that these vendors have a technological edge when compared to Spark Streaming. Terracotta DB. Terracotta Streaming Spark Spark Streaming has been historically been based. That is, basedon a time window. In other words, processing data asa micro-batch. The problem with this that is b) and sequence events where you have a start and end time that need to be correlated with one another. As a result, Spark Streaming not is suitable for many streaming applications. The Spark community has started to rectify this though true event based processing will not be introduced – that the is plan – until version (version 2.4 2.3 was released March in 2018). In the meantime, “structured streaming” facilities for handling things such as late arriving data. This a start is but Spark has still got some way to go. able to include an evaluation of Databricks this in report. that, Failing we have considered including an evaluation of Spark Streaming based on publicly available information. This would, however, be unfair on Spark as the commercial supporters of Flink and Heron have added additional capabilities that go being replaced by Kappa though, sadly, some vendors seem not to have woken up to this fact. vendors are increasingly deploying databases as a part of this architecture. Thus Kafka now includes serialised storage and Streamlio offers Bookkeeper conjunctionin with Pulsar and Heron. Other vendors thein open source community are relying onApache projects such as Druid, while proprietary vendors are adopting a similar approach with, for example, StreamAnalytix integrating with Kyvos (both products are developed by Impetus Technologies), IBM introducing its Event Store and Software leveraging AG , consists of a Figure 1 More specifically, thereis a clear trend towards The preceding comments have been justified only © 2018 Bloor environment, thereby significantlysimplifying any deployment. All our research suggests that Lambda is and a (database) server to unite the data from these two environments. A Kappa architecture conflates the streaming and batch pipelines into a single between Lambda and Kappa architectures. Lambda A architecture, illustrated in streaming pipeline, one or more batch pipelines, Lambda and Kappa architectures Lambda and Kappa Going beyond this the is debate – largely over – than replacements. should We also add that Striim proprietary(a vendor) also is targeting the data integration market, as well as analytics. (both of which have event processing capabilities of their own) they may be able to supplant them for particular use cases. More generally, however, they targeting new deployments (successfully) are rather integration space. While we do not think that these products are a position in to replace traditional data integration vendors such as Informatica and Ab Initio vendors extending their offerings to incorporate batch as well as stream processing and to target the data be a warning about the potential frailties of companies thein Apache space: we do not expect that DataTorrent will be the last of these to go out of business. DataTorrent, the company behind , in this report. However, May in it 2018 closed it’s doors, despite having raised over $20m funding. in This probably means that Apex a dead is duck. And it should because challenge constant. is just recently. had We originally intended to include It remains to be seen whether the market accepts this proposition but the trend clear is that only very few of the Apache projects can sit on their laurels beyond earlier products. Thus, Streamlio will claim that Pulsar a step is forward from Kafka and that Heron goes beyond Spark Streaming, Flink or Apex. that used to get a lot of press – for example, Storm – no longer do so. On the other hand, new projects continue to arise and their supporters will claim that these represent next generation capabilities that go Apache projects Evolution the in Apache space noticeable: is projects consolidation this in space. do not We believe this can continue and we expect to see acquisitions both in the open source and proprietary sections of the market. rather than place them all within a single section we have broken these down into three segments. will, We however, comment that we have not seen much (any) Market trends Market There are multiple trends within this marketplace and MarketUpdate 4 In the context of timeseries databases it between interesting divergence an There is processing. are We seeing substantial take-up of language) mark-up modelling (predictive PMML to enable the portability of models amongst the former but not from the latter. This may be to do with the fact that streaming analytics products can often train models on their platform, while stream processing offerings cannot. This not is universal: it typically is the case that only some models can be trained as well as scored in-platform but we expect the number of model types that can be so trained to also grow. We anticipate that the range of models supported by PMML will increase. As a result, it seems likely that the stream processing providers will need to extend their platforms to support in-platform training and scoring order in to compete with proprietary solutions. is worthis commenting that InfluxData not only targets environments IoT but also IT operations and infrastructure. For example, by monitoring application requests a microservices in environment you can use InfluxData to automatically scale up thenumber of containers use.in Not only this is difficult to do using Kubernetes it a use is case not being addressed by other vendors this in report. learning models and machine Algorithmic There a wide is disparity the in support for algorithmic modelling and machine learning the different by provided vendors. provider One only supports Java models, others support Python, Scala or R but not Java. It therefore is difficult to identify trends though Python and R seem to be popular. most providers of proprietary streaming analytics platforms and suppliers of open source stream ) lists www.db-engines.com Increased support for time series data also is In particular, there are more and more database © 2018 Bloor detailed this in Market Update. series databases this in report because we believe that they can genuinely compete performance – in terms – with other streaming analytics products which might or would be suitable for supporting streaming analytics. have We included both InfluxData and SystemsKx as exemplars of time 23 time-series23 databases but does not mention Ancelus, IBM Informix, InterSystems or OneTickData, for example, all of which support time series and company’s Apama streaming analytics platform. ( db-engines relevant: such as InterSystems IRIS, Splice Machine and VoltDB. In the context of this report it also is worth referencing Terracotta DB, which engine an is HTAP from Software AG, and which integrates with the targeted by both mainstream vendors Oracle, (SAP, IBM and Microsoft) but also various SQL on Hadoop engines (for example Esgyn) as well as others transactional/analytic environments processing) where you want to combine analytic processing on real-time and historic data. This market being is vendors targeting streaming analytics and stream processing environments, especially (hybrid HTAP in offerings may be appropriate deployment options. deployment appropriate offerings be may requirements processing onerous with less Readers would be well advised to consider the deployment database-orientedof solutions. are many Internet of Things (IoT) environments where this sort of scale perfectly is adequate and where suitably scalable and performant database InterSystems to process satellite telemetry data but this at is around 100,000 transactions per second rather than much faster rates. The truth that is there processing millions or tens of millions of events per second. This not is necessarily the case. As an example, the European Space Agency uses Database-oriented streaming analytics streaming Database-oriented An important point to make that is it tends to be assumed that streaming analytics involves AG acquiringAG Cumulocity, which proves connectivity to many sensors and actuators; and TIBCO acquiring Statistica and Alpine Data Labs. StreamBase respectively – have been rounding out their product portfolios by acquiring companieswith complementary Software example, For technologies. on the acquisitions being made by Software and AG TIBCO. Both of these companies – with two of the most mature products the in market with Apama and Streaming analytics platforms analytics Streaming Perhaps the most notable trend this in space based is MarketUpdate 5

“mutable” to what extent the is product at this time. As we do how amenable the is platform to “perfect” We haveWe split the scores across stream processing The scores below are solely related to the beyond structured and semi-structured data? Does it support text, voice and so forth? API support to access machine learning libraries arealso relevant. – of)(Breadth integration platform integrated with other solutions, either from the same or third-party vendors. In other words, this is part of a larger solution stack with significantcomplementary capabilities? If so, how Does the that? product is include comprehensive integration with third party provided) (or databases for storing event and other forms of data, and how good are the facilities for combining event analytics with historic data stored a data in warehouse, mart what extent or To lake? has the product been designed to play a role a larger in stack? Self-service – use by business analysts? Are there self-service and collaborative capabilities Are there built-in? visualisation capabilities provided and/oris there connectivity at the front-end to support visualisation tools such as Tableau? on the Bullseye diagram, as well as the diagrams accompanying each vendor evaluation, also encompasses company issues such as support, geographic presence, stability and so as on; well as factors like innovation and the ability to support data-driven towards a enterprise.moves We recognise that some aspects of these requirements requirements aspects these some of that recognise We will be more important for some users than others. So, while all of the scores for individual products are included the in detailed descriptions that follow later, the tables below represent the comparative scoring for each of the areas set out above. Note that each score out is of 5 but, unlike AmazonAdvisor, it is or Trip impossible to score 5 on any topic. A score of 5 would represent a not believe perfection in no product can be awarded a maximum score. platforms and streaming analytics solutions. However, Striim and InfluxData appearin both sections because, while theyboth offer solution-oriented products they also target markets (data integration and IT operations respectively) that are not specifically about streaming analytics.. have We also included EsperTech thein former category because its Esper and NEsper (the .Net version of Esper) products, although intended are engines, (CEP) processing event complex for deployment within a larger stack and thus are platform rather than solution oriented. products under evaluation. However, the positioning • • going processing? Is it processing? measures the extent measures “exactly once” “exactly how easy it is to develop how easy it is to scale the how extensive are the connectivity what platformsdoes the product run

category would be the range of data types thesupported: does product for example, extend Connectivity – options sources for as IoT well as more traditional connectivity requirements? Also within this functions such as tumbling windows, sliding functions such as tumbling windows, and so forth?windows capabilities built into the product? Does thecapabilities built into product support if the former and, or window-based event-based does it also support Does it support time windows? support and transformation data integration with batch functions? Does the product work Are there workflow as streaming data? as well cluster underpinning the solution? (Non-analytic) streaming– functionality does the platform what extent to analytics, beyond is theis process of deployment? Also, what facilities are provided to monitor streams flowing through the environment as well as the performance of the Deployment – it Is availableon? both in-cloud and on-premises? What administrative tools are available? How easy Or hundreds of thousands? Further, what is the footprint of the solution: it is suitable for deploying edge in devices or gateways? Architecture – solution? the Is platform capable of handling tens of millions of events per second? Millions? versus a proprietary a versus Data required). language, is preparation capabilities built into the platform are also relevant, as workflow. is visual development environment, development whethervisual a common IDE such as available, is and whether language training (for example, SQL applications and/or analytics using the tools (if any) that are provided? This will include considerations such as whether there a is PMML. Data preparation capabilities built into the platform are also relevant. – Development trained within the platform or only outside of it, the extent of support for models built Java, in Python, Scala, R and so forth, as well as import via as embedded functionality within the product or integration with third party tools and libraries. Issues would include whether models can be Analytics and modelling – to which the platform supports analytics, either © 2018 Bloor

• •

• •

• this report we have used the following metrics: • Scoring score theTo various vendors/products discussed in MarketUpdate 6 SERVICE ANALYTIC STREAMING ANALYTIC - - INTEGRATION BREADTH INTEGRATION SELF NON CONNECTIVITY VENDOR VENDOR VENDOR VENDOR InfluxData Streamlio Striim Confluent data Artisans EsperTech Streamlio InfluxData Confluent data Artisans InfluxData EsperTech Streamlio Striim Striim Confluent data Artisans InfluxData Streamlio EsperTech Confluent Striim data Artisans EsperTech ANALYTICS AND MODELLING AND ANALYTICS DEPLOYMENT ARCHITECTURE DEVELOPMENT

© 2018 Bloor VENDOR VENDOR VENDOR VENDOR InfluxData Striim data Artisans EsperTech Streamlio Confluent InfluxData Streamlio Confluent Striim data Artisans EsperTech EsperTech data Artisans Confluent Streamlio InfluxData Striim data Artisans Confluent InfluxData Streamlio Striim EsperTech Scores for stream processing platforms for processing stream Scores MarketUpdate 7 SERVICE ANALYTIC STREAMING ANALYTIC - - INTEGRATION BREADTH INTEGRATION SELF NON CONNECTIVITY

VENDOR VENDOR VENDOR VENDOR TIBCO InfluxData Systems Kx IBM StreamAnalytix Striim Software AG Software SAS SQLStream Software AG Software Striim IBM TIBCO Systems Kx SQLStream InfluxData StreamAnalytix Striim SQLStream StreamAnalytix SAS InfluxData TIBCO IBM AG Software Systems Kx SAS TIBCO SAS StreamAnalytix InfluxData Systems Kx SQLStream Striim IBM AG Software ANALYTICS AND MODELLING AND ANALYTICS DEPLOYMENT ARCHITECTURE DEVELOPMENT

© 2018 Bloor VENDOR VENDOR VENDOR VENDOR IBM Striim InfluxData TIBCO Kx Systems Kx SAS AG Software SQLStream StreamAnalytix SQLStream StreamAnalytix Striim InfluxData Systems Kx SAS AG Software TIBCO IBM StreamAnalytix IBM Systems Kx Striim InfluxData AG Software SAS SQLStream TIBCO InfluxData Striim TIBCO Systems Kx SQLStream SAS AG Software StreamAnalytix IBM Scores for streaming analytics platforms analytics streaming for Scores

MarketUpdate 8 20 7043 9750 www.Bloor.eu ) [email protected] 0 United Kingdom United ( LONDON 7GU N1 Web: 20–22 Wenlock Road20–22 +44 email: email: Tel: Bloor Research International Ltd Ltd International Research Bloor © 2018 Bloor you can build a solution on the other. its original developers. therefore We stand by our statement that there are no weak products this in arena: the principle choice will bebetween a solution stack on the one hand and a platform upon which Streamlio – are very new indeed. However, it tends to be the case, even with Streamlio, that its technology environments by demanding tested been has in of the proprietary vendors, It plus might EsperTech. seem that this was not true of the suppliers offering notably – products. these of Apache-based Some of the offerings havebeen available for the better part of two decades and they are as mature as you might expect. This true is for most, though not all, Conclusion There are no weak products this in market. Many MarketUpdate 9 and Application Application Operating IT-Analysis.com and and on behalf of Cambridge Cambridge on behalf of Away from work, Philip’s primary primary Philip’s from work, Away In addition to the numerous reportsIn addition to Outside of work, Daniel enjoys latin Daniel enjoys work, Outside of Daniel primarily (although by no (although by Daniel primarily IT-Director.com Development News Development System News He has also (CMI). Intelligence Market various magazines and to contributed reports a number of written published by The Financial such as CMI and companies at Philip speaks regularly Times. throughout and other events conferences America. Europe and North skiing, leisure activities are canal boats, playing Bridge (at which he is a Life and dining out. Master), databases and data warehousing, data databases and data warehousing, data master data quality, integration, data data governance, management, and metadata management, migration, data preparation and analytics. Bloor on behalf of Philip has written regularly Philip also contributes Research, to both of editor was previously where his previous experience can be put experience where his previous is increasingly but he the most use, to areas. related branching into and cooking skiing, and ballroom dancing, playing the guitar. table. Shortly afterward, Daniel left IPL to to IPL Daniel left afterward, Shortly table. researcher as a Bloor Research for work at least) is history. and the rest (so far, his alongside works means exclusively) expertise, technical providing father, perspective 'on-the-ground' the and insight both of in the form developer, a (former) of articles. and written explanation verbal DevOps, research is principally His area of

PHILIP HOWARD HOWARD PHILIP Management / Information Director Research About the authors About the HOWARD DANIEL Senior Researcher

hilip started in the computer hilip started in the computer in 1973 back way industry as worked variously and has relatively recently, in only 2014. 2014. in only recently, relatively his of the completion Following aniel started in the IT industry industry aniel started in the IT

Information management includes Information After a quarter of a century of not of a quarter a century of After In the summer of 2016, Daniel's father, father, Daniel's 2016, In the summer of

P D © 2018 Bloor that data. It involves diverse technologies technologies diverse It involves that data. to) that include (but are not limited anything that refers to the management, the management, to anything that refers of and storage governance movement, of and analysis to as access as well data, time and he is now Research Director, Director, Research time and he is now Management. on Information focused Philip working for the company as an the company for Philip working His relationship with analyst. associate that since has continued Bloor Research being his own boss Philip set up his own boss Philip set up his own being his own in 1992 and his first client was company with (then ButlerBloor), Bloor Research product management, for a variety of of variety a for product management, GPT, Marconi, including GEC companies and NCR. Raytheon Philips Data Systems, a systems analyst, programmer and analyst, a systems and as in marketing as well salesperson, experience that Daniel could bring to the bring to could that Daniel experience Philip Howard, approached him with a approached Philip Howard, be that he thought would work of piece and testing the development enriched by high standard, including a stint working at including a stint working high standard, at Nationwide. lab' 'innovation an tester at IPL (now part of Civica Group). His part Civica Group). of (now at IPL tester software manner of there included all work usually and testing, development and web a to environment and usually Agile in an Masters in Mathematics at the University of in Mathematics at the University Masters and he started as a developer working Bath, MarketUpdate 10 . . No part of this publication may be No part this publication may of . © 2018 Bloor Whilst every care has been taken in the preparation of this document to ensure that ensure that this document to in the preparation of taken has been care Whilst every For over years, 25 Bloor has assisted companies to intelligently evolve: by embracing We provide actionable strategic insight through our innovative independent innovative our through actionable strategic provide insight We We’ll show youWe’ll the future and help you deliver it. © 2018 Bloor the information is correct, the publishers cannot accept responsibility for any errors or any for responsibility cannot accept the publishers is correct, the information omissions. Research’s intent to claim these names or trademarks as our own. Likewise, company company Likewise, these names or trademarks claim to as our own. intent Research’s and the owner of with the consent reproduced been shots have graphics or screen logos, copyright. that owner’s to are subject Due to the nature of this material, numerous hardware and software products have been been products have numerous hardware and software this material, the nature of Due to these product names are the cases, of if not all, In the majority, name. mentioned by It is not Bloor the products. that manufacture the companies claimed as trademarks by Copyright and disclaimer Copyright This document is copyright Bloor Research of without the prior consent method whatsoever any by reproduced we will help you challenge assumptions to consistently improve and succeed. growth and profitability. technology to adjust their strategies and achieve the best possible outcomes. At Bloor, technology research, advisory and consulting services. assist We companies throughout their transformation journeys to stay relevant, bringing fresh thinking to complex business situations and turning challenges into new opportunities for real Bloor brings fresh technological thinking to help you navigate complex business situations, converting challenges into new opportunities for real growth, profitability andimpact. but if you do not adapt then you will not survive. So the in age of Mutable business Evolution Essential is to your success. Bloor overviewBloor Technology enabling is rapid business evolution. The opportunities are immense Bloor Research International Ltd 20–22 Wenlock Road LONDON N1 7GU United Kingdom

Tel: +44 (0)20 7043 9750 Web: www.Bloor.eu Email: [email protected]