00:02good afternoon everybody thank you for
00:04being here last session of the day so
00:07that's great I love Boston I'm very
00:11happy to be here today and I have two
00:14reasons for that the first reason is
00:16that Boston was the first city in the
00:18United States of America visited 20
00:20years ago I came here to study English
00:22for a month and hopefully all of you can
00:25notice my Boston accent second reason
00:29and a little bit more important is this
00:31today we're introducing hyperscale for
00:34up Anna stack it's a struggle defining
00:36and storage solution that has been built
00:38for OpenStack cloud today we're going to
00:41cover how hyper scale for openness that
00:44can provide quality of service
00:46performance and data protection bill on
00:49software that can be used with the
00:51commodity Hardware of your choice to
00:54help customers to move more loads
00:56into OpenStack clouds my name is Carlos
01:01Guerrero and a produced man a year
01:02working for very terse for the last 16
01:06today I have the pleasure to have with
01:07me Ruth Harrison CEO for Fairbanks one
01:12of the key partners in Netherlands they
01:14are OpenStack and bass udders in
01:16Netherlands they have been working with
01:19us on all this way providing feedback
01:21and validation of what we're doing so I
01:23will introduce you in a while before I
01:26want to introduce my colleague and also
01:28friend a busy day vice-president for
01:31engineering just because you have been
01:33two years more than me in the company
01:35when you cover a little bit about you
01:37know what's very tells what's the
01:40division and what are we doing here okay
01:43thanks Carlos yeah I believe some of you
01:46know me at most of you know me been in
01:49this company for very long time actually
01:5018 months 18 years sorry 18 years
01:53I wish 18 months yes yet so you know we
01:58will present the company but you know
02:00today where it has has emerged from you
02:04know we used to be acquired by Symantec
02:05one year back we got freedom we are run
02:09by world's biggest equity investor
02:11carline investment in a start-up mode
02:15it's great for us who built the original
02:17startup called visitors to kind of come
02:19out into a stealth mode startup now and
02:22it gives us opportunity again re-emerge
02:25as the storage and data protection
02:28leader that we used to be right so today
02:30the company is organized into really
02:31three broad sections data management or
02:36storage management that we proudly do
02:38for the last 25 years we did the first
02:41file system volume manager the industry
02:44had seen over unix the second portfolio
02:47is the data protection or backup backup
02:51is our flagship product today we call it
02:54net backup and then the third portfolio
02:58is the data insight data insight ranges
03:02across the solutions that we have around
03:05data governance data insight looking at
03:09data visibility of the data that's
03:10really what Veritas is it's very simple
03:12data management data protection and data
03:14governance right so what what the heck
03:18happened you had to click on the right
03:19bottle that's the right okay cool so
03:23quickly on our history right because
03:25it's it's from the past you know how to
03:27learn the future we have been you know
03:29you know wedding in storage management
03:31for a long time first volume manager
03:33UNIX world had seen first file system
03:36truly POSIX based file system which we
03:39still ship called cluster file system
03:41and then the first h.a or VCS solution
03:44the industry had seen back in the days
03:461999-2000 moving on to that we are
03:49adopting and emerging and reshaping
03:51ourselves to do multi cloud customers
03:53are going to go to multi-cloud no matter
03:56what that's the next world we are going
03:58to see for many years and then
04:00software-defined storage software
04:02different storage is a term industry I
04:04used last for five years but we actually
04:06define software-defined before it was a
04:09cool term really if you look at volume
04:12manager file system this is a way of how
04:14do you provision storage managed storage
04:17in a software-defined way so we are
04:19emerging there right that's really where
04:21we are coming into OpenStack OpenStack
04:24is a great venture it's a in front I
04:28it's a great infrastructure for us to
04:31add values in storage and backup the
04:33things that we do and we do better than
04:35many people right many other vendors
04:38right and we are trying to add value
04:40this is a great opportunity for us to
04:41really add value in a more open
04:43architecture agnostic way we are
04:45checking in all the sources to open
04:47source so this is a great idea
04:48personally OpenStack this is my 9th or
04:5110th whatever number of OpenStack has
04:53happened it's it's almost like a
04:55festival or religions thing for me to
04:57show up and talk in this so I'm
04:59extremely honored to be now with Carlos
05:01and root to talk about and launch the
05:04product on stage in OpenStack summit
05:07Boston right so with that I'll let talk
05:10about his company Thanks
05:12ok I would you thank you good afternoon
05:15ladies and gentlemen my name is root
05:17Harmsen I'm the CEO of fairbanks and we
05:19are a Dutch IT year company and we are
05:23in the open-source business now since
05:252011 and the reason we choose for
05:28open-source is because it gives
05:30customers ability to innovate give them
05:32flexibility and scalability and also
05:34cost savings and we see a lot of
05:38examples like uber and air B&B what can
05:44give them a competitive advantages by
05:46innovative and flexible IT solutions and
05:49that's the reason we are in this
05:51business in the last couple of years we
05:56built a lot of clouds for our customers
05:58and we did it with several distributions
06:01but now we have a very very good
06:04relationship with canonical and we build
06:07our clouds with Ubuntu and canonical and
06:10in 2014 I received a phone call from a
06:17friend of mine within Veritas and he
06:20found me and he said Ruth before I going
06:24you have to sign an NDA and at that time
06:27I thought wow if somebody calls me in
06:29the evening and wants to explain me
06:32something and I have to sign an NDA it
06:33should be important so I received the
06:36NDA I signed it in a couple of days
06:37later I went in a meeting with with
06:41Veritas and I am really very proud that
06:46we as fairbanks have the opportunity to
06:50help Veritas develop hyperscale
06:53as where it is right now and yesterday
06:56it was the launch of a real very very
06:59good product to work in the open source
07:04community you have to add value and we
07:08do that in several ways we have partners
07:11some partners in Europe who don't have
07:15the expertise of OpenStack and we help
07:18them developing and improving ICT and
07:21marketing concepts to bring their
07:23business forward we also have customers
07:26where we act as a cloud specialist and
07:29broker of modular and open source cloud
07:32solutions and of course we have our end
07:34users who we help and build our clouds
07:39the target market for for fairbanks we
07:43have service providers and end users and
07:45as I said before you can see the picture
07:48for our server providers our customers
07:50we developed a service extender which
07:53means that above on his services we can
07:56help them to improve their business and
07:58of course our end users where we build
08:01an internal of external private cloud
08:04for the customers in the Netherlands but
08:06also in Europe good well thank you I
08:11think it's great integration and again
08:14thank you very much for being with us
08:15the reason all this all this time so we
08:17take a look to what problems are we
08:19solving for up in a stack we concentrate
08:21on three key areas the first one is that
08:23up we have to change the mindset about
08:27thinking about the story to the whole of
08:29the story it's our volume that you go
08:31and use and you have to think more about
08:33the workloads so we charge the specific
08:35needs you have the workloads you can
08:37have things like MySQL you can have
08:40things like Cassandra do they need the
08:41same requirements so typically what you
08:44find is that it's one size fits all so
08:46you ain't having maybe three copies for
08:49everything do you need three copies for
08:51your analytic floor role which is
08:52running on memory and you have to be
08:55based on memory and then having another
08:56three copies spread 3 copies play three
08:59copies and it's just about the number of
09:01copies it's about how you are using the
09:03i/o which is go to the next point is
09:06that how do you really get the
09:08performance that you need and definitely
09:10there is some progress and working on
09:12the community about quality of service
09:14okay but how you can guarantee the
09:16quality of service that is defined for
09:19your specific workload and make sure
09:22that the end you get the mass of your
09:23hardware because you can be using
09:25commodity hardware but how do you make
09:27sure you get the base of that anything
09:31here we're going to be talking about the
09:32architecture and the mindset so we do to
09:34introduce a little bit of the history
09:36three years ago we started this project
09:38called open flame the today's hybrid
09:40scale okay and that hyper scale was
09:42mindset change about we have to do
09:45something different in how we provide
09:47this and this is the one of the key
09:49things is that change the mindset about
09:51the workload because that is what really
09:54ok and when we think about that the
09:57third point which is not very well solve
10:01or unsolved is how do you protect those
10:03clothes workloads what is the protection
10:06we as very test as the backup company
10:08it's key for us and we will only keep
10:10notes that more customers or moving
10:12workloads into production that's clear
10:15because more customers are asking us
10:17about how do I protect my workload if
10:19you need protection that means it
10:22matters for you so to solve this we are
10:25introducing hyper scale for upon a stack
10:28the first distribution that we announced
10:30yesterday is based on canonical and
10:32really appreciate canonical all the help
10:34to to go da first in our NDA is to be
10:38agnostic and of course we're working
10:41with other vendors we're working very
10:42closely with Red Hat and helping a lot
10:45we are planning to be with them by the
10:48end of the year and we have suze we have
10:51mentees it's our NDA so I believe talked
10:53about so we're defining a storage we did
10:56that agnostic to any hard work to any
10:58store eternal network to any cloud and
11:00of course we have to be agnostic to any
11:02technology that you guys may want to use
11:05that's in our NDA so what we really
11:08wait Hyper a scale for up in a stack so
11:11hyper a scale for up on a stack is still
11:13with the finest storage that is going to
11:15allow you to define which is the
11:17resiliency level that you need for your
11:19workloads maybe you need three copies
11:22maybe you need to support to compute
11:24failures one compute failures that's
11:26something you are going to define at the
11:29workload level the second point is that
11:32how you can get the predictable
11:35performance with quality of service
11:37there is a term very well known in the
11:40market which is a noisy neighbor of
11:42course we're solving the noisy neighbor
11:45but not only that that's we can
11:47guarantee a minimum miles per second so
11:50we're going to see a little bit our
11:51architecture how it works to make sure
11:54that at the end your workloads is going
11:56to get the performance that I need
11:58because it's a cloud environment so I
12:01don't know what instances are going to
12:02be running and where they're going to be
12:04running so I had to make sure I create
12:07mechanisms within my storage so that my
12:10workload can run whatever they want but
12:12they're going to get the performance
12:14that they need and the third point is
12:17the backup integration or not backup I
12:20call data management without impact well
12:23what I see is an architecture that
12:25really separates the performance for the
12:28data management and it's going to help
12:31you to really do a what we call zero
12:33backup window or run a backup whatever
12:36you want right so before jumping into a
12:40bigot if we take a look to what we can
12:43compare with that traditional so we
12:46define a storage architectures that are
12:48based on a single layer okay is that
12:50where you have your hyper converse the
12:52server's the storage the memory right
12:55and we have been there I mean again we
12:57create our fastest than 1995 1920 m4 we
13:02have cluster file system the last
13:03version for cluster falsus then that as
13:05you guys I love support after 128 nodes
13:09using commodity Hardware a local storage
13:12but the problem with that is that
13:14everything is on the same place so you
13:18have this noisy neighbor problem where
13:20all the instances are going to be
13:22and you have to make sure you fix that
13:24because you don't know what they're
13:26going to be running the second issue is
13:28that everything is on the same plane
13:30what I mean with that is the copy that I
13:34need now and the copy I need for
13:37resiliency are happening on the same
13:39place and if I go to typical things like
13:42one size fits all where you get three
13:45copies for everything it's not about the
13:47number of three copies because you're I
13:49was talking yesterday is that the number
13:51of i/os that I have to do because I make
13:54a right and then I make nine right but
13:57what I do those nine rights on the same
13:59way here so everything is there so we
14:02took a step back I think it was three
14:04years ago and say we have to rethink how
14:07we make this thing okay so I think that
14:10was a challenge for a reigning area that
14:12was really well presented and a bit I
14:15can to let you cover sure what you guys
14:17have to do thanks guys and you know
14:19interestingly what you said it appeared
14:22to me again and again that if you have
14:24multiple copies it's not a bad thing
14:26right necessarily because you need
14:27copies for resiliency and H a purpose D
14:30our purpose so you need copies but point
14:32is when you go into hyper-converged or
14:34cloud computing right you are mostly
14:37going to be using flash for your local
14:39computer storage now we have this copies
14:41of the data so many times in the flash
14:44storage and that's going to be also very
14:46expensive because you're using
14:48two-thirds of your data center on flash
14:50storage which are just simply copying
14:52the primary data and that's what we
14:54wanted to rethink right and so you know
14:56and thanks Carlos that is a great
14:58introduction so if you really look at it
15:00we thought about this for 20 years and
15:03we challenged our self to say well for
15:06none of this storage architecture which
15:08we have today in our legacy and also
15:09many other vendors today we will scale
15:11because you are stuck with one plane
15:14which is called the compute plane where
15:16applications are running like where Akal
15:17or CouchDB and where you take the backup
15:20right data really has two layers right
15:23one is the primary consumption which is
15:25applications run that's the primary
15:26usage we call it primary data the data
15:30has the secondary consumption which is
15:32backup ETL archiving these are secondary
15:36the problem with our architecture
15:37traditionally has been we morphed this
15:38secondary and primary data into one
15:40plane so what we did was very simple now
15:43that we have it and we are under NDA
15:45said we introduce another data plane we
15:48call it a data plane as you can see the
15:50top is the compute plane which is a
15:53scale out compute plane which serves up
15:55the application I ops and the lower one
15:58is the data plane which is here right
16:01here which is also storing the data
16:04right but what we do here is by doing
16:06that like carlos's quickly we can now
16:09look at solving this problems of quality
16:11of storage because now we have one
16:13compute to worry about one plane to
16:15solve where multi-tenant applications
16:17are happening we can look at optimizing
16:19the storage consumption because we can
16:21push off all those extra copies to a
16:23load clear secondary storage data plane
16:25we can look at how do we do higher
16:29density storage and finally how do we do
16:31data protection because data protection
16:33has to be thought of now that we have
16:35two claims one is the primary compute
16:38and the secondary secondary data plane
16:40we can actually rethink the whole
16:42concept of data protection where we can
16:44say you know what we can just take that
16:46backup of the data from the secondary
16:48plane we don't have to impact the
16:50applications which are running in the
16:51primary plane to do backup that's really
16:54the IP here right this is very
16:56disruptive and we have protected this
16:58for almost two and a half years except
17:00with partners and chosen customers that
17:03we have got a lot of feedback there are
17:05customers among you who have given us a
17:07lot of I mean I can see three and other
17:08people who have given us a lot of
17:10feedback on how do we do better we have
17:12definitely morphed and today we were
17:14ready to be GA so we released the
17:16product so I'll move on yeah is that so
17:21what are the key things is when talking
17:24about resiliency is that donor how many
17:27of you are familiar with their OpenStack
17:28flavor I've been talking to some people
17:31or maybe not familiar with that but
17:32that's what you define your policies
17:34your essays the tying off machine that
17:36you need that is where like in Amazon
17:38you say I want this number of CPUs and
17:41in this amount of memory but what about
17:43the storage with the storage you today
17:46just defined I want changing about what
17:49to have a real integration is extend on
17:52the flavors to say I need this
17:55resiliency factor so this is the
17:57protection level I want to get and this
17:59is the number of copies I want to have
18:01because again my SQL is going to be
18:04different from Cassandra so like I can
18:07define different flavors hyper SQL is
18:10going to define by default what we call
18:12a bronze and silver and gold very
18:16magnetic names right so but you can
18:19define your own flavors or you can add
18:21the the properties to the capstone
18:24flavors that you want so you can create
18:26your Cassandra flavor if you want and
18:28defined that an immediate need of this
18:32is the reduce on the hard word needs so
18:35we take a look to the economics and you
18:37take a look remember that single layer
18:39right when I have my first copy my
18:42second copy my third copy my second and
18:44my third copy if I need a fair copy are
18:47going to be welcome to move that down to
18:49the data plane the data plane is going
18:51to be high resiliency and cheaper
18:53storage and the third thing is this is
18:57based on commodity hardware ease so we
18:58defined a store it's based on commodity
19:00Hardware of your choice so it's not this
19:04is my software solution and you have to
19:06buy my servers right now this is really
19:08customized so you choose whatever you
19:11want and we are going to be talking
19:12about the architectures a little bit so
19:15describe a little bit that how it works
19:19so this is the compute plane we have
19:21used this traditionally in any storage
19:22computing right this is the computer in
19:24which runs your virtual machines or
19:25applications right so what we need this
19:28is OpenStack the instances are the Nova
19:31virtual machines which are running your
19:32applications workloads right there's
19:34nothing new here now we have extended
19:36the flavor like Carla's mentioned about
19:38OpenStack flavors OpenStack gives you
19:40quality of service in our way right it
19:41gives you bronze gold silver which can
19:44be extended we did that and then we said
19:46you know when application rights are
19:47coming the primary the primary data is
19:50in this node right which actually gets
19:53the application and then we replicate
19:55the Delta changes we do not copy the
19:58entire file over and over again into
20:00this peer computes that's what the
20:02traditional solutions would be
20:03that's how you end up with multiple
20:05copies all we do is we take the Delta
20:07changes from the last 15 minutes or so
20:08which is a tunable and you say let's
20:10only keep the last rights right into the
20:14peer compute nodes that's the resiliency
20:16level but then you'd ask them well you
20:18don't really have a full copy of the
20:20file anywhere right that's the next
20:22question that's where we said you know
20:23what we can disrupt this we can say well
20:26you can you can use this data you can
20:29store this data from the primary compute
20:32plane every episodic interval we call it
20:35episodic data sync but every episodic
20:37interval you can say well I can store
20:39this flash them to a secondary storage
20:40now the beauty of this design is once
20:42you are done flushing them into the
20:44secondary storage you can remove those
20:46deltas or replicas that you had in the
20:48primary computer so you end up with one
20:51primary copy in your flash tier and one
20:54secondary copy in your chip and deep
20:56secondary storage tier that's really the
20:58solution right now if you really look at
21:01it this is all works with OpenStack of
21:03course we had to upgrade the cinder
21:06to look at local flash storage we had to
21:09update the Nova to look at the filters
21:11that Nova provides to upgrade them to
21:13use this design so think of this like us
21:15to enter into its storage management for
21:18OpenStack right now this is a storage
21:21technology but we call it also end to
21:23end because now that you have the data
21:25in your secondary node right you can
21:27take a backup simply from that node
21:29that's really the architecture so see
21:31the other thing to point out is that
21:33this four nodes we are showing four here
21:35these are scalable infinitely scalable
21:37because there is no strong clustering
21:39between these nodes they are not
21:41connected by a heartbeat Network like
21:42other solutions even our legacy
21:44solutions had right there layer 3
21:46network each node comes and goes you can
21:49infinitely infinitely scale the compute
21:52node you can infinitely scale the data
21:55node so your storage needs
21:57don't dominate your application grows
21:59faster application needs the storage
22:01does not have to grow that way they can
22:02scale linearly right so I'll pass over
22:06to root to talk about thank you
22:09well you heard something about
22:11resiliency and quality of service but
22:14one of the first slides of
22:17this presentation you also saw something
22:19about performance and this is a real
22:22case where we were dealing with a few
22:24years ago this is a Dutch supermarket
22:27it's called the spar and they came to us
22:29and said Fairbanks can you help us
22:32building a cloud where we can have a ops
22:35and a lot of higher-ups because they
22:37build an application on an Oracle
22:41database and what they what they do is
22:44and that's their unique selling point
22:46there is a guy who was working in the
22:48supermarket at the time there is a
22:50product he the the shelf is empty he
22:55pushed the button the Gauss information
22:58to the warehouse there is a guy who
23:00picks the products put it into a
23:03container container into the lorry lorry
23:05to the store and they build it very
23:08sufficient very high qualified but if
23:12you look about for elastic clouds then
23:16you definitely need a performance so we
23:20did a lot of testing and at the end we
23:22had to decide and not to use chef
23:25because Saif was not able to solve this
23:27issue and they had to go back to the
23:30standard solution likes and technology
23:32and because of the the stress test we we
23:38did for this customer you saw with the
23:41elastic cloud necessary they were
23:43looking for it was not possible to
23:45Yousef and we really believe that
23:48together with the product like hyper
23:49scale we definitely can solve solutions
23:52for our customers in the market I will
23:58thank you then we will make sure we get
24:01that so that goes to their predictable
24:03performance with quality of service do
24:05we see that and and we have seen
24:07different solutions all on the market is
24:10that do you have quality of service that
24:12you can define or each of the flavors or
24:15each of the workloads again how I can
24:17have critical database running today in
24:20OpenStack and I can make sure I can run
24:23another word roads like web servers
24:25application servers and everything is
24:27going to get that so using my
24:30artistical skills I try to represent
24:32that with different regions where you
24:34can go and share and say this is I have
24:36this workload that need these
24:39I have these other workloads that need a
24:41different performance one of the things
24:43we do by default is to avoid the noisy
24:45neighbor so that means is that now any
24:48instance is going to keep under control
24:50and it's going to is not going to make
24:52any noise that it's going to affect the
24:54performance of other ones and one of the
24:56key points is accelerate performance so
24:59what we do is that we keep that first
25:01copy in the compute plane which is
25:03running where your workload is there so
25:05the CPU and the story chart together we
25:08also use SSDs to our flash to accelerate
25:12the performance to accelerate writes and
25:14do that protection that the bigot was
25:16talking about and make sure that that
25:17that right is protected but it's
25:19protected in a very effective way so
25:26let's see the how part Carlos gets to
25:30tell me what to do and I get to do the
25:32how part of it with my team some of you
25:35are here so get as well so basically so
25:38basically what again just listen to me
25:41will the slides cannot explain this is a
25:43pretty complicated algorithm but
25:44OpenStack actually you have to mean it
25:47inside the algorithm here yes OpenStack
25:49introduces quality of service by adding
25:51flavors to the virtual machine when you
25:53create or instantiate a virtual machine
25:55these are called gold silver bronze with
25:57the each tier you get a performance
25:59guarantee but then OpenStack does not
26:01really solve the storage problem as an
26:03IAS provider it's not open stacks job
26:05it's vendors job to look at it but
26:07OpenStack tries to solve it QoS or
26:11quality of service and the source level
26:12at the disk and the source level but the
26:14virtual machine issues the i/o what
26:16really needs to be done is to solve the
26:18storage quality of service at the
26:20devices in we call it the destination
26:23what the data has been written so what
26:25we do is we simply assign the
26:29application a sort of credit system we
26:32say how prioritize is the application is
26:34it my Oracle or is it a noisy neighbor
26:35the noisy neighbor is not prioritized
26:37the Oracle gets prioritized we take that
26:40weights or credits we call it and we
26:44and in the system then we go back to the
26:46device and we understand the device
26:48latencies for that application then we
26:50map that too and then we say who gets to
26:52write if the Oracle came with a very
26:54high max max I ops and mini ops a higher
26:57priority with the gold level will let
26:59the Oracle flow in and the noisy
27:01neighbor automatically does not get the
27:03share of the storage its deprioritized
27:05that's really the quality of service now
27:07this is something of course happens
27:09within a physical node you have a
27:11multi-tenant physical node you can say
27:13well I'll assign priorities and do
27:14quality of service between the workloads
27:16that's how we do quality of service our
27:18vision is to also take that and map that
27:20into inter compute plane where you can
27:22do quality of service across physical
27:24nodes that's coming up later in the
27:26roadmap so with that I'll let Carlos or
27:30do talk about Carlos so this is quality
27:34of service on an action and I understand
27:36it's a very complex it's like let me go
27:39through the key things this is think
27:41about your database running where you
27:43have to find your flavors you want to
27:45minimize per second top 10,000 and a
27:47maximum of 20 thousand hours per second
27:49you have some application servers
27:52between 5 and 10 K and what happened
27:55here is that I start my database running
27:57it goes to the limit of 20 K I have the
28:01application server goes to the 5k
28:03and then what I'm doing it's a 13 what I
28:05call noisy neighbors with no control
28:07what is happening is that my database
28:10gets impacted because I have a number of
28:13iris per second like I'm sure in my
28:16let's get impacted but it's going to get
28:18yes impacted to the minimum gaius per
28:20second that's the the low band that I
28:23have on the 10k I use per second for my
28:26database for my goal so again what I can
28:29do is to have a better share of
28:31resources within my system but also what
28:34I have is the capability to at one point
28:36in time go and limit say now you are web
28:39servers we're not going to get more than
28:411,000 iOS per second at the moment I do
28:44that and I can do that online is that
28:47then my database and my application
28:49server goes back to the valance that
28:51they had something I can do also is to
28:53recover that and say yeah now I have
28:56more bandwidth so the database
28:57going to raise to 25k so I can give you
29:0025 K and get the performance so this is
29:03an interesting example because we're
29:06having 60,000 iOS per second in a
29:09compute plane and of course this is
29:11going to depend on what hardware you
29:13have this is one SSD car on for hdds
29:16that's all and this is a scaling in this
29:19testing we're running eighth notes
29:21having the same performance from each of
29:23those eight notes so that's close to 400
29:2680k iOS per second in eighth note
29:29cluster using commodity Hardware right
29:36we discussed them pick this example out
29:39of our customers it's the name of the
29:43company it's revised and it's a service
29:44provider and what we see in the area
29:48where we are working which is the
29:50Benelux Netherlands Belgium Luxembourg
29:53and the area around a bit of UK and
29:56Germany we see that those service
29:58providers they work still on a
30:01traditional way they used their
30:04traditional software they earn money
30:06with the customers the customers bring
30:07these VMs and on their platform and
30:11that's why they make money with and we
30:12believe that together with the cloud
30:16software like OpenStack you really can
30:18bring more technology and more service
30:22for your customer so revised came to us
30:25and said Fairbanks can you help us
30:27building a cloud we are really looking
30:30for reduce of costs and we don't want to
30:33have a vendor lock-in about the quality
30:36of services for a service provider could
30:39be and maybe should be very interesting
30:42for their customers so he has his SLA to
30:45his customers and with a different
30:47quality of services he can help his
30:49customer so now with this product like
30:53hyper scale well we are we can't wait
30:58object we talked several times now we
31:00can't wait to bring this to the market
31:02because we really believe it brings
31:04something for our customers like the
31:07service providers but also the large
31:11thank you root anything lastly but not
31:13less important or even more important is
31:16the data protection part okay if you
31:19have understood the explanation from
31:20grabbing it and I hope everybody
31:21understood we get quite well you don't
31:24have a Boston accent like me but that's
31:25fine you have this port in Tyne copy
31:28which is happening at the data playing
31:30every 15 minutes that's my first line of
31:32defense if you want those copies are
31:35going to be in your time line your view
31:36of horizon so in your UI you're going to
31:39see how many copies you have you can go
31:41any time click on any of them and do a
31:44restore more important is that we're
31:46using that data playing and the copies
31:49we have at the data plane to provide a
31:51zero backup window I hope you understand
31:54right now you can run a backup whatever
31:55you want that backup is going to go to
31:58the data plane and of course running
32:00were integrating with net packet and
32:02they're going to show you a very quick
32:04demo about how it works yes with one
32:06click you can define what do you want to
32:09backup when and how cool yeah so thanks
32:14Carlos so so basically it's now that we
32:16have gone through many times but will
32:17reinforce right this is the data plane
32:19you're looking at you are looking at the
32:21point in time copies that's coming from
32:23the compute node right on the on the
32:24Left arrow there so the it's a version
32:27storage right so you got the original
32:28data of the file could be a VMDK and
32:31then you got the incremental snapshots
32:3415-minute deltas that are now
32:35accumulated in the data node now it's a
32:38matter of taking that data from this
32:40data node into any backup vendor in this
32:43case our flagship product need backup so
32:45now you've got the data you just over
32:47API right you send it to a backup device
32:51and then the beauty is now you can
32:53restore it back and once you restore it
32:55back it's a one-click restore from the
32:57UI horizon once you restore it back now
33:00you can rehydrate it back we call it
33:02rehydration which is meaning taking the
33:04data from the data plane back to the
33:06compute lane now you can reinstate rerun
33:08your application based on that data that
33:11you are restoring it from that's really
33:14OpenStack need backup or OpenStack
33:16backup that's something that not many
33:18vendors have been able to do but because
33:20of this disruptive design we have now
33:22not only solved an OpenStack cloud
33:25but also more importantly we are not
33:27going to impact the application the data
33:29is already here so we don't have to go
33:31back and take an application wising all
33:33the time for crash consistency this is
33:35perfect for application consistency you
33:37still have to say backup now from the
33:39compute node the data still shows up in
33:41the data node and then it's a matter of
33:42taking the data and tear it to your
33:44backup device you could even take it to
33:46the cloud right that's kind of how the
33:48data protection portion of the solution
33:51works good thank you
33:58another example and that's what you said
34:02Abhijit and also you Carlos I really
34:06believe and I see a lot of examples in
34:09the market enterprises but also
34:12universities and other companies they
34:16really have a problem if you talk about
34:18private cloud and then what can you do
34:20with your backup and you said that maybe
34:23not many vendors I think you are the
34:26only one who really has a solution for
34:29backup for instance for this customer
34:31like University of Luxembourg what they
34:35do is they work with several
34:38universities all over the world and do a
34:41research on very difficult diseases and
34:45bring information to all the other unit
34:48to all the other universities and every
34:50time when I speak with the CIO together
34:53with my colleagues when you visit this
34:54customer they ask how can we solve this
34:57issue because it's very very high
35:00valuable you can't put a price on it
35:03because that information is so so
35:06valuable for those organizations that
35:08they really want to keep it for
35:11themselves with all the other with all
35:14the other universities but they don't
35:16want to lose the information so they're
35:18really looking for a solution like what
35:20you have right now I'm pretty sure about
35:22that yeah so let me show you a quick
35:23demo how this really works and is this
35:27is out of the box integration Wynette
35:29backup the way it works is that now this
35:32is your horizon UI you can see hyper SQL
35:34tap you are going to see backups so in
35:36the net backup site you just create a
35:39that policy is going to define when you
35:40back up with in OpenStack you're going
35:44to define for this policy when this
35:47policy is happening this is what I want
35:49to do and you want to do is I want to
35:51back up based on a flavor or a flavor
35:53and a name or a flavor on the state so
35:56you didn't really define what do you
35:58want to back up then when that back up
36:00is coming that's going to data plane the
36:03data playing moves the data to the net
36:04back up and then we think that you are
36:06you are going to have the visibility of
36:08what backups that you have you can guess
36:10go and clicking one of the images you
36:13have to a restore that restore is going
36:15to talk to net backup it's going to
36:17bring the data from whatever it is back
36:19to the data plane and then from the data
36:21plane you are going to instantiate a new
36:22instance that your way having guess with
36:25really are one click restore so again
36:30out of the box integration so I asked
36:33for my fortune engineering about I want
36:36to see a proof of how this house works
36:41you know how that's not affect
36:42performance and it's a very simple thing
36:44because I have 25 instances running
36:47having this perform and I said ok now
36:50run a backup you have the same
36:51performance so I don't know if I make a
36:53graph with just as flat line okay will
36:56be a caelius like but that is the result
36:57that is what really happens so when
37:00talking summarizing a little bit about
37:02the value probes that we have and when
37:04we think about the economics of the
37:06solution ok is that you think about that
37:09single layer that I presented in a
37:12traditional so we define a storage
37:14solution and if you are ending having
37:16three copies everywhere ok the proposal
37:19with hyper SQL architecture is that that
37:21first copy is going to be in your
37:23compute planes if you want to do an all
37:25flash all SSD and you have lot of
37:27vendors here talking about flash I can
37:30keep the first copy in flash if you want
37:32that's what I need the second the third
37:35copy whatever you need it's going to be
37:37on your data plane that's going to be
37:39higher density and cheaper storage and
37:41not only that is that hopefully you
37:44realize that what's happening is that
37:46the i/o pads are much more better
37:50utilized so now I have the easy
37:53West Network what I have Maurice Lucy my
37:55computer my computer of optimize for
37:57performance my data playing is made for
38:00data management so at the end if you
38:02have the hardware you have to serve
38:04where you have all the components we
38:06really have we can have a very good
38:08proposal in our TCO model for our
38:11customers so thanks Carlos
38:16so if you really look at it some early
38:19this is open stacks horizon console
38:21right what we have done as a storage
38:24vendor right who understands storage we
38:26have added a storage persona to open
38:28stacks horizon so when you have
38:30OpenStack installed you will see these
38:32dashboards which you don't see with
38:34OpenStack vanilla you can look at it
38:36there's really nothing here about IP so
38:38we are actually showing the primary
38:39compute plane that's the primary compute
38:42on the top the date the secondary
38:44storage data plane is there you will see
38:46all your virtual machines you can drill
38:49down to the applications to your
38:50physical nodes you can see the
38:52performance characteristics carlos
38:54talked about backup you can drill down
38:55which virtual machine to backup you can
38:57do all of that right from Horizon so we
39:00have pinned down into horizon checked in
39:02the changes of course it's needed to
39:03open source contributing into the
39:05community which we want to do and then
39:07that's what you see when you install the
39:10product on OpenStack beat canonical or
39:13red hat or my dentists or any other
39:15destroy because we are destroy agnostic
39:17right we are cloud agnostic the other
39:20last point I want to mention is that
39:21this is a storage technology we are
39:24taking this technology from OpenStack
39:27into containers like Dockers and
39:28kubernetes and we are already working on
39:30solutions because this storage
39:32technology is a cloud scalable hyper
39:34convert software-defined technology
39:36there is nothing really pinning down to
39:38OpenStack OpenStack is a launch ground
39:40to get the product going in solutions
39:42tested with you customers who are going
39:44into the OpenStack journey but we are
39:46going to take this technology moving to
39:48Dockers and other cloud friendly ACO
39:50system with that I'll move to collars so
39:53please visit us in the booth we are just
39:56here this is Venki aware director of
39:58engineering thinking about what's the
39:59next thing he is going to be building
40:00and with that we will finish with a
40:03video from our friends from noise
40:06so from the stack is it working good hi
40:11we are Fairbanks we are part of the
40:14OpenStack community a community of users
40:16and developers building the ubiquitous
40:18open source cloud computing platform the
40:21use of OpenStack cloud software gives
40:23you the ability to innovate the get up
40:26and running is very quick and efficient
40:28we help you to avoid a vendor login with
40:3024/7 production grade support save cost
40:34and you pay-as-you-go
40:35increase your strength with innovative
40:38cloud solutions we build a stable
40:40working OpenStack cloud for your unique
40:42needs and provide you with training and
40:44support the OpenStack NL sandbox is the
40:47simplest way to try out OpenStack cloud
40:49software therefore we felt a
40:52continuously growing OpenStack test
40:54environment with hardware that has been
40:56made available by several suppliers the
40:59use of the OpenStack top NL sandbox is
41:01free here you can test how your
41:03applications operate on an OpenStack
41:06environment sign up for the OpenStack
41:08dot NL sandbox good so that's great and
41:13I think it's an invitation for yes yes
41:17so we built the sandbox just to give
41:22everyone the ability to see how
41:25OpenStack works but also now with with
41:28hyper scale I think we did the sign up
41:33when you arrived here in this to listen
41:36to this presentation so we will come
41:38back to you and give you the ability to
41:42assign into for the for the sandbox and
41:44there you can see how OpenStack together
41:47with hyper scale works for you good so
41:51we really want to thank you all of you
41:53for me here I don't know we have time
41:55for questions very late with any
41:57question you may have any quick
41:59questions now ahead we can always meet
42:01at the booth - we can stay here too
42:07thank you guys I think nice solution
42:11I think the curious and tieran Keller
42:13kind of helps some of the solution we've
42:16been challenged with October as
42:17awareness was how do we provide a
42:21instance local storage with low latency
42:24and high bandwidth that could be coach
42:28base or Cassandra or some other
42:30databases that you have that requires
42:33lowest latency possible so that she can
42:35you can do a lot more throughput across
42:37the instances that you have the cluster
42:39that you build so I think you what the
42:42hyper scale kind of we're trying to
42:43figure out trying to work with it so far
42:46is to provide the ability for the
42:49instance to get the local storage and
42:52have that affinity defined so that the
42:55instance will have local storage but a
42:58high performance residents a local
43:00storage don't mistake me for just local
43:01storage we want to put SSDs or nvm es or
43:05maybe flash cards to get that high
43:06performance route but coming from the
43:08reference to the local instance so that
43:11you can build that cluster and say ok
43:13instance you have your high performance
43:15storage locally available to you and go
43:19crank it up your database as fast as you
43:21can so I think that's the kind of use
43:23because we're trying to do the end I
43:24think thank you guys
43:26Abhijeet and inhalation all these guests
43:28to kind of get some pure sea going on so
43:31just want to let you know we're almost
43:35thank you thank you script good to hear
43:38from Fairbanks also that you guys are
43:39already putting into production I think
43:41you're missing the useless is already
43:43that's good so I think we just get lot
43:45more cranking up to do but in the
43:47meantime just good good to be working
43:53thank you effective we really appreciate
43:54you coming a larger helping us ship the
43:57product and again again the open source
43:59community this is like fun for us
44:01engineer side you can see the source
44:03code of other vendors who knew and then
44:05you can contribute right that's that's
44:06really what we are doing again of course
44:08is a business aspect of it but it's glad
44:10to be here among the community and today
44:13is doing off really historic day for us
44:15to be doing something for OpenStack
44:16actually contributing into the open
44:19helping customers as they move to their
44:21cloud journey together fortunately it's
44:24a dream come true for me honestly but
44:26Carlos wait mama thank you everybody
44:28thank you for sharing the moment you're
44:30here I see how many times 10 p.m. I
44:32trying local trying your face on a WebEx
44:35every wake so you guys are there thank
44:38you for all the work and thank you for
44:40coming it's pretty late in the day we
44:41understand thank you very much you
