00:00
So welcome. Uh, everyone, um, good to be here. And my name is, uh, Roy Mikes. I work for pure now, roughly, um, three years. Um, MS. Thank you. It's good having you, Um, I work as a partner, technical manager in the Netherlands.
00:20
Um, mainly in the past, I have had many roles from storage technology to disaster recovery, business continuity and so on and so on. And in the last decade, I'm working together with partners on partner Enablement to make sure that the partner is sufficient, trained and certified and well equipped to sell to sell pure.
00:50
In that case, So, um, my introduction. OK, up to you. Good afternoon, everybody. My name is I'm from the Netherlands. So my English is English. Not really English. So apologise for that.
01:09
I'm working 25 plus years in the industry. At this moment. I'm a lead architect with open line with a focus on storage and backup. And this is a story about storage systems in our big, uh, manage service company with an amazing outcome.
01:29
Um, first of all, who's open line? Open line is founded in 2002. I was one of the first I'm executive defence employee with Open Line and we start growing. Now we are over here and there are a lot of of course, but the most important thing is the 14% year over year annual growth we made in the last couple of five years.
01:53
And our ambition from our sea level is to do next year. 17% and you can make some more revenue by selling more to people or changing innovating your systems. And this is about innovating your systems. What we are going to talk about a lot of more numbers at this moment. We are with 300 plus employees.
02:16
That's a colleague of mine who he's the victim today. Um, when you look at net promotor score of 8 78 and a revenue of 75 million most important things, we have specialisations in technical specialisations, healthcare, local government, housing and logistics. And this is the most important one at this moment, our E Health services for hospitals.
02:40
We have generated a virtual platform to discuss patient cases, and in the Netherlands we call that MBO multiple Discipline Organisation, where several doctors from several different hospitals are working together looking into a patient case without copying data from the from the original or the organisation, and that's important. Otherwise you can multiply copies and with our
03:10
current laws, it's not allowed. You have to be. So we present them a virtual environment. They can sit on the workspace and work with those documents and talk about the patient. The treatment of the patient is there explained, covered the document and centre around how we got to treat it. It's a very new one and another one is very
03:28
important at this moment where it's not security is one of our biggest things at this moment. That's open line and we are located in 10 locations. This is the most orange picture I could find for the Netherlands. We are not there. This is Amsterdam.
03:47
Yeah, Normally every service provider in the Netherlands is in Amsterdam or Surroundings of Amsterdam. We are not. We are in a little booth from the over there and there is one unique selling point. Everyone knows that Holland is below sea level. Those two data centres are not. They are both 155 and 114 metres.
04:11
There are some other things. We have a secure A security operation centre in Valencia, Valencia. There's a university with a scholarship on security, so that's why we're there. We have some in England and in Slovenia. Those are the guys who are making that virtual infrastructure on E health.
04:29
So a little bit of our company, that's me. In better years. I had some here, Um, my team has seven architects and we are struggling with our environment and now talking about the situation of three years ago, we're struggling with our environment, So we had a lot of visionary things. These are 33 visionary things about what to do
04:56
with the environment. First of all, most important things. Better service provider revenue. If you want to make some money, that's why we're there. So we want to stretch the systems to the limit. When the system, when the seller says you can do 80 we want to do 85 because bigger heads bigger money. That's what we want.
05:18
And this is not about money. This is about innovation is not only money is by selling new getting new customers, but optimisation of your systems. The second thing three years ago, we noticed that our energy cost was one third. One third of our total operation costs a lot of service, a lot of storage. And we had to do something about it for the future.
05:46
And now we are there. We had to do something about it for the future. And we don't sell storage. We don't sell computers. We don't sell networking. We sell services on top of it. So our environment has to be stable for boot up time. 100%.
06:03
That's what we sell because of E health system. When your patient in the hospital and you want to discuss your case, God prays, it must work. That's the most important thing. So it has to be scalable and robust with that question. Roy, that was my question to you. What can we do with it? Pure. Can you help us?
06:26
Here you go. All right. Thank you. Question. Sure. So a sales promise to go to be true. That was, in fact, um, our starting point. So, um, the account team took open line to an
06:47
ABC in Mountain view, and they talked with cars, and the result of that was CP waiting time. So implementing faster storage led in fact, to reducing servicers and CPU costs. And that's of course, good for your money, but not only good for your money.
07:14
It in the end when you have less, uh, servers, then you, of course, have also less Microsoft licences. So there were three pillars, of course. The no brainer, the financially one, the technical one and the environmental one. And if we look to the to the middle one, with the reduction of complexity,
07:38
I mean back then had 36 different storage systems and that was by organic growth, so perfectly normal, but how to manage and very complex. So to reduce that complexity, we reduced the CPU cost. We reduced the latency and, um and, of course, no migration anymore. The only time they had to migrate was in fact migrating the data from the old systems
08:13
to the new systems. And I mean, with 36 different systems, you are migrating on the clock, right? It's unbearable. He did it so and and and and the environmental of course, by the reduction of arrays, reduction of san supports, you get,
08:33
of course a reduction of power and reduction of um, of space. So to summarise this, the assumption was was made that, um you need when you when you have a certain amount of capacity, you have to manage that with a certain amount of people. So the assumption was done with seven people, which was brought back to two,
08:59
but must come back to that, uh, later licencing. Of course, we went back 10 20% because the CPU cost and the latency, in fact, no more migrations anymore. You back from nine per year to zero. OK, just one. And the wreck space went back, the power went back, service went back.
09:26
And, um, yeah, and especially if you look at the power 766%. I know back then that was a big deal. But go figure now, you know, with those energy prices insane. So, um and yeah, of course, all numbers were based on the assumption back then. So we went back from 770 U to 30 U.
09:59
I mean impressive, right? Um but that doesn't say much, you know, if you if you have to imagine how that they look in a typically a household in the Netherlands with an average of 3500 kilowatt hours per household. You get something like this and bear with me because it's building up.
10:19
But there you go. So the energy they you made that you made that animation it was crazy. But I know, I know, I know anyway. But the energy energy that they that they used was was comparable by 273 households, just the data centres of open line and that the assumption, of course and
10:55
migrating to you get something like this. So we went back to 36 households. It is insane, of course. And we told Marshall that And you, uh, still flabbergasted. This was a proposal three years ago and this morning in a keynote,
11:22
there was a guy as bald as me blue shirt sitting in a chair, telling the same story as we experience. So now you can go because it's the same story on but still amazing still entity, by the way. Oh, yeah, that one. But I'm I'm going to explain a little bit more. This was my first reaction.
11:45
When the technical be awesome, I want to judge it myself. I have to convince rob some colleagues and I have to convince our CEO because he's the biggest tech we have. So that's second one. My dad told me the sun is for free. And I doubt if Vegas the so is for free this? I think so. I think not.
12:07
And everyone was changing I. I love the animation they made. So we all were flabbergast and look at the animation. But could it be true? 12.2 kilowatts an hour. And it was something like 95. Very close.
12:23
Yeah, OK, first of all, I have to explain our problem. And that was the biggest problem. We had our problem with latency. May you look at a normal latency of our old system and ladies is in a mirror. Situation is doing about right. Uh uh, acknowledgements to the host.
12:45
And we have a round trip of 11.4 milliseconds. This was the lady we could realise with our old systems. Was it good? Was it bad? But then we stretch the systems. There we go. The latency went up to amazing 12.6
13:06
milliseconds. And then it's bad. Then it doesn't work, so you have to lower your utilisation of storage system to maintain a good latency, and that's not what we want. We want a full loaded system with a low agency. That's what we want. So there is your problem, and the guys told us the problem is solved
13:30
with metal cloth or active Closser. Sorry, active class. So then it's sold. Now you have to know how active clusters working and there is. A big difference between active cluster and normal, uh, auto storage systems is that there is no double round trip to get acknowledgement back.
13:52
There's only one little bit bit additional 0.1 on the front and the back a little bit different additional latency. And they presented those of 1.4 milliseconds latency in total on the load. So the first thing we did was approve a concept two of those systems give up the the assignment, stretch it to the limit and see what it does. And, yes, this is not the production data. It's simulated something like that,
14:23
but it stays on 1.4 no, three years later, 12 systems richer. This is one of our systems, and this is an average one. Is doing 1.4 mills. Still an amazing number. And yes, we discussed by lunch We stretch one
14:45
system to 106% and it still was. How do we do? 106%. I don't know how we did it, but we did it and it still was 1.4 milliseconds. And that's what we want as a service provider. Stable, robust and scalable. 12 systems. 36 systems we had.
15:05
It's the difference. Second thing, storage efficiency. First, some explanation what we do with storage. In the past days, we bought storage and you buy physical terabytes. This and then you do something called raid groups Hot space. We have to calculate a little bit between terabytes and tibi bytes,
15:33
but you lose some capacity. This is this is what you're selling due to the fact that we had sure not that good storage systems that we bought and the lady she went up and we raised utilisation. We couldn't use all the capacity, but still buying this.
15:54
Then the it data efficiently applied at where this was SSD systems. And, uh, tomorrow is SD is not the same as an NVME. Something like that. It simulated diss. And this is what we're selling. We're selling virtual machines and data on top of it. When you look at numbers, this is one side.
16:17
We bought 2.7 raw petabytes in storage. And at the end, the servers are storing 0.9 petabytes in storage. So you're losing something like 64 percent in capacity. This is and he told us, Don't do this with you are a service provider. Do it with pure one pure one.
16:48
I think everyone knows pure one is someone who doesn't know pure one. OK, all good commit. Do you know this is what we are paying? This is buffer capacity for pus, configuring for us. So when you look at the previous sheet left side, it's not my problem.
17:15
It's your problem Exactly. This is what we provisioned. This is data. This is where we now and be aware. Buying war capacity and K versus capacity written data.
17:33
There's a price difference. This is more expensive, but for a managed service provider, pay as you use. That's the thing. I'm only getting money from customers who are using data so don't have to invest any money upfront. This is what we do.
17:50
But you have to do something else. You have to stay, stay close to the wind. You saw the on commit and on the there is a 12 terabyte margin between on the on the amount and commit when I go 38 terabytes per month and on the month I have to raise my commit because that's cheaper and it's
18:16
everything about margin. So this is the most thing I do all day. I'm not a technical architect. I'm more financial architect because our sea level wants to see numbers. They know for sure that the system is working, but we have to calculate every month what's new. What new customers, what to migrate. What's the growth?
18:37
What the is. So that's the most important thing this morning and me sail close to wind Play with it. You can change your commit every month. There's a minimum amount 50 terabytes on a pure X, but you can do it. The only difference is the amount you will over commit is to the contract.
19:02
And another thing. What does your think? We had three storage consultants. Senior guys know everything about our previous brand all the kind of systems, all the 67 different flavours of systems they can manage and then came in. And my first question is, I have to train those guys.
19:29
Yeah, so maybe to add something, um, So what is a typically thing with a with a legacy Storch vendor every three years, a new system? And of course, that's a better system than the previous one. But still, it's a 1.0 version of the new one and probably new interface new system where you have to train people.
19:57
And that wasn't necessary because of the the ease of our interface, our systems. And of course, those guys get trained because they have to install it. They have to upgrade it, et cetera, et cetera. But the easy of management and installation was done by the right, of course.
20:19
So, yes, we are training those guys Now when we do a little bit more by ourselves, it's better for timing. It's better for all things, but we do it ourselves, but it's not necessary. In Africa, one pure does everything for you. The last storage migration ever.
20:36
We aren't there yet. We are now the third year, four year contract. So next year I'm gonna experience if this is true. But tomorrow, uh, this morning, I, I heard from a guy who had 50 85 migrations upgrades done in the last couple of years without any outage.
20:55
And what we did with 36 systems maintaining every year we have to migrate forklift, nine systems with all the cost of things and outage and whatsoever. 94 cliffs a year. We did it with three. And now we did three years of migration with some forklifts to and
21:20
hopefully this is the last one. We did some forklifts because they promised us that this was the last one. And we're going to see this next year. Then we have to change to the generation four. Thank you. Exactly. I mentioned to the generation four new system new, faster systems of pure because the end of the contract Another thing
21:44
doing a lot more with the same when you are growing three years revenue growth. I know revenue growth is not equal to storage growth, but the sea level doesn't know now and now, you know, because this is recorded. Um, I always had storage systems have grown capacity at 55 0%. 3 guys doing a lot more that three years ago.
22:12
Data migration. Um and then you ask me to do this a couple of years with three guys. I said Yes, no problem. Because we don't do any immigrations anymore. That's the difference. And and this is an important one. Unexpected help.
22:31
My fourth member, even my 4567 member in my team, the business critical service team of Pure. Be aware. This is not an extension of your evergreen one. This has to do with a number of storage systems. You have to be a big one, and I think we are a big one in the Netherlands. We are a big one.
22:55
Then there is a help. There's a business critical services team, four guys an account manager and CE and some real technical guys who are helping us from day to day basis with our technical infrastructure guarding performance, looking at trends advising if hosts are not double connected, single connected host,
23:22
looking at the systems on technical details, something we did before ourselves. But they are helping us, and every month there is a meeting where these are the main topics, discussing what's with our environment telling us They asking us if we had some problems with support? Of course, normal ticketing system is working, no problem at all.
23:51
But when there is a ticket open for a long time, we discuss it over there. It's a very nice help. He checks our users OK? They are telling me so real. They are telling me you get a new system. Why? Because those four systems or five systems are
24:14
at the max. We put in a new system and I don't pay for the new system. It's not a it's a I pay only for the data that's on the system. Host written data you will see and recommendations next steps. We are discussing which version of pity to go to. Now we are 63 11 at this moment.
24:36
Yeah, so 64 is our. But it's not enterprise ready. And I think we are an enterprise. Uh, but that's the discussion. What are you doing with your system? How do you upgrade it? When do you upgrade it? And with the self service portal and self serve, it's working.
24:52
If you ask your own data put in and pure one. That system. Yeah, that's working. Yeah, So it's a very good help with those from those guys. Then I think this is the most important sheet of the whole presentation. I told you he showed us in an animation. 36 homes in the Netherlands equals
25:16
12.2 kilowatts an hour and storage. This is the result. This is from last month. I look at the numbers and the circles. 19 185859 equals is almost 12 kilowatts. And I still think it's amazing because you have
25:38
to do how many power system is using. That's not the problem because they know they invested our environment, looked at our data and said, You need this amount of storage systems, these types three years ago and the only thing they had, I mentioned them. Name is live optics, and I think someone knows what live optics is and don't explain a bit more.
26:09
That's the information we gave them, and they came out with this number. Another important thing. Orange is for pure the best scholar. I don't think so, because this is what I mean with sailing close to the end using your commit for 94% and when I'm going above about 1%
26:31
I'm buying more. Commit the top one where it is 84. We increase that one because we are migrating from a new customs into there. So that's why we increased. And you see, we have two licence agreements separate from each other. This side one and side two must break on the order this long half.
26:53
So this is a very nice one. These no efficiency are nice to know and all Law and the Netherlands, we have to convince what we do with C. Two kind of levels, kind of things. These numbers we need for that this one fibre channel ports about 50% less fibre piano points.
27:18
We didn't migrate to a higher speed. We are still working on eight gig or five, but it's working because you spread it out over a lot of systems. We gained the numbers by eliminating those systems and losing some hosts.
27:39
We didn't need to connect more hosts, even put on less host with more memory and one of the most convincing things did. Our CEO went to the States put him in a meeting room with and he talked for. I thought 1.5 hour something like that, and he was convinced about the thing. Cost all them about the weight state from a processor. A processor is doing nothing.
28:08
You can't fill them up first, is put in additional memory and put in faster storage. And those things resulted. In the last two years, we didn't had to add some additional host, but we grow 14% with no additional additional memory.
28:30
And when you buy memory at a good place, it's not that expensive. Um, but only additional memory, no additional host. And then when you look at cell funding, no training, no this no desk, no additional servers less Now we are doing a project within the end of this year next year to migrate to 16 gig New fibre channel environment by Fibre Channel.
28:55
It's robust. It's made service quality. It's predictable, but there's also a problem with it, but I'm gonna get it back there later. But this is what we did. And yes, 90% of the racks empty.
29:13
We were discussing a third data centre. Here are your racks for the next coming year, so that's also different. It's all about money. It's all about better systems, better usage, less power usage. And this is real. This is a road we did a trip we did over three
29:36
years. This is our result. Have do I have some advice for you? I think the first one is the most important. We are a ego tripping service provider in the Netherlands. We are the best. There's nothing better. We are the best.
29:52
We also are Internet service provider, so we have the best Internet connection there is in the Netherlands. When you make a active cluster, there are some four domains. The system itself. The application between those systems has to be separated. Connection to the out to the immediate to pure
30:13
one has to be separate. But when you Internet service provider, you think, Ah, the best Internet connection of the Netherlands. Those mediator talk thing can be done by those Internet feats. Don't do it. Don't learn it the hard way.
30:34
Take my advice, order some, keep as Internet connection and spin it off in your main infrastructure and when there's something hitting your main infrastructure whatsoever, and both of those things are gone. Then a middle class that the perfect side will survive, of course, but you don't don't make any management mistakes.
31:01
So when you have the perfect side on 11, there are no horse on it because we have no uniformed access. We have a non uniform access, so horse is only talking to one side. Then you have an outage. I see the audio is gone.
31:25
Is gone. I was to good. Can you bring me Mike Ho? Can you bring me a ho mic? My, That's 12. What?
31:58
It's a hip hop mike. So? So there is some advice to the audience. Make sure you fold them in. That's what I told. Did you hear everything about or what I going on at that moment? Are you here?
32:15
The second thing. We look at the system on a financial basis, not on a technical basis anymore, because what I said it was working. It works now. Nowadays, the network is the problem, not the storage, but you have to stay close to the wind, and then it's interesting.
32:36
The other thing I ask, I now look at Roy for more than a year to make the active closer and an active passive mode. And everyone will say muscle. You have to use active DR that's active passive, but active DR has an RPO recover, uh, point of the form somewhere about zero, and we have 12 systems.
33:05
Then a customer is divided over more than one system. So one system is doing an RP over zero the other system of a minute, and then I have to do a fail over. So I want to have a RPO zero system with not the mediator kind of thing with some policy. This ignore mediator activity because
33:28
mediator can manage the whole cluster. When I say you, your preferred size has to live and the pre facci is a little bit slow reacting to the mediator. It will die. And on the other side, the secondary side. Oh, no. So in an active passes way, I want to have this know that I'm asking this for a
33:52
year. Another thing Balancing 12 of storage system is a pain in the ass. Sorry for the word, but it's difficult. We have to migrate Lu between flash array Couple 12 flash a ray couple two because one is full and two is empty or one has and I want something called active work.
34:16
Yes, it is 64, but not enterprise. Ready. The other thing. Most of you can use fusion as well, but they are. I'm using Fibre Channel in a brown field. So no, I que in a green field. So that's always as well. I want to have fusion in a brown field with
34:40
fibre channel. Do we have another one? Yeah, you could expect this one in pure one. There is a for every one at this moment, a confirmation kind of tool where you can put in your workload. And he says, You have to expand your system. That's technical. That's his problem.
35:03
I want to know the financial impact, and that's not in pure one. And it would be nice that the cheese I made for myself to guard and sail in close to the wind should be in one. And I know it's difficult because there is a partner between it, and there is a reseller between it, but figure it out to make it work in a financial kind of
35:26
way. But after three years working with doing a lot of migration. I only can say one thing. Wow, what they promised us three years ago, it all went come through. And hopefully, uh, the the the the the migrations will go come through as well. And thank you very much. Is has anyone some questions give you the mic
35:58
for the active cluster? You said active passive. Why don't you guys don't use preferred path? That's we have a problem with that. So speak up. So the problem that we had was that both arrays had their own No.
36:21
Had shared an Internet connection. Um, and that's the network problem that we had. Uh, so the internet connection went down and, um, the replication link, which is also running over Ethernet when it was also down. And, um, I think because both the mediator and the replication link were down,
36:46
both, uh, PR couldn't talk to each other. Um, the last known pure that was alive happened to be the wrong pure. There is a service provider. You have one Internet feed. Otherwise you have to use lips or other kind of things. So the one Internet feed is living in a virtual space data centre.
37:17
one or data centre? Two. Yes, and we replicate data for the Internet to. It's a redundant pair, but one is active. The active site is faster, as less latency as a non active side. When you prefer, uh, pot is over there, he has to travel this way.
37:40
He's always slower as a non preferred, and that's the problem. There's only eight seconds, eight milliseconds difference. And when your round trip is 1.4 then it was It's a bit tricky. That's roughly our problem. What our problem Do you know what I mean? How about like, having like a mediator on speak up so that
38:10
then you need a third data centre. When it's in the data centre, you have two. You have to need a third. You had to have it on a location. That's another problem. Yeah, so the best solution is give me a policy, which is overriding the A cluster connection to the mediator
38:32
because the system, who is not preferred says sorry, I'm gonna die. That's no problem, because there is no compute on that system. The system who is staying alive with a has also thing. So in a Brit split brain situation for me, Always a good one is living. I don't need those discussions.
38:56
I want to live that one. That's what I configure. Yeah, yeah, yeah. We had some problems in the past that the wrong the wrong part stays alive to make an override, but so that before sign or refer to the array would always be This is the one that stayed alive after we had That's that's what we want And our
39:28
even our worst old system had that.
