CloudStack allows you to use older hardware for a longer time in your cloud environment. By using older hardware for a longer time you can save money and the environment by not producing new hardware.
—————————————–
The CloudStack Collaboration Conference 2023 took place on 23-24th November. The conference, arranged by a group of volunteers from the Apache CloudStack Community, took place in the voco hotel, in Porte de Clichy, Paris. It hosted over 350 attendees, with 47 speakers holding technical talks, user stories, new features and integrations presentations and more.
—————————————–
👉 Read more about Apache CloudStack:
https://cloudstack.apache.org/
⏬ Download CloudStack: https://cloudstack.apache.org/downloads.html
👉 Get Social – Follow us Online
Apache CloudStack Blog ► https://blogs.apache.org/cloudstack/
LinkedIn ► https://www.linkedin.com/company/apachecloudstack/
Twitter ► @CloudStack
#CSCollab2023 #CloudStack #opensource #cloudmanagement #IaaS #cloud #cloudorchestration
Hello welcome all uh so I’ll introduce to you to today’s talk from Widow about how to reuse All Storage all Hardware with cloudstack okay well good morning morning and online of course as well so um some of you might have been in London where I gave the same talk I tweaked the
Talk slightly because I got more information since we did did a talk last last time so a bit about myself first you know working as at a Dutch hosting company for more than 20 years um very open source enthusiastic work a lot with SEF with cloudsac and I think we should
Implement IPv6 everywhere so that’s a task for you everybody to do do so now um the company I work for is your online we have hosting companies throughout Europe uh big presence in France in the Netherlands and in Spain and expanding through towards other countries I think
This year um and probably next year as well uh but we use cloudstack and SEF a lot for the environments we run mainly in Spain in the Netherlands we have large deployments of cloudstack and SEF so um yeah indeed our large um infrastructure is part powered by cloudstack Harlem and Amsterdam although
Refer to Harlem as Amsterdam from abroad but uh way knows the IRA Mountain data center of Le Webb that’s uh always referred to as an Amsterdam Data Center and in Valencia in in Spain so we have hundreds of hypervisors running thousands of virtual machines but we don’t like to throw away properly
Functioning Hardware it’s a waste of money so the financial people like it but it’s also pretty bad for the environment and we care about both that’s the honest answer we care about the money of course and we care about the environment so let’s take a look at
Servers 5 years old are they old servers do you think they need to be thrown away need to be replaced who replaces his servers every five years because it’s company policy okay nobody’s doing so but if you look closely at a fiveyear cycle as written here is that vendors
Are telling you that you need to replace everything every five years but actually they will operate for about four years you have six months from delivery to production and then before they hit the 5-year Mark you want to to start replacing them so you have a 4year window where the machines are operating
And then are not even operating um at their full capacity for the full four years so yeah you as I said replacement before they go end of life in a financial perspective after 60 months but is it really wise to do so well I think it’s not because that’s the whole
Point about this talk because we can look at a few things there’s a global warming potential for greenhouse gases um there’s a carbon dioxide equivalent the the equivalent means that we have other greenhouse gases like methane which have a higher warming potential than CO2 but they calculate it back
Towards a CO2 equivalent and a lot of products have a product carbon footprint um you look at Dell Dell does it very very well they also have a life cycle analysis of the servers you buy from them you know how much CO2 equivalent is emitted when producing those machines so
Let’s take a look at a Dell server running time four years operational 247 the global warming potential is 8 and a half tons of CO2 which this machine emits over its four-year life cycle and this is information which Dell provided you can look at other Machines of Dell the
R6515 um uh running the AMD epic um and you see that actually the manufacturing process is quite a big impact of the server itself there’s of course course the use and there’s a bit of transport and and all the and the EOL getting rid of the machine but the production of the
Machine has a very high impact uh on on on global warming why we’re producing um chips um the the metals which are being used to produce your machine um and you can find this in the life cycle assessment of Dell server so just Google for life cycle assessment and the model
Of machines you’re using and again Dell was really good at this and Publishing these numbers all the manufacturers will be actually mandated to uh produce these numbers and publish them but it is an impact we need to take into account but what if you would run your service for a bit
Longer so there’s a part where you have no influence that’s the production part of the machine if you run for four years it’s about 50/50 if you start running the machines for a longer periods there’s a part we have an influence because that’s the energy they are consuming while they’re
Running what we’ve seen over the last few years is that CPU are becoming faster and faster and actually more energy efficient but it also slowed down a bit in the fact that a CPU from six years seven years ago can still work just fine for many applications Sven we
Were talking about this yesterday that an E5 CPU an Intel E5 CPU from six seven years ago is still fine to use for a lot of applications and when I talk with Engineers they say yeah but you need to buy the new service because they’re more energy efficient and so that’s better
For the planet well actually if you look at at this part it’s it doesn’t really matter that much you can run servers older servers and um be much better for the environment of course it depends for your energy mix if you um where it depends on where you are on the planet
However if we look at the part where there’s no influence and then we can see that if you’re using a machine with ssds what we’re doing 80% of the CO2 emissions come from the ssds those are the most horrible to produce if you look at uh CO2
Emissions and I’ll get back to that in a moment so 80% is just the ssds because there’s a paper from uh October this year uh you can Google for this and I have the quote here is that for a single gigabyte of flash memory 0.16 Gil kilogram of CO2 is emitted so a
1 tbte SSD has 160 kg of CO2 emitted as a footprint one terabyte so go towards two four terabytes whatever you might be using you’ll see they have a major impact on on the environment ssds and they got the same numbers um from all kinds of different uh research
It’s it’s a good paper I think it’s like 10 pages um but you can see here they come up with the same number in a server about 80% of the CO2 emissions come from the ssds not from the rest and if you would look at a equivalent with hard drives
The number would go down to about 38% compared to 80% hard drives are much better for the environment than ssds are so we were looking at the part where we had no influence there’s part where we have influence of course which is energy but then you can also look at the CO2
Emissions which countries have the European Union they publish information about uh the CO2 emission per for a kilowatt hour well the country where I’m from the Netherlands we’re not doing that well with 418 G per Co for every kilowatt hour which is produced countries like France for example do
Much better there a lot of nuclear power in this this country here so those kinds of things need to be taken into account as well but it’s a no-o right using refurbished Engineers will always ask I need the newest the most shiniest machines because that’s the best we need
To spend a lot of money because if we spend more money we get better machines and spending more money is just better because then you have the best stuff well if you look at it from a different perspective you say 50% in the CO2 because you’re simply not manufacturing new
Machines it’s also significantly lower on kexs can be 50% usually 30% depending uh on where you buy them if somebody has tested them for you and did some proper refurbishing on them but if you buy them secondhand it could even be going Beyond 50% and my opinion is is that you should
Have your customers are asking for a service and you need to deliver the service if that’s of Hardware 5 years old 6 years old 7 years old well I think it doesn’t matter as long as you’re meeting the as laa which you have agreed on with your
Customer so take a look at refurbished machines and there’s a big Market out there where you can go and just buy refurbished servers and I would also say they’re ble tested in the last two years we we’ve been buying a lot of AMD Epic Systems as we were throwing out a lot of
Really old Hardware um but I found that the stability of these new machines for the last years was pretty bad we’ve been buying super micro andell all kinds of glitches and crashes and we need to do bios updates micro code updates for the CPUs to get them stable but the
Refurbished machines we’ve been buying they’ve been much more stable than actually the new machines we bought so I would say they’re battle tested okay but we’re on a cloudstack conference here so why we’re talking about CO2 emissions right so tell us something about cloudstack right that’s
The the whole point well Cloud stack has a few interesting things you can use um for um determining which Hardware which service needs to run on which hardware and it’s the tagging I think swen you mentioned I think the from Wii systems they said we’re using different types of
Hardware in the same Cloud stack environment well we do so as well so let’s take an example of our environment in Amsterdam we have many generations of Hardware the Intel xon so the E5 the Broadwell the scalable so the Skylake CPUs and the AMD epic generation one and
Two so we came up with an idea of having an offering tier one it’s always the latest generation of Hardware in our case AMD epic but there’s also the Intel zon scalable because we acquire companies which might have gone for the Intel zon scalable so they’re called
Tier 1 a tier 1 B and the tagging we use it’s still tier one tag and then the tier one a and a tier 1 B tag you can read the the table here so we go down towards older Generations even have e5s V2 and V3 CPUs still running in the
Environment so the service offering with the T1 tag um so you can go very specific saying I want this to be running on an AMD epic t1a or you said I just want a a service offering with proper CPUs which is the latest generation I’ll go for T1 so tier
One high performance websites like e-commerce which really need a f low first time to first time to bite um tier one but many back office applications tier two and spam filters they can run on tier three it’s not needed to run them on the latest generation of
Hardware so we try to use the hardware as long as possible that is our goal again as La up time that’s something which we Benchmark we check if the systems are still functional but it’s not that we say that Hardware needs to be replaced every 5
Years and how do we do so with live migrations well the same types of Hardware are in the same cluster so in the Amsterdam data center One locations there’s a rec room 138 A5 these are super micro one used with the AMD epic uh generation one um and then we have
All kinds of different RS so per rack we have a generation of Hardware in one rack and that’s one cluster Inside cloudstack So within the cluster then you need to make sure that in the KVM agent you set the CPU model to be identical between all the nodes in the
Cluster so live migrations work as as as expected indeed so same CPUs in the same cluster which is one Rack in our case and if we start a new cluster we we use a new rack so you’ll find some of our RS are really full with a lot of machines
But some can only house 10 machines because this rack is limited to 32 amps and those AMD epics the machines consume about one amp no no more than two amps per machine so we can fit 10 machines in one rack and then it’s full but some other Generations we can put more
Machines in a single rack and uh yeah we’ve been using Dell most of the time but we have a lot of Super Micro as well so this is the re we have deployed in Amsterdam Valencia and I really recommend using colored power cables that’s that’s something I I find it cool
But it’s also very beneficial to uh see that uh all the cables are connected properly just as a side note because why I’ve seen that if these were black cables I’ve seen that some people made mistakes when pulling in the machine so one machine will be connected to feed a
One machine will be connected to feedb it was all beautifully cabled but with black cabling and then one feed went down and then a Servo went down because there was a mistake with cabling and that’s something in oversight you see very quickly when you’re using colored power
Cables it’s a lesson we learned the hard way a couple of times so uh colored power cables much better anyway we also use SEF and SEF is designed for failure so why should we buy the newest Hardware all the time for SEF again I even think that older Hardware is more stable the
Newest one why not use refurbished um hardware for a SE environment does everything require the highest performance I don’t think so so we run a few SEF clusters we’re still with SATA based ssds and they were just fine for back office applications no need for high performance just run and I think we
Have the oldest cluster now hitting eight years and still running just fine it’s 10 gig based environment updated to the latest software and just running and running so occasionally replace an SSD but um only when needed so my summary is and my main point here is is that Hardware production has a
Significant impact on the environment it’s not often talked about but it’s something which you should be aware of so you can use host tagging to do cool stuff but older Hardware can still be used and think twice before replacing equipment and also think twice ah let’s just replace all the ssds because I’ll
Go back to this one 80% of your CO2 equivalent of the production of a machine comes from ssds they are really cool with performance but really bad for the environment and then there’s an open question for the community do we actually know how much kilowatt hour
Cloud stack consumes do we know how much a single trillion run uses when we do something on on on GitHub or a blue or utang run if you ask can you please test my pool request do we know how much CPU power is being used and how much energy is consumed I
Don’t know maybe there something we should try to find out as a community as a project and publish about that okay you just are you sure you want to run this test because it’s going to emit so much CO2 and consume so much energy do we ever think about it as developers
Because cicd is really cool right we just build all day let it building and testing and we do one commit and the whole uh shebang goes again and we start testing everything but is that really required so maybe it’s time to start thinking about this as well that doing a commit and having
It tested is fine but maybe you should think twice before asking a testing environment to run it because yes it runs on a server Far Far Away doing the testing for you but it is consuming electricity which in then in turn is actually producing CO2 so that is
Happening it is real um so kilowatt hours you probably know what what they are so I found an image about this but maybe we can find out or make an educated guess and inform users and tell people saying hey this is the impact of a single test you’re running on
Cloudstack so my main point is just awareness be aware that Hardware has a major impact on the environment think twice save money in your wallet save the environment by using Hardware a bit longer so thank [Applause] you thank you weo there are no questions from the online um any question from the room
Anyone uh so my question is uh uh when you have to replace some hardware for example after 10 years uh will it be a nightmare to replace the hard drive of sefe cluster well no I if you look those SEF clusters which are running ssds they used to be running on the Samsung
PM 988 with the eight something starting with an 8 pm8 Samsung is still producing SATA ssds uh for servers and they’re now the nine Series so whenever an old hard drive an old SSD fails we replaced with a latest generation of Samsung ssds and they might have a slight performance
Difference but you don’t notice on that back office application thanks and the the thing I think what we did is with our Rex is that we have so one generation of Hardware of hypervisors but the same goes with SEF there’s only one SEF cluster in one rack
If we build a new SEF cluster it goes into a new rack so the cabling is also still very nice per rack thanks we have a new question from the online uh from Yosef do you buy single CPU servers for SEF because of Numa or two CPUs are fine as well okay
There a sear related question we do use one CPU the AMD epic so indeed Numa makes life much easier just using one CPU so we use a single CPU yeah great thank you we okay thank you thanks