www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Slides share: DMesos - Not only a re-implementation of Mesos

reply =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
Dear all,
    I am a D-language amateur from China, and just want to share 
you with a slides from me that post on MesosCon Asia 
2017(Beijing):
    
https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
described in this slides,
your help and suggestion are very welcome!
    Thanks!

BRs,
Koo
Jul 10
next sibling parent reply Joakim <dlang joakim.fea.st> writes:
On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to share 
 you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!
Wow, you've done your research! Lot of good D stuff in those slides, nice to see the ldc team's work on ARM and GPUs has been noticed (glad to see a mention of my Android builds too). I had never heard of Mesos until now, shows how I don't track cloud at all. I'm actually skeptical of cloud- I think mobile p2p will eat most of the cloud- but a lot of these distributed D cloud libraries you want can be repurposed for mobile too.
Jul 10
parent reply "Nick Sabalausky (Abscissa)" <SeeWebsiteToContactMe semitwist.com> writes:
On 07/10/2017 02:16 PM, Joakim wrote:
 I'm actually skeptical of cloud- I think mobile p2p will eat most 
 of the cloud-
I've been REALLY hoping p2p will eat...cloud^H^H^H^H^Hcentralized internet services[1], but if I were a betting man I'd bet heavily against it. For one thing, for p2p to kill "cloud" we'd realistically need IPv6 to become much more ubiquitous, and that just isn't happening. And I think the #2 reason IPv6 isn't happening (behind plain old inertia) is that it would *allow* p2p to overtake cloud. Which brings me to the next reason I don't think p2p will kill "cloud": All the big players with all the money and the power all LOVE "cloud" because it allows them to hoard more power, control and money, whereas p2p would completely destroy that frontier for them. Also, replacing "cloud" with p2p would mean more reliance on user's devices actually having decent storage and upload bandwidth, but non-power-users (ie the vast majority of people, if you don't live in hipster valley) are ambivalent towards that, and it would raise the price of their devices, AND they don't want to deal with running low on storage, or backing things up, so they love "cloud" too. Both the engineer and the humanitarian in me both REALLY want to see p2p eat "cloud", but I just don't see it realistically happening. [1] The word "clould" bugs me to no end. It's the tech sector's equivalent to "smurf" - stupid word is used to mean *anything* internet-releated, even "internet" itself.
Jul 10
next sibling parent Joakim <dlang joakim.fea.st> writes:
On Monday, 10 July 2017 at 18:45:34 UTC, Nick Sabalausky 
(Abscissa) wrote:
 On 07/10/2017 02:16 PM, Joakim wrote:
 I'm actually skeptical of cloud- I think mobile p2p will eat 
 most of the cloud-
I've been REALLY hoping p2p will eat...cloud^H^H^H^H^Hcentralized internet services[1], but if I were a betting man I'd bet heavily against it. For one thing, for p2p to kill "cloud" we'd realistically need IPv6 to become much more ubiquitous, and that just isn't happening.
I don't think it's necessary, but it would be nice to have.
 And I think the #2 reason IPv6 isn't happening (behind plain 
 old inertia) is that it would *allow* p2p to overtake cloud.
I think you underestimate how well ipv6 has already done on mobile: https://blogs.akamai.com/2016/06/preparing-for-ipv6-only-mobile-networks-why-and-how.html That post is now more than a year old, extrapolate accordingly.
 Which brings me to the next reason I don't think p2p will kill 
 "cloud": All the big players with all the money and the power 
 all LOVE "cloud" because it allows them to hoard more power, 
 control and money, whereas p2p would completely destroy that 
 frontier for them.
Sure, many of the big tech players have built or are building giant cloud businesses, so they're not going to be interested in p2p, especially since there's no proven p2p business model. Look at how BitTorrent Inc. floundered for more than a decade, before finally sinking recently. This just means someone new will have to come in and knock the big boys off, maybe you. ;)
 Also, replacing "cloud" with p2p would mean more reliance on 
 user's devices actually having decent storage and upload 
 bandwidth, but non-power-users (ie the vast majority of people, 
 if you don't live in hipster valley) are ambivalent towards 
 that, and it would raise the price of their devices, AND they 
 don't want to deal with running low on storage, or backing 
 things up, so they love "cloud" too.
A midrange smartphone or tablet these days comes with 2-3 GBs of RAM, 32-64 GBs of flash storage, an ARM chip about as powerful as a desktop chip from 3-5 years ago, and a 4G LTE chip that provides dozens of Mbits/s both ways. Please tell me what they're missing, you can buy all that for $200-300.
 Both the engineer and the humanitarian in me both REALLY want 
 to see p2p eat "cloud", but I just don't see it realistically 
 happening.
All you have to think about is whether the hardware is capable enough: we both know that it is. Then, the software will flow downhill to where it can be executed most cheaply and reliably, _once_ there's a p2p business model. It may take time for the dumb engineers to figure this out, but someone finally will, and when he strikes gold, the p2p gold rush will follow.
 [1] The word "clould" bugs me to no end. It's the tech sector's 
 equivalent to "smurf" - stupid word is used to mean *anything* 
 internet-releated, even "internet" itself.
Heh, I was surprised recently to hear someone actually use the term as a real "cloud," ie they were referring to both mobile devices and servers in the data server, a true distributed cloud. But yeah, most of the time it's just a new buzzword for ye olde client-server computing.
Jul 10
prev sibling parent Patrick Schluter <Patrick.Schluter bbox.fr> writes:
On Monday, 10 July 2017 at 18:45:34 UTC, Nick Sabalausky 
(Abscissa) wrote:
 On 07/10/2017 02:16 PM, Joakim wrote:
 I'm actually skeptical of cloud- I think mobile p2p will eat 
 most of the cloud-
I've been REALLY hoping p2p will eat...cloud^H^H^H^H^Hcentralized internet services[1], but if I were a betting man I'd bet heavily against it. For one thing, for p2p to kill "cloud" we'd realistically need IPv6 to become much more ubiquitous, and that just isn't happening. And I think the #2 reason IPv6 isn't happening (behind plain old inertia) is that it would *allow* p2p to overtake cloud. Which brings me to the next reason I don't think p2p will kill "cloud": All the big players with all the money and the power all LOVE "cloud" because it allows them to hoard more power, control and money, whereas p2p would completely destroy that frontier for them. Also, replacing "cloud" with p2p would mean more reliance on user's devices actually having decent storage and upload bandwidth, but non-power-users (ie the vast majority of people, if you don't live in hipster valley) are ambivalent towards that, and it would raise the price of their devices, AND they don't want to deal with running low on storage, or backing things up, so they love "cloud" too. Both the engineer and the humanitarian in me both REALLY want to see p2p eat "cloud", but I just don't see it realistically happening. [1] The word "clould" bugs me to no end. It's the tech sector's equivalent to "smurf" - stupid word is used to mean *anything* internet-releated, even "internet" itself.
May be this companay can make it happen http://www.piedpiper.com/ ;-)
Jul 12
prev sibling next sibling parent reply Andy Smith <andyrsmith gmail.com> writes:
On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to share 
 you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
Wow - very cool stuff. Wish I could have heard the talk - was it recorded at all? D could really benefit from some sort of 'project X' that will propel it to the next level. (Much like rails did for ruby, numpy/pandas did for python, akka/spark for scala). A D re-implementation of the mesos/akka/raft stack was one of the things that could be it. Best of luck with the project. I'll be watching it with a lot of interest! Cheers, A.
Jul 10
parent =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Monday, 10 July 2017 at 20:45:01 UTC, Andy Smith wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
Wow - very cool stuff. Wish I could have heard the talk - was it recorded at all? D could really benefit from some sort of 'project X' that will propel it to the next level. (Much like rails did for ruby, numpy/pandas did for python, akka/spark for scala). A D re-implementation of the mesos/akka/raft stack was one of the things that could be it. Best of luck with the project. I'll be watching it with a lot of interest! Cheers, A.
In general, the videos of MesosCon will be uploaded to Youtube within one or two month. But unfortunately, I gave the talk in Chinese since the content of this slides is relatively rich and my English is poor:).
Jul 12
prev sibling next sibling parent =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to share 
 you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
Thanks to your attention! As this slides exposed: I do have some ideas about implement the data center in D language via kernel space/user space/runtime repartition and redesign, the corresponding experiments are on the way, but many technical challenges still need to be overcome, I will update relevant information timely.
Jul 12
prev sibling next sibling parent reply Iain Buclaw <ibuclaw gdcproject.org> writes:
On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to share 
 you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!
GDC: - complete support armel, armhf - partial or bare-metal only support aarch64 Only bare-metal? As far as I'm concerned, all targets have full compiler support. The druntime and phobos libraries less so on that front. ;-) Iain.
Jul 13
parent reply =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Thursday, 13 July 2017 at 19:32:07 UTC, Iain Buclaw wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!
GDC: - complete support armel, armhf - partial or bare-metal only support aarch64 Only bare-metal? As far as I'm concerned, all targets have full compiler support. The druntime and phobos libraries less so on that front. ;-) Iain.
Such declaration is from the official D compilers wiki: https://wiki.dlang.org/Compilers It says the aarch64 support in GDC is not "Complete"... Since D front-end for GCC 8 is on the way, maybe we could replace GDC with GCC soon...
Jul 15
parent reply Iain Buclaw <ibuclaw gdcproject.org> writes:
On Saturday, 15 July 2017 at 15:40:38 UTC, 鲜卑拓跋枫 wrote:
 On Thursday, 13 July 2017 at 19:32:07 UTC, Iain Buclaw wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!
GDC: - complete support armel, armhf - partial or bare-metal only support aarch64 Only bare-metal? As far as I'm concerned, all targets have full compiler support. The druntime and phobos libraries less so on that front. ;-) Iain.
Such declaration is from the official D compilers wiki: https://wiki.dlang.org/Compilers It says the aarch64 support in GDC is not "Complete"...
Wikis tend to be a little behind, in any case I think baremetal probably doesn't quite give that list justice. None of the packages available on Debian (as listed by the wiki) are baremetal targets, they all for Linux. There's a few more levels of language support that aren't made obvious: - Has a compiler. - Can build druntime. - Passes all compilable tests in the testsuite. - Passes druntime unittester - Can build phobos. - Passes all runnable tests in the testsuite. - Passes phobos unittester Granted that the top marks only goes to x86 and Arm. The others are not far behind, and have at least the first three covered.
 Since D front-end for GCC 8 is on the way, maybe we could 
 replace GDC with GCC soon...
D frontend for GCC and GDC are the same thing. ;-) Iain.
Jul 15
parent =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Saturday, 15 July 2017 at 16:16:01 UTC, Iain Buclaw wrote:
 On Saturday, 15 July 2017 at 15:40:38 UTC, 鲜卑拓跋枫 wrote:
 On Thursday, 13 July 2017 at 19:32:07 UTC, Iain Buclaw wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!
GDC: - complete support armel, armhf - partial or bare-metal only support aarch64 Only bare-metal? As far as I'm concerned, all targets have full compiler support. The druntime and phobos libraries less so on that front. ;-) Iain.
Such declaration is from the official D compilers wiki: https://wiki.dlang.org/Compilers It says the aarch64 support in GDC is not "Complete"...
Wikis tend to be a little behind, in any case I think baremetal probably doesn't quite give that list justice. None of the packages available on Debian (as listed by the wiki) are baremetal targets, they all for Linux. There's a few more levels of language support that aren't made obvious: - Has a compiler. - Can build druntime. - Passes all compilable tests in the testsuite. - Passes druntime unittester - Can build phobos. - Passes all runnable tests in the testsuite. - Passes phobos unittester Granted that the top marks only goes to x86 and Arm. The others are not far behind, and have at least the first three covered.
 Since D front-end for GCC 8 is on the way, maybe we could 
 replace GDC with GCC soon...
D frontend for GCC and GDC are the same thing. ;-) Iain.
Thank you very much for the clarification:)!
Jul 15
prev sibling parent reply Dsby <dushibaiyu yahoo.com> writes:
On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to share 
 you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
我也希望能用D做出来。 我们也在研究过raft, akka这些技术。 raft算法的翻译我同事也有个port到D版本的。只是具体什么样,我没参与也没关心。 我们也在上海,我看链接里介绍,你现在也在上海工作,在EMC? 只是不知道你们开始做没?
Jul 13
next sibling parent =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Friday, 14 July 2017 at 05:07:11 UTC, Dsby wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
我也希望能用D做出来。 我们也在研究过raft, akka这些技术。 raft算法的翻译我同事也有个port到D版本的。只是具体什么样,我没参与也没关心。 我们也在上海,我看链接里介绍,你现在也在上海工作,在EMC? 只是不知道你们开始做没?
有更多人关注并实现基于D语言的数据中心真是太好了! 若你们能公开相关Raft算法的D语言实现那更好了:)! 我近期已经离开EMC和上海了,不过还是经常有机会去上海的。 就像上文和幻灯片中描述地那样:目前DMesos的开发者只有我一个,DLMDB我在写,希望能在1,2个月内公开出来。但我主要精力会放在底层实现上面(操作系统和运行时级别的重构),希望能有更多人能帮 实现DMesos的user space部分,因为这部分的工作量非常大。 让我们一起努力吧:)!
Jul 15
prev sibling parent reply =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Friday, 14 July 2017 at 05:07:11 UTC, Dsby wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 Dear all,
    I am a D-language amateur from China, and just want to 
 share you with a slides from me that post on MesosCon Asia 
 2017(Beijing):
    
 https://mesosconasia2017.sched.com/event/AZc6/dmesos-not-only-a-re-implementation-of-mesos-ce-feng-li-emc#
    I do really wanna to implement the design or "dream" as 
 described in this slides,
 your help and suggestion are very welcome!
    Thanks!

 BRs,
 Koo
我也希望能用D做出来。 我们也在研究过raft, akka这些技术。 raft算法的翻译我同事也有个port到D版本的。只是具体什么样,我没参与也没关心。 我们也在上海,我看链接里介绍,你现在也在上海工作,在EMC? 只是不知道你们开始做没?
做DMesos的目的一是Just for Fun, 二是在实战中提高D语言的功力。 D语言若能有更多一些大项目,则能更好地扩大其影响力并促进进一步的发展:)。
Jul 15
parent reply Dsby <dushibaiyu yahoo.com> writes:
On Saturday, 15 July 2017 at 15:57:18 UTC, 鲜卑拓跋枫 wrote:
 On Friday, 14 July 2017 at 05:07:11 UTC, Dsby wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 [...]
我也希望能用D做出来。 我们也在研究过raft, akka这些技术。 raft算法的翻译我同事也有个port到D版本的。只是具体什么样,我没参与也没关心。 我们也在上海,我看链接里介绍,你现在也在上海工作,在EMC? 只是不知道你们开始做没?
做DMesos的目的一是Just for Fun, 二是在实战中提高D语言的功力。 D语言若能有更多一些大项目,则能更好地扩大其影响力并促进进一步的发展:)。
D 的确缺少大点的项目。 那个raft实现具体我没关心。 而且最近我在公司推nogc呢。
Jul 15
parent =?UTF-8?B?6bKc5Y2R5ouT6LeL5p6r?= <hkli2012 126.com> writes:
On Sunday, 16 July 2017 at 03:45:09 UTC, Dsby wrote:
 On Saturday, 15 July 2017 at 15:57:18 UTC, 鲜卑拓跋枫 wrote:
 On Friday, 14 July 2017 at 05:07:11 UTC, Dsby wrote:
 On Monday, 10 July 2017 at 17:29:01 UTC, 鲜卑拓跋枫 wrote:
 [...]
我也希望能用D做出来。 我们也在研究过raft, akka这些技术。 raft算法的翻译我同事也有个port到D版本的。只是具体什么样,我没参与也没关心。 我们也在上海,我看链接里介绍,你现在也在上海工作,在EMC? 只是不知道你们开始做没?
做DMesos的目的一是Just for Fun, 二是在实战中提高D语言的功力。 D语言若能有更多一些大项目,则能更好地扩大其影响力并促进进一步的发展:)。
D 的确缺少大点的项目。 那个raft实现具体我没关心。 而且最近我在公司推nogc呢。
嗯 D在内存管理这块相比其他商业化已经成熟的主流语言确实还有一定差距。
Jul 18