Cloud Operating System - Are you interested in helping?

[tsdnz]

Could you answer the following simple questions:

0) does it have a custom bootloader or it uses GRUB?
1) does it have memory management or multitasking?
2) can it load simple files from the disk and what file systems does it support?
3) does it have a proper CLI?
4) how many device drivers do you currently have?

Hi,
0 = Custom
1 = Kernel is multitasking, userspace is single threaded, multi-cored. Both have memory management.
2 = No storage yet, planning to show the core scaling functionality.
3 = No CLI, it has no UI. All controlled remotely. Small admin screens to see internals. One client wants a small command line, for the banking system.
4 = Keyboard, Ethernet (down to one, just supporting 10 GB Intel NIC).

Ali

[kzinti]

You just described your IDE and build environment. Still haven't provided any information about your OS, or answered any question.

It's all in his head, just be more positive!

[AndrewBuckley]

Forget cygwin altogether. Assume I handed you a small operating system with nothing more than a desktop, Visual studio(that I stole the source code to compile) and gcc setup for native compile. no cross compile needed. it has mouse,keyboard, video and a file system to store everything, all of the garbage you are asking for. you have the code on the machine, and it looks like windows so you know how to use it. Name a single thing you would do with this that would convince us to waste any time with it.

[Nutterts]

You're only hope is to provide some "proof of life" for your OS.

[tsdnz]

You're only hope is to provide some "proof of life" for your OS.

I agree, planning to.

Ali

[ggodw000]

i am bit confused by very definition of your "cloud OS" term.
To start with, lets define the cloud:
ability to do a live migration of VMs seamlessly between physical machines based on the resource arbitration policy. Now I am not huge expert in this but that is how I think essentially work. Vmware ESXi, MS Server with Hyper V support and other Virtualization OS should all be able to do it. So how about you start from the basic: what is your cloud OS's differentiator from others already on the market?
Those are OS that support virtualization, as far as I know i havent heard anything regarding cloud OS. For me virtualization technology is the basis and foundation of cloud.

[jojo]

It's been well over a month. I sure hope his demo went well.

[tsdnz]

Hi all,

Sorry about the "Cloud OS" def..

Yes the demo went well, thanks for reminding me.
The first investor is on board.
I have one years development money (wages), plus a development/server hut (I work from home), plus nice cash in hand (the entry fee for the investor), plus money to buy servers/equipment, and free flying money.
I hope to have it ready for the next investor(s) in three months time, but this depends on the features we will add.
The accountants and lawyers are drawing things up now, hopefully ready next week, although this does take some time.

I have just brought the first switch, a 48 port 10Gb SFP.
Just about to buy a laptop so we can develop on the move to show others, currently using a very cheap desktop as my development machine.
About to buy 20-30 servers, each with 2 x 10Gb NICs.

Then on to the next stage of development.

Parts of the project are processing packets fast, for example the cache OS can handle around 40 million per second, not ready for production, so the pps will drop.
The pps are a little lower than expected but the server I have has an old-ish PCI Bus, and I an not willing to spend money on a flash server just to test. But I hope to contact a server supplier and ask if I can borrow a server for 1/2 a day.

In the User-space OS I have increased the pps from 7 million to 9 million, with everything that needs to be done, and a slow-ish server this seems to be my limit, for now, things always pop into the head after a little time passes.
On the old servers I already have, Dell 1950, slow memory, PCI 1, 4 cores, I am processing around 4.5 million per second.

I would love to test some 40Gb NICs, but this would take time to write the drivers, which I am not willing to do right now.
But very much looking forward to the chance to do this one day.

Things are going very well for me, it is very exciting, and I get to program 24/7.
Just the best of fun.

All the best guys and gals, Ali.

[Brendan]

Hi,

I have just brought the first switch, a 48 port 10Gb SFP.
Just about to buy a laptop so we can develop on the move to show others, currently using a very cheap desktop as my development machine.
About to buy 20-30 servers, each with 2 x 10Gb NICs.

Don't.

20 to 30 identical servers is a large amount of $$$ for nothing. It'd make more sense to buy an eclectic mixture of 10 very different servers so you can test how well the OS supports different hardware and how well it copes with different conditions. Don't forget that (for testing "many computers") you can always run your OS in virtual machines (e.g. 10 servers running 20 virtual machines each = network of 200 computers running your OS).

Also note that the cash you have now is probably going to have to last for many years, and in about 12 months you're going to have to explain to your investor that there's no hope of them ever getting any return on their investment.


Cheers,

Brendan

[tsdnz]

Don't.

20 to 30 identical servers is a large amount of $$$ for nothing. It'd make more sense to buy an eclectic mixture of 10 very different servers so you can test how well the OS supports different hardware and how well it copes with different conditions. Don't forget that (for testing "many computers") you can always run your OS in virtual machines (e.g. 10 servers running 20 virtual machines each = network of 200 computers running your OS).

Also note that the cash you have now is probably going to have to last for many years, and in about 12 months you're going to have to explain to your investor that there's no hope of them ever getting any return on their investment.


Cheers,

Brendan

Cheers for the feedback Brendan, the cash is for 12 months.
The servers are just to test the OS, it should cost around $ 500 - $750 New Zealand Dollar per server.

The investor really wants to spend the money on hardware, I am a lot more conseravite, I only want to spend x dollars, and leave some for unexpected, and travel costs. It is quite funny, we talk all the time, he wants me to buy, buy, buy, send him the bill. And I am no, no, no, lets just use what we have and a few extra's.
I am used to having just enough money to keep my family feed, with a little left for extra's, and if lucky some computer stuff.
He can see the potential, he has a great point, if it helps get the project finished faster, then get it.
The hardware money is there to be used, it is only there for a year.

I do not need to test the OS on heaps of different hardware, it is for a specific task, it is not for general public use.
Although a general range of servers, like AMD/Intel 64 are supported.

I might start with 10 servers?
Still thinking about it.....

Ali

[Brendan]

Hi,

Cheers for the feedback Brendan, the cash is for 12 months.
The servers are just to test the OS, it should cost around $ 500 - $750 New Zealand Dollar per server.

What sort of server do you get for $750?

For most "cloud" providers, the purchase price of the server is fairly irrelevant, and what matters is running costs (specifically, the power consumption and air conditioning costs), reliability (things like ECC), and the physical space consumed. They tend to go for more expensive servers with high "performance per watt", not your cheap $750 "fake servers" (that are often just consumer grade/desktop stuff in a different/rack mount case).

I might start with 10 servers?
Still thinking about it.....

One thing you'd want to think about is; Intel will be releasing the next "Xeon Phi" some time in the next 6 months, and this is a chip that's almost ideal for cloud - it's a massive ~72 core chip, with MCDRAM and a huge amount of RAM bandwidth, which is probably capable of handling several hundred virtual machines in a small "blade server" configuration.


Cheers,

Brendan

[tsdnz]

What sort of server do you get for $750?

* 2 x Intel Xeons 5160 Dual Core 3.0 GHz VT-x (4 cores total)
* 16 GB DDR2 FB-DIMM RAM
* 2 x 146GB SAS HDD
* 2 x Gbit NICs
* Dual PSU
* RAID Controller card
* DRAC Card for management (which is similar to IBM RSA and HP iLO cards)
http://www.trademe.co.nz/computers/serv ... 033281.htm
http://www.trademe.co.nz/computers/serv ... 049694.htm

For most "cloud" providers, the purchase price of the server is fairly irrelevant, and what matters is running costs (specifically, the power consumption and air conditioning costs), reliability (things like ECC), and the physical space consumed. They tend to go for more expensive servers with high "performance per watt", not your cheap $750 "fake servers" (that are often just consumer grade/desktop stuff in a different/rack mount case).

Yes, I agree. For me I do have a dollar limit, so just to test I will buy cheaper models.

One thing you'd want to think about is; Intel will be releasing the next "Xeon Phi" some time in the next 6 months, and this is a chip that's almost ideal for cloud - it's a massive ~72 core chip, with MCDRAM and a huge amount of RAM bandwidth, which is probably capable of handling several hundred virtual machines in a small "blade server" configuration.

Yes, a very nice chip. I am looking forward to having a play on it when I get a chance.
I had a 10x blade server when I first started the project, this was a mistake.

Ali

[Brendan]

Hi,

What sort of server do you get for $750?

* 2 x Intel Xeons 5160 Dual Core 3.0 GHz VT-x (4 cores total)

Oh my...

These CPUs:

• were "end of lifed" 7 years ago, and because they're old (e.g. using a 64 nm manufacturing process) they're much much worse on the "performance/watt" scale than anything more recent.
• are based on Woodcrest/Core 2 where the memory controller was in the northbridge (and not built into the CPU) so RAM latencies are much higher than modern (Nehalem or later) Xeon; and with 2 of them sharing a memory controller (via. FSB) you'll have "RAM bandwidth bottleneck" issues (in addition to higher latency).
• don't support ECC, which means they're only for "fake servers" and not useful for cloud
• don't support hyper-threading

For all these reasons, they're fairly unsuitable for "cloud OS" development - anyone that is interested in your cloud OS (if/when it's finished) won't be interested in these systems. I also suspect that (for $750 each without any hard drives), they're over-priced, and for twice the price you can get something with 4 times better/more suitable.

Note that the "Dell Power Edge 1950" systems (from your second link) have most of the same problems - twice as many cores with twice as much power consumption and similar "performance/watt"; except that you'd have twice as many cores fighting for the same RAM bandwidth. They do have ECC though, so that's one thing in their favour.

Yes, I agree. For me I do have a dollar limit, so just to test I will buy cheaper models.

To test what?

To test if the OS works efficiently on computers that cloud providers are likely to use you need computers that cloud providers are likely to use.

I had a 10x blade server when I first started the project, this was a mistake.

Blade servers are for "density" (cramming as much as possible into a limited space), which is why they're more likely to be used by cloud providers than larger individual systems. Excluding that; there's very little difference between blade servers and individual machines.


Cheers,

Brendan

[ggodw000]

working on the upcoming Broadwell-EX it is a monster with nearly 200 cores.

[tsdnz]

Oh my...

These CPUs:

• were "end of lifed" 7 years ago, and because they're old (e.g. using a 64 nm manufacturing process) they're much much worse on the "performance/watt" scale than anything more recent.
• are based on Woodcrest/Core 2 where the memory controller was in the northbridge (and not built into the CPU) so RAM latencies are much higher than modern (Nehalem or later) Xeon; and with 2 of them sharing a memory controller (via. FSB) you'll have "RAM bandwidth bottleneck" issues (in addition to higher latency).
• don't support ECC, which means they're only for "fake servers" and not useful for cloud
• don't support hyper-threading

For all these reasons, they're fairly unsuitable for "cloud OS" development - anyone that is interested in your cloud OS (if/when it's finished) won't be interested in these systems. I also suspect that (for $750 each without any hard drives), they're over-priced, and for twice the price you can get something with 4 times better/more suitable.

Note that the "Dell Power Edge 1950" systems (from your second link) have most of the same problems - twice as many cores with twice as much power consumption and similar "performance/watt"; except that you'd have twice as many cores fighting for the same RAM bandwidth. They do have ECC though, so that's one thing in their favour.

Thanks for responding, you make excellent points. You know a lot more about the hardware than I do. What hardware would you suggest?
I would like:

• 4-8 cores
• Memory not an issue, minimum 8GB, quick-ish, ECC
• 64 Bit, SSE2+, would like AVX, PCIe 2+ 8xSlot
• 2.0 GHz +
• And cost efficient

I was planning a trip to the server provider, taking a test OS with me and testing different servers.
I thought that between them and me testing, and the costs, we would come home with something.
If you can help with what you think is required that would be great, I might be able to Skype them and save one days travel.

To test what?

To test if the OS works efficiently on computers that cloud providers are likely to use you need computers that cloud providers are likely to use.

To test the code I need multiple servers running, a minimum of 5 just to get it started. Currently I have code running as separate "OS's" inside an OS....., this was fine but I am at the stage where this is impossible to do. (Things happened in life that reduced my money, time and assets. Not a divorce, very weird experience.... Now to rebuild)

Blade servers are for "density" (cramming as much as possible into a limited space), which is why they're more likely to be used by cloud providers than larger individual systems. Excluding that; there's very little difference between blade servers and individual machines.

NICs are shared between the blades, this does not matter now, a lighting strike about 2 years ago took care of them.

Thanks for all your input, it is greatly appreciated, Ali.