Telcos are now turning to the Open Compute Platform

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

RobertFontaine

Active Member
Dec 17, 2015
663
148
43
57
Winterpeg, Canuckistan
Very cool. While the open compute gear looks cheap, it removes significant cost from the hardware both in branding and in overbuilding.

With 3 year turnover cycles, having hardware designed to last 20 is a recycling challenge.

As one of the beneficiaries of overbuilt hardware I will be sad if corporate gear ends up built to the minimal requirement but it does make a lot of sense both financially and environmentally to take this path.
 
  • Like
Reactions: gigatexal

cesmith9999

Well-Known Member
Mar 26, 2013
1,417
468
83
does OCP have standards that fit 19" 2 post racks? I do not see them retrofitting anything that is out there. it seems like there will be a hard cost up front for changing standards.

the MS OCP contributions are too deep to be considered for 2 post racks

Chris
 

PigLover

Moderator
Jan 26, 2011
3,184
1,545
113
does OCP have standards that fit 19" 2 post racks? I do not see them retrofitting anything that is out there. it seems like there will be a hard cost up front for changing standards.

the MS OCP contributions are too deep to be considered for 2 post racks

Chris
No - there is not and there are no plans to add support for 2-post racks in OCP.

2-post racks exist throughout the legacy wire centers and in the past there were some limited active elements placed in them (showing my age here, things like digital channel banks and then moving to routers and lightweight servers). As long as there is wire in the plant this will continue.

However, the part of the telecom business that cares about OCP already looks like a big data center, using standard data center racking, hot-cold aisle cooling optimizations, etc.

The next generation of Telco office will be virtually indistinguishable from any other highly virtual, cloud centric data center. They have many of the same problems to solve as the big ISP hosts like FB, Google, etc. There are a few complications to be addressed that for now require a separate Telecom working group - but I think over time even these differences will disappear.
 
Last edited:

PigLover

Moderator
Jan 26, 2011
3,184
1,545
113
But far better to have them in generic designs that are multi-sourced/open-sourced. Long term you've got a lot better resale/surplus market for solder-on CPUs in OCP modules than say solder-on CPU chips in proprietary designs like HP or IBM blades or Moonshot modules.
 

PigLover

Moderator
Jan 26, 2011
3,184
1,545
113
Also, don't forget that the Quanta switches we see flooding the market today represent the future. With OCP you'll see a lot more open/whitebox switch and router platforms too - which for the after-market means newer/faster inerconnect solutions should be available more quickly and at better prices.
 
  • Like
Reactions: Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
Also, don't forget that the Quanta switches we see flooding the market today represent the future. With OCP you'll see a lot more open/whitebox switch and router platforms too - which for the after-market means newer/faster inerconnect solutions should be available more quickly and at better prices.
I think that is true, however, the Quanta BMS switches require some sort of software to sit on them. I got the Sunnyvale lab's Quanta switches with the Quanta OS but the BMS ones would require Cumulus/ Big Switch or something.
 

PigLover

Moderator
Jan 26, 2011
3,184
1,545
113
The BMS switches are open designs with uboot, right? Don't loose context here - the CPUs and motherboards you buy used today require an OS to make them useful too and Linux/OmniOS/etc are widely available. That is the future view. If companies like BigSwitch does not go opensource they will likely die. Their future is in selling services (ala RedHat/Canonical/Mirantis/etc).
 

bds1904

Active Member
Aug 30, 2013
271
76
28
What most people don't know is that telco's that have video services already use an all 4 post model in their data centers that serve/process video. When I say all 4 post, I mean it. Even the routing and fiber distribution is 4 post.

That being said, all the new hardware is 19" and requires adapters to be fit into 23" racks. The new routing platforms I've seen installed recently are even installed in 19" 2 post racks. All-IP switches (providing legacy POTS service), DSLAM's, OLT's and new fiber patch panels have been coming in 19" for years now.

Having knowledge of the space and design of a central office, changing to a 4 post standard is not difficult at all. Decommissioning & consolidating legacy TDM POTS switches frees up plenty of space as they are typically 40" deep. A new IP switch that takes up 42u replaces 6 rows of equipment. Each row is 20' long. The single 42u switch serves the same number of customers also. Do that math... Getting rid of legacy TDM switches also frees up hundreds of amps of power.

The issue with opencompute hardware will be DC/DC power supplies, I'm not sure that have a standard for dual 48V feeds. Central offices have no AC power running to equipment, and there is no sense in running it. The DC power setup for most large offices is way more than enough for an entire datacenter. In fact the datacenter that was built 10 years ago locally to provide video services to a large area uses 1/3rd the power of a single office.
 
Last edited:
  • Like
Reactions: T_Minus

Patrick

Administrator
Staff member
Dec 21, 2010
12,511
5,792
113
The BMS switches are open designs with uboot, right? Don't loose context here - the CPUs and motherboards you buy used today require an OS to make them useful too and Linux/OmniOS/etc are widely available. That is the future view. If companies like BigSwitch does not go opensource they will likely die. Their future is in selling services (ala RedHat/Canonical/Mirantis/etc).
You are 100% correct! When I looked in mid-2015 there was no real good open source distributions out there for the switches. If there was that would have saved me a ton!

What I think we really need is one of these SDN guys to do Open Source + a WebGUI management plane. Then you make the BMS model attractive for SMBs as well.
 

mstone

Active Member
Mar 11, 2015
505
118
43
46
The issue with opencompute hardware will be DC/DC power supplies, I'm not sure that have a standard for dual 48V feeds.
The 48V power supplies should be the first standard out of the telco project, and it really shouldn't take very long. The equipment is mostly premised on 12V power distribution, so you just replace the AC/DC chassis power supplies with smaller/simpler DC/DC supplies to step the incoming 48V to 12V internally. I actually see that as one of the big draws for the telcos: this makes it easier for them to buy the same commodity hardware as everyone else instead of specially priced "telco" versions which are almost identical but with different power input.
 

bds1904

Active Member
Aug 30, 2013
271
76
28
The issue with opencompute hardware will be DC/DC power supplies, I'm not sure that have a standard for dual 48V feeds. Central offices have no AC power running to equipment, and there is no sense in running it. The DC power setup for most large offices is way more than enough for an entire datacenter. In fact the datacenter that was built 10 years ago locally to provide video services to a large area uses 1/3rd the power of a single office.
Forgot to add that the entire datacenter is run on the telco style dual 48V feeds to servers, comprised of mostly HP servers. The entire setup uses standard telco rectifiers and batteries, no UPS necessary.

I think the real reason telco's will have an attraction to the Open Compute Project is a standard form factor. The reason telco's had/have/use/used "proprietary hardware" is because it conformed to the necessary standards required to use it in any central office. Having a (near)industry wide standard of form factor, power supply and most importantly cooling will enable telco's to integrate hardware into a wide array of locations.

Hardware ingratiation standards are a 100% necessary evil in the telco world. Without a multi-vendor standard a roll-out of new hardware would be impossible in a central office. Facebook & Google (the big 2) don't hold a candle to the scale of a hardware build-out that telco's take on. "The big 2" have what, 50 locations between the two? Telco's have hundreds of thousands of locations to intall hardware into.
 

archmagos

New Member
Dec 16, 2014
18
1
3
Seattle, WA, USA
You are 100% correct! When I looked in mid-2015 there was no real good open source distributions out there for the switches. If there was that would have saved me a ton!

What I think we really need is one of these SDN guys to do Open Source + a WebGUI management plane. Then you make the BMS model attractive for SMBs as well.
Has anyone looked into OpenSwitch:

OpenSwitch

Could hold some promise if your switch supports an ONIE environment.
 
Last edited:

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
Not used it, would be great if it really works well, Google just released the code for their open load balancer.
Right now the project I am doing will use NSX for network and virtual F5's for LB