Results 51 to 64 of 64
Thread: 80% rule in datacenters?
-
05-05-2012, 03:33 PM #51Problem Solver
- Join Date
- Mar 2003
- Location
- California USA
- Posts
- 13,681
With my colo stuff I am doing, I designed what I am doing to leave me a few amps spare for future use. It only costs a small amount of money for me to do, and doesn't hurt my margins at all.
Steven Ciaburri | Industry's Best Server Management - Rack911.com
Software Auditing - 400+ Vulnerabilities Found - Quote @ https://www.RACK911Labs.com
Fully Managed Dedicated Servers (Las Vegas, New York City, & Amsterdam) (AS62710)
FreeBSD & Linux Server Management, Security Auditing, Server Optimization, PCI Compliance0
-
05-05-2012, 03:33 PM #52Problem Solver
- Join Date
- Mar 2003
- Location
- California USA
- Posts
- 13,681
Steven Ciaburri | Industry's Best Server Management - Rack911.com
Software Auditing - 400+ Vulnerabilities Found - Quote @ https://www.RACK911Labs.com
Fully Managed Dedicated Servers (Las Vegas, New York City, & Amsterdam) (AS62710)
FreeBSD & Linux Server Management, Security Auditing, Server Optimization, PCI Compliance0
-
05-05-2012, 03:44 PM #53Aspiring Evangelist
- Join Date
- Apr 2008
- Location
- Tulsa, OK, USA
- Posts
- 376
Using power at 16 amps is legal, but continuous loads greater than that are not.
Non-continuous load periods are 'three hours or less.' Any load for a period longer than that must be classified as continuous.
Following the rules above, "bursting" would be load periods of 'three hours or less.' But a lot of datacenters I have worked with treat power usage like data usage and have some other strange definition of "bursting." So, YMMV.
What datacenter is this? Also, a lot of PDUs are not entirely accurate on their power consumption data. Yours may be under-reporting, or the DC PDUs may be slightly over-reporting.
If the datacenter has someone who is not afraid of electricity, the most accurate way to determine who is in the wrong here would be to stick two probes onto the circuit and measure the EM resistance between the probes, using ohm's law.
How many servers are you running on this circuit?0
-
05-05-2012, 03:49 PM #54
Our PDU only reads out to the nearest amp, and I could certainly be convinced it's not even 100% accurate there. I got a graph showing I was reasonably well over 16a (around 16.5) for a significant portion of the last month, so I'll go ahead and reduce my usage shortly.
Looking at racktables, it looks like we've got 18 servers and 3 ethernet switches on this circuit. From previous experience, one of the 3 ethernet switches uses about 1.5a, the second one uses a little less than 1a, and the third uses about 0.3a. So really doing quite well in terms of amps used for each server, with 18 servers using a little under 14a it would seem. Go go gadget E3 Xeons.IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers in under an hour
★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
Contact Us: sales@ioflood.com ★0
-
05-05-2012, 03:51 PM #55Aspiring Evangelist
- Join Date
- Apr 2008
- Location
- Tulsa, OK, USA
- Posts
- 376
240.4(B) means that if you have a circuit rated for 16 amps, then it must be able to carry 20 amps for a non-continuous load period, not that you can go over 16 amps continuously.
[Edit: To clarify: the circuits you get from the DC are rated at 16 amps, not at 20 amps. Many people sell circuit capacity based on maximum non-continuous load rate. In reality, you're being sold a 16A circuit and you're allowed to use 100% of that 16A. This is kind of like dedicated RAM verses burst RAM on OpenVZ -- the burst RAM may be available to you, but you shouldn't be using it constantly.]
NEC and IEEE standards define going over the rated continuous load (even for a non-continuous load period) may cause undefined behaviour in the power equipment. This may translate to deterioration of the circuit breakers, making them more likely to trip in the future.
Given that overcurrent state on a circuit even for a non-continuous load period may cause undefined damage, it is not unreasonable for both the building and the datacenter itself to impose rules designed to protect their distribution equipment.Last edited by kaniini; 05-05-2012 at 04:04 PM. Reason: clarification
0
-
05-05-2012, 03:59 PM #56Aspiring Evangelist
- Join Date
- Apr 2008
- Location
- Tulsa, OK, USA
- Posts
- 376
A good rule is to allocate 1 amp for each device, yielding a maximum of 16 devices per circuit. Even if you are undershooting as a result, it will ensure that your circuit won't trip.
Limiting the number of devices per circuit is also a good idea because crappy PSUs may cause line noise which could damage the equipment.0
-
05-05-2012, 04:01 PM #57Virtually Flawless ;)
- Join Date
- Apr 2009
- Location
- USA / UK
- Posts
- 4,577
So conforming to the requirements set out in the National Electric code is not responsible?
You should have quit while you were ahead.
No, I simply prefer not to have to be concerned about power overages, or a datacenter claiming I used more power than I did.
I prefer flat rate network ports for the same reason. I like predictability.
You should not assume just because I have a certain opinion, that automatically means I must be abusing resources, or assuming I must somehow be violating a datacenter's AUP/TOS.
Yes, this is true. Likewise, however, nobody should assume all datacenters define rules regarding this that go beyond what the law says.
Same thing. You should not assume that all datacenters set usage rules that go beyond what the law requires.
This is a portion from an Equinix contract I have been privy to:
A Customer may order power to the extent offered and approved by Equinix. Power Circuit Policy: A power circuit at all times meets following conditions: 1) The aggregate draw of such Power Circuit does not exceed the rated capacity pursuant to the National Electrical Code.Last edited by ramnet; 05-05-2012 at 04:12 PM.
0
-
05-05-2012, 04:04 PM #58
We find that we routinely use dramatically less than 14a on a circuit if we only have 16 outlets, we switched to 24 outlet PDUs and will never look back. It's extremely difficult to balance a load properly with only 16 outlets. You'll have one PDU hit 16 with only 12 outlets, and the other with 10a of use and no outlets free. Then you either have to put up with paying for 5a of power you're not using (1/3 of a circuit essentially), or you have to play the game where you hope you can move things around without causing any problems. Really not a game worth playing.
On this PDU for example, we're using 22 outlets, 18 for servers, 3 for switches, and 1 for a wifi access point. If we need to hook up a crash cart it would be nice to be able to do that too, and luckily we have a spare outlet for that. Impossible on a 16 outlet pdu. Even with our current overusage, we could move simply one device (the 1.5a power hungry switch) off of the circuit and then be far enough under our allowed power to safely hook up another of these 0.75a servers we love so much, so we'd still be at 22 outlets either way.IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers in under an hour
★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
Contact Us: sales@ioflood.com ★0
-
05-05-2012, 04:06 PM #59Virtually Flawless ;)
- Join Date
- Apr 2009
- Location
- USA / UK
- Posts
- 4,577
While I won't dispute that, I will add that
Underwriters Laboratories (UL) tests all thermal circuit breaker designs to ensure they can handle 100% of the stated load on the breaker in open air at 40C ambient temperature.
A 20 amp breaker is actually capable of handling 20 amps. Just not constantly (as NEC rules restrict this further).0
-
05-05-2012, 04:11 PM #60
After being able to actually look at the NEC rules, it appears to be a little more strict than you're making them out to be.
The circuit size needs to be
1) Equal to 1.25 times the size of the continuous load
PLUS
2) Equal to 1 times the size of the non-continuous load
If you never drop below 14a of usage, then your continuous load is at least 14a. 1.25 times 14a is 17.5a. If you have a 20a circuit, then your "non continuous load" is allowed to be an extra 2.5a from this point. Meaning your bursts should be no more than 16.5a according to NEC under this scenario. Not even close to "tripping the breaker" as you seem to think you should be able to do.
I do appreciate everyone in this thread pointing me in the right direction such as to the relevant electrical codes, definition of continuous / non continuous etc. I also appreciate the datacenter getting back to me with a proper usage graph so I can square it against those NEC rules.
However, ramnet, even though it sounds like you're "on my side", really it just feels like trolling. I should know, I'm good at it. "oh wait, we can really get this thread going if I claim it's every customer's right to trip their breakers, bet that's good for a half a dozen replies".
Mods can please close this thread, I've gotten the information I needed, thanks.IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers in under an hour
★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
Contact Us: sales@ioflood.com ★0
-
05-05-2012, 04:16 PM #61Virtually Flawless ;)
- Join Date
- Apr 2009
- Location
- USA / UK
- Posts
- 4,577
What your continuous load may be, and what the maximum allowable continuous load is, is different.
If you have a 20 amp breaker, max continuous load is 16 amps. This is not questioned by anyone (20 / 1.25).
16 amps continuous load times 1.25 allowed spike load is, guess what? 20 amps - the size of the breaker.
If your datacenter doesn't define operating requirements in their environment stricter than those defined by laws / the NEC, then you should be able to spike non-continuous load to 20 amps without an issue.0
-
05-05-2012, 04:20 PM #62
Nope, that's wrong. If you continually use 16a, it means you REQUIRE a 1.25 times sized circuit / breaker in order to be allowed to do that, leaving ZERO room for bursting on a 20a circuit. If, however, you have ZERO continuous load, you may then use the entire circuit size for your non continuous load.
Multiply continuous load times 1.25
add non continuous load
Result of the two added together must be less than rated circuit / breaker size.
In my case, I'm never using under 14a on that circuit, so there is no room for argument that my continuous load is under 14a. 14 times 1.25 = 17.5. If I'm using a 20a breaker, this means there is 2.5a "left over" for non continuous loads, which would mean, a total of 16.5a.IOFLOOD.com -- We Love Servers
Phoenix, AZ Dedicated Servers in under an hour
★ Ryzen 9: 7950x3D ★ Dual E5-2680v4 Xeon ★
Contact Us: sales@ioflood.com ★0
-
05-05-2012, 04:34 PM #63Virtually Flawless ;)
- Join Date
- Apr 2009
- Location
- USA / UK
- Posts
- 4,577
The NEC code does actually provide examples on how to size a circuit. It's fairly easy to follow.
The 125% sizing of the Over Current Protection Device (or 80% loading) is only applicable when continuous loads are involved. Circuit Breakers and other OCPDs can be sized at 100% of their rating for noncontinuous load applications.
What this means is, if you mix continuous and non-continuous loads on the same circuit, the circuit must be rated for the larger 125% sizing, and not simply 100% sizing. You do not add these together to come up with 225%
It is extremely rare for continuous and non-continuous loads to be on the same circuit, which is why this isn't very clear from reading the NEC as it assumes you will use separate circuits for the different loads types. If you mix loads you use the stricter continuous load definition at 125%, otherwise you use the 100% definition. You don't use both definitions at the same time.Last edited by ramnet; 05-05-2012 at 04:48 PM.
0
-
05-05-2012, 06:23 PM #640
Similar Threads
-
UK Datacenters
By VN-Ken in forum Colocation, Data Centers, IP Space and NetworksReplies: 31Last Post: 09-15-2006, 12:23 PM -
datacenters?
By deskbar builder in forum Web HostingReplies: 4Last Post: 10-25-2004, 03:19 PM -
Iptable firewall rule for icmp ping rule
By Cephren in forum Hosting Security and TechnologyReplies: 2Last Post: 08-24-2004, 05:59 PM -
DataCenters
By Real_FAN in forum Dedicated ServerReplies: 5Last Post: 03-23-2004, 06:16 PM -
Datacenters...
By IGobyTerry in forum Web Hosting LoungeReplies: 1Last Post: 09-30-2002, 01:17 AM