Slashdot Mirror
The Data Center Density Debate: Generational Change Brings Higher Densities (datacenterfrontier.com)
1sockchuck writes: Over the past decade, there have been repeated predictions of the imminent arrival of higher rack power densities. Yet extreme densities have remained focused in high performance computing. Now data center providers are beginning to adapt their designs for higher densities. One of these companies is Colovore, which is among a cliuster of companies adopting chilled-water cooling doors for their cabinets (LinkedIn is another). They say the move to higher densities is driven in part by a generational change in IT teams, as younger engineers are less worried about high-density strategies using water in the data center. "A lot of them grew up with PC gaming and water cooling right in their living room," said a Colovore executive.
How reliable is it, really? What happens if a pump breaks down in a high-density liquid cooled rack versus a breaking fan in a low-density air cooled rack?
How likely are those breakages? At the every least they are less than 100% reliable; they WILL break at some point. It might be perfectly fine, but a single PC at home does not scale to an entire datacenter of racks.
Slashdot social media options: AIM, ICQ, Yahoo, Jabber and Mobile Text. Why no MySpace?
Don't bother. The article is clickbait. The submitter has a history of frequently submitting articles from the same website. I suspect 1sockchuck is affiliated and not disclosing that fact.
"National Security is the chief cause of national insecurity." - Celine's First Law
Each new generation of electronics has a higher density, so this isn't really news.
During the 70's a computer room was huge with a lot of stuff physically. Today that computing power fits in your pocket.
In addition to this we have more and more data stored electronically, so the increased density is somewhat counteracted by that.
Another factor is that earlier there were only a few terminals locally, but today we have networks giving every household a connection where a large number of terminals can reside. Soon your toaster is on the internet as well. All that counteracts higher density in the data centers because they need to be able to manage more connections and a higher traffic volume every year.
And even when you get a higher density you need to cool the electronics in a way that avoids hot spots in the electronic devices shortening their life span.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
Speaking from experience here running an HPC system with water cooled doors on the back of my racks for just shy of four years now. My personal view point is if you are not doing racks with water cooled doors then you need the sack unless you are using one of the in rack water cooling systems.
There is no pump in the rack/door. You hook up to the cool water supply that is coming from the chillers, and they provide the "pump". You are no worse off than before with air handling units in the room which also don't have pumps in them.
The doors we use have a number of large aka ~400mm diameter fans in the back that spin relatively slowly, and these are N+1 redundant. Besides that they are the sorts of fans than can spin for decades before going wrong.
The other advantage of having the cooling in the doors is that you save floor space in the room as you don't have to so many air handling units in the room.
The first point to note is you probably still need some air cooling in the room. Things like tape libraries don't come with water cooled doors for example. Though there is an argument that the tape libraries should go somewhere other than you data centre.
The second problem is suppliers that want to sell you something prepackaged in a rack that is not water cooled. Trying to explain to them that you don't want their shitty none water cooled rack is in my experience is like taking to a brick wall. Another reason why you probably still need air handling units. That said our doors are currently providing net cooling to the room, that is inlet temperature at the front of the rack is higher than the outlet temperature at the back.
The third problem is strip off to tee shirt and shorts if you open the rack door while the system is going full tilt, it can be sufficiently hot.
Now go crawl back under your rock and stop spouting nonsense about things you clearly know nothing about.
So those of us that still have a lot of empty space available and don't have to change to high density yet "need the sack"?
WTF is it with the insulting zealotry? It completely demolishes credibility, making any of your points questionable whether they are correct or not.
Also you've said nothing useful about water treatment, condensation, descaling or corrosion. They used to be a pretty big deal back in the day and are today in other systems cooled with water, why not worth mentioning now?
Redundancy for pumps is easy at data center scales, but the very real problem over non-trivial equipment lives is leaks.
Way back when, we used concrete pumping hoses for chilled water to CRAC units, to facilitate relocation of the units and reduce the risk of leaks after an earthquake. Worked pretty well, they had a huge safety factor compared to concrete pressures, and they were easy to test before placing into service. Seismic performance was very predictable, and it all worked well. Then, a bad (metallurgicly) batch of fittings came in, and two or three years into operation the fittings separated from the hoses and two or three floors of data center were flooded.
Same things have happened to copper pipes, victaulic fittings and welded connections on steel pipes, PEX pipes... everything imaginable.
Water cooling is great, but there are a lot of weak links in the chain from a risk management perspective.
Shame nothing ever came of HP's technology to "print" water droplets on the die and use some latent cooling in data centers.
Nothing requires water cooling, it is just a water cooled door on the back of the rack gives us a better PUE, we don't waste floor space on air handling units and we don't have to worry about getting all the air containment for hot/cold isles either.
Off the top of my head we are about 35kW per rack we the nodes are all spun up, though of course 100% utilization is super hard because we are always having nodes idling while to gather enough free for the next job to run. Nodes are about 85-90% busy.
"A lot of them grew up with PC gaming and water cooling right in their living room,"
A lot of us grew up working on water cooled mainframes right in our own data centers.
What's that saying about 'what's old is new again...' Now get off my lawn.