Over the last couple of days I've been doing some tests of temps. The issue of datacenter cooling has come up again, and I want more data. I strongly suspect that some past-practices we've inherited don't scale well to pizza-box and blade-server densities. Heck, when I got here in December of 2003 the only rack-dense servers we had in the datacenter were the three student nodes in the NetWare cluster (2U servers, even though 1U would have worked), and a pair of Active Directory root-domain DC's. Since then we've installed 19-20 blade-servers, converted a whole bunch of circa 2000 server hardware to be test and development servers, and picked up a couple more 1U and 2U servers as well.
One of the bigger problems in my opinion are the doors to the racks. The front doors are plexiglass fronted doors with maybe a 5-10% perforation, and the back doors are big metal doors with perfs at the top and bottom for maybe 2% perforation if that much. To handle airflow, there are big ole fans in the tops of the racks. This setup was just fine for whitebox and other machinery that isn't rack-dense.
Just by changing the front door to the blade-rack (back when it was 16 blade-servers) from the plexi-front to a fully-perfed door we dropped exhaust temps 10-15 degrees. There was recirculation going on there!
Right now it looks like rack-ambient temps are about 10 degrees F above room-ambient. When the rack-fans are turned off the increase goes to 18-20 degrees F, so those rack-fans are doing work. If I open the front door and leave it there, rack-ambient drops to room-ambient. Clearly, a fully perforated door is what we want.
On the other hand, even with the rack-fan off the servers are still within their operating spec. But then, who wants to run near the top of the spec when these things are supposed to last 5-6 years. Perhaps this data will convince TPTB to liberate fundage to start converting our doors to the fully perforated ones.
Another area we'd get benefit from once the doors are in: blanking panels. We're getting noticible recirculation. Happily, the heat of the recirculated air is not enough to run the servers out of spec, but recirculation is an abomination unto Nuggen and can not be tolerated. Ahem. We'll do that once we yank the CRTs out of the racks too. THEN we'll be sitting pretty for true rack-dense stuff!
One of the bigger problems in my opinion are the doors to the racks. The front doors are plexiglass fronted doors with maybe a 5-10% perforation, and the back doors are big metal doors with perfs at the top and bottom for maybe 2% perforation if that much. To handle airflow, there are big ole fans in the tops of the racks. This setup was just fine for whitebox and other machinery that isn't rack-dense.
Just by changing the front door to the blade-rack (back when it was 16 blade-servers) from the plexi-front to a fully-perfed door we dropped exhaust temps 10-15 degrees. There was recirculation going on there!
Right now it looks like rack-ambient temps are about 10 degrees F above room-ambient. When the rack-fans are turned off the increase goes to 18-20 degrees F, so those rack-fans are doing work. If I open the front door and leave it there, rack-ambient drops to room-ambient. Clearly, a fully perforated door is what we want.
On the other hand, even with the rack-fan off the servers are still within their operating spec. But then, who wants to run near the top of the spec when these things are supposed to last 5-6 years. Perhaps this data will convince TPTB to liberate fundage to start converting our doors to the fully perforated ones.
Another area we'd get benefit from once the doors are in: blanking panels. We're getting noticible recirculation. Happily, the heat of the recirculated air is not enough to run the servers out of spec, but recirculation is an abomination unto Nuggen and can not be tolerated. Ahem. We'll do that once we yank the CRTs out of the racks too. THEN we'll be sitting pretty for true rack-dense stuff!