Focusing on process technology by the choice of memory in shared services of webhosting industry contributes dramatically to save power
May. 2010
By Peyman Blumstengel;
Samsung Semiconductor Europe,
- Objective: how does the right choice of memory could contribute to power saving efforts in shared services of a real Webhosting provider.
Saving Power in a very cost competitive environment is key competitive factor in this industry. - Setup: we have worked out the following proof of concept with one our close customers. The company would like to be anonymous, since the server setup and usage is a competitive advantage in a very cost sensitive environment. We have populated one of the running systems with 30nm class memory and kept the comparable system on 40nm class. The systems are part of the pay-as-you-use concept. So the customer could increase/decrease the allocation of resources over the time and pay only that share that it use in a cloud environment. In this respect the average values of power consumption fluctuates about 20-30 W depending of usage.
- Results: the cloud servers B populated with 30nm class memory consume between 32-60 W less power depending on system loading.
(See below for details)
| systems | Cloud server with Green DDR3 40nm class | Cloud server with Green DDR3 30nm class |
|---|---|---|
| platform | Supermicro 2U 826TQ-R800LPB | Supermicro 2U 826TQ-R800LPB |
| Supermicro X8DTN+-F | Supermicro X8DTI-F (~same as X8DTN+-F; with less DIMM slots) |
|
| Raid | LSI MegaRaid SAS 9260-8i with Battery Backup Unit iBBU07 |
LSI MegaRaid SAS 9260-8i with Battery Backup Unit iBBU07 |
| CPU | 2x Intel Xeon L5630 | 2x Intel Xeon L5630 |
| Memory |
68GB Samsung REG 16x DDR3 ECC REG 1333 Mhz, 4 GB M393B5270CH0-CH9, 2Gb base 2Rx4 40nm 2x DDR3 ECC REG 1333 Mhz, 2 GB M393B5670FH0-CH9, 1Gb base 2Rx4 40nm |
72GB Samsung REG 9x DDR3 ECC REG 1333 Mhz, 8 GB M393B1K70DH0-YH9, 2Gb base, 30nm, 1.35V |
| HDD | 2x Seagate 3.5" 160 GB (ST3160318AS) | 2x Seagate 3.5" 160 GB (ST3160318AS) |
| 8x Seagate 3.5" 1000 GB (ST31000528AS) | 8x Seagate 3.5" 1000 GB (ST31000528AS) | |
| Idle mode* | 252 W | 220W (32W less than DDR3 40nm class) |
| Burn-in* | 340 W | 280W (60W less than DDR3 40nm class) |
*We tested the power consumption of both systems in idle state (system fully booted under Linux kernel 2.6.18) and while running a Burn-In with the CTCS tool
(see link below) for min. 60 sec.
CTCS TOOL :
http://sourceforge.net/projects/va-ctc
We have reached comprable results(<5% deviation) with Goodle stress test for Memory : http://code.google.com/p/stressapptest
Please note that these are average values as the power consumption fluctuates about ~1-7% since we have performed the test on Live systems over the period of observation.
OPEX & Capex consideration
| TCO | Cloud server with Green DDR3 30nm vs 40nm | Value (Idle) | Value (fully loaded) |
|---|---|---|---|
| Capex | module extra cost | 0 euro | 0 euro |
| Opex | Saving on system level | 32W | 60W |
| Saving in cooling | Not measured | Not measured | |
| Total saving in DC* (PUE=1,5) | 48 Watt | 90 Watt | |
| Cost of energy | 3,5* Euro /Watt/life cycle | 3,5 Euro /Watt/life cycle | |
| Life cycle | 4 ~ 5 Years | 4 ~ 5 Years | |
| Total cost saving per system per life cycle | 168 Euro / life cycle | 315 Euro / life cycle | |
| Total cost saving per 100 system per life cycle | 18.600 Euro / life cycle | 31.500 Euro / life cycle |
* Webhoster's Datacenter and cost values
For comments and feedback and involvement into PoC at your site please contact Peyman Blumstengel
(peyman.b@samsung.com)
BACK TO THE LIST
