@jcl333,
first - I wish you a Happy New Year!
Sorry for the delay in responding, had been away and busy.
To address your questions:
1) I realized that this community has a strong affinity with LSI but I do not have any preset preferences with regards to Raid controllers. Most of the controllers I used in the past were fine for the job at hand, but the speed and performance of multiple SSDs put a different level of stress on all parts of the chain (hardware and software alike).
For the high performance workstation LSI proved to be the better solution:
1) When I started the project, it was the only PCI 3.0 x8 capable controller - a prerequiste in my case, as 8 fast SSDs would saturate a PCI 2.0 x8 interconnect
2) The LSI 9207-8i controllers were reasonably priced
3) The LSI 2308 RoC is a good performer
4) The drivers for WS2012 are rock solid, even at 2.2 mio IOPS or more than 20 GB/sec seq transfers. (I did not know that upfront, but in hindsight)
The reason to go for the Adaptec was easy. It seem to be a better fit for the power/port ration I was trying to optimize. Most of the energy saving 1155 boards don't have multiple PCI x8 slots and multiple LSI controllers would consume much mor energy than the rest of the system. So the goal was to use only one controller with as many ports as possible. As I am probably amont the first batch of customers using this new controller, I can't claim that I based my decision on previous and other experiences. I hoped and expected that the new controller is a good fit for the purpose. and for a reasonable price.
The system is now running 6 weeks and the controller did not show any issues during operation with my setup.
When I tested the controller in the high perf workstation before I built it into the server, I connected it with 24 SSDs - just out of curiosity. Pushed to the limits, the controller did not exhibit the same stability than the LSI controller. It is hard to say from the superficial check I did, but my assumption is that the driver hasn't undergon the same level of regression testing than the LSI drivers which are much longer in the market. As an example: When pushed to max IOPS in some cases the driver disconnected the controller from the OS and the only remedy was to reboot the system. This test was way beyond the usage envelope of my intended use, and all experiences so far with 24 hard disks is very positive. It does, what it is meant to do. in my case, I configured it with 2 Raid5 volumes defined in the controller to reduce the load on the CPU (keeping the lower power envelope when transferring data): This decision is based on the observation that the power consumption of the intel CPU fluctuates stronger than the controller's power between idle and fully loaded.
I did not have any airflow in the case where the RoC of the adaptec is located. Consequently, the temp of the RoC was reported by the Raid utilities to be in the 85 degree celcius range. While it seemed hot to me, the utilitiy reported the temp to be "ok". Due to my concern on the long term probability to get issues with this temp, I installed a tiny 4cm vent on a creative 3 bracket structure to lead the airflow over the chip and connected the vent to the motherboard. Energy consumption = not measurable. Temp dropped by 40 degrees and is now at 45 degrees which I like much more.
I haven't done any play with jbod, software raid or Raid6 configurations but the performance with Raid5 is more than sufficient to saturate my installed network bandwidth (4 GBit/sec).
hope that helps & regards,
Andy
first - I wish you a Happy New Year!
Sorry for the delay in responding, had been away and busy.
To address your questions:
1) I realized that this community has a strong affinity with LSI but I do not have any preset preferences with regards to Raid controllers. Most of the controllers I used in the past were fine for the job at hand, but the speed and performance of multiple SSDs put a different level of stress on all parts of the chain (hardware and software alike).
For the high performance workstation LSI proved to be the better solution:
1) When I started the project, it was the only PCI 3.0 x8 capable controller - a prerequiste in my case, as 8 fast SSDs would saturate a PCI 2.0 x8 interconnect
2) The LSI 9207-8i controllers were reasonably priced
3) The LSI 2308 RoC is a good performer
4) The drivers for WS2012 are rock solid, even at 2.2 mio IOPS or more than 20 GB/sec seq transfers. (I did not know that upfront, but in hindsight)
The reason to go for the Adaptec was easy. It seem to be a better fit for the power/port ration I was trying to optimize. Most of the energy saving 1155 boards don't have multiple PCI x8 slots and multiple LSI controllers would consume much mor energy than the rest of the system. So the goal was to use only one controller with as many ports as possible. As I am probably amont the first batch of customers using this new controller, I can't claim that I based my decision on previous and other experiences. I hoped and expected that the new controller is a good fit for the purpose. and for a reasonable price.
The system is now running 6 weeks and the controller did not show any issues during operation with my setup.
When I tested the controller in the high perf workstation before I built it into the server, I connected it with 24 SSDs - just out of curiosity. Pushed to the limits, the controller did not exhibit the same stability than the LSI controller. It is hard to say from the superficial check I did, but my assumption is that the driver hasn't undergon the same level of regression testing than the LSI drivers which are much longer in the market. As an example: When pushed to max IOPS in some cases the driver disconnected the controller from the OS and the only remedy was to reboot the system. This test was way beyond the usage envelope of my intended use, and all experiences so far with 24 hard disks is very positive. It does, what it is meant to do. in my case, I configured it with 2 Raid5 volumes defined in the controller to reduce the load on the CPU (keeping the lower power envelope when transferring data): This decision is based on the observation that the power consumption of the intel CPU fluctuates stronger than the controller's power between idle and fully loaded.
I did not have any airflow in the case where the RoC of the adaptec is located. Consequently, the temp of the RoC was reported by the Raid utilities to be in the 85 degree celcius range. While it seemed hot to me, the utilitiy reported the temp to be "ok". Due to my concern on the long term probability to get issues with this temp, I installed a tiny 4cm vent on a creative 3 bracket structure to lead the airflow over the chip and connected the vent to the motherboard. Energy consumption = not measurable. Temp dropped by 40 degrees and is now at 45 degrees which I like much more.
I haven't done any play with jbod, software raid or Raid6 configurations but the performance with Raid5 is more than sufficient to saturate my installed network bandwidth (4 GBit/sec).
hope that helps & regards,
Andy
