06 December 2012 Source:
Over the past few weeks the team at Boston Labs have been evaluating the new MegaRAID 8100 series. Using the same technology and RAID processor as the existing MegaRAID 6Gbit/s range and the CacheCade Pro 2.0 SSD acceleration kit, the MegaRAID now includes both products together in one SKU and even includes integrated SSD's.
This amalgamated product brings users several benefits:
There is no need to place the SSD inside your chassis and lose potential hotplug bays
Users don't have to manually install the CacheCade software key and register online
SSD's are isolated from the disk backplane, they don't have to share an expander with the storage HDD's giving optimum performance
The performance can be tuned by Avago to give maximum IOPs and throughput
Support from Avago is a simpler process as they do not have to validate numerous SSD vendors drives
MegaRAID Product Range
This card is available in several different flavours, using both SLC and eMLC memory to give better performance or higher cache capacity respectively.
Nytro MegaRAID 8100-4i - 100GB SLC SSD Cache
Nytro MegaRAID 8110-4i - 200GB eMLC SSD Cache
Nytro MegaRAID 8120-4i - 800GB eMLC SSD Cache
All controllers are based on the LSI dual core RAID on chip 2208 processor, utilise PCI Express generation 3 for improved bandwidth to the host, have 4 ports and support RAID 0, 1, 5, 6, 10, 50, and 60 as standard.
In our lab, we tested the Nytro MegaRAID 8100-4i which includes 2 x 50GB SLC SSD's on-board. The SSD devices utilise additional ports directly from the controller, leaving all 4 ports available to connect your disk devices or expander backplane. This means it's likely that most implementations will be with an expander backplane to allow more than 4 disks to be used, giving the best price per GB metric.
The on-board SSD's can be configured as a 50GB RAID 1, or as a 100GB RAID 0 Nytro caching device. For applications where the Nytro cache is used for both read and write operations then RAID 1 is strongly recommended or data loss would occur in the event of SSD failure. In our testing we used a mirror configuration as this is likely the most common real world implementation.
The Nytro cache accelerates the hot spots of data activity that are common in a variety of different applications such as SQL, web, and exchange. As a result it is not suited to every application, however where it can apply the gains are huge. To simulate a hotspot of data we configured the well known IOMeter to utilise only the 1st 100 million sectors of a RAID 5 volume to determine how the Nytro Cache might accelerate access in a real world application. Four runs were completed, without Nytro Cache, then with an empty Nytro Cache and 2 subsequent runs allowing the Nytro Cache to fill with hot data.
Our test bed was configured as follows and our results are below -
Supermicro X9DR3-F Motherboard
Supermicro CSE-846A-R900B Chassis
64Gb DDR-3 1600MHZ ECC Registered Memory
LSI Nytro MegaRAID 8100-4i - 23.4.1-0052 FW - 188.8.131.52 Driver
4 x WD1000DHTZ 3.5" 1TB 10K S-ATA HDD - 6A00 FW - Configured as RAID 5
Windows 2008 R2 SP1 x64 Enterprise
RAID Settings - Write Through Cache, Always Read Ahead, Disk Cache Off, Direct IO, Nytro Cache On / Off
Testing & Results
The results of our tests are depicted in the graphs below -
As you can see the results are very impressive. For the small 4K block random access pattern test there is a clear improvement in performance as the cache warms up. By the third run the performance is more than 10 times that of an un-cached array in read, write, and mixed load performance - a clear win for Nytro MegaRAID.
The large block sequential throughput test does appear to show a drop in performance for sequential reads. In our previous testing with the older Cachecade Pro 2.0, this was more pronounced so there has been some clear progress here in the optimisation, allowing the controller to use data stored on the HDD in favour of SSD cache for purely sequential operations.
Write performance for large blocks does improve by as much as 50% and mixed operations are performing as much as 3-4 times better. With most workloads being a mixture or varying block sizes, reads and writes with both random and sequential patterns - Nytro Cache makes excellent sense and we can thoroughly recommend it.
Using special tools, it's now possible to simulate the acceleration on a running system and truly understand the potential benefits. If you're interested in Nytro Cache but are unsure if it could really accelerate your workload then please drop us a line, we'll gladly help you decide.
For further information please contact us.