Recent Host Bus Adapter (HBA) testing by Demartek labs comparing Emulex and QLogic 16GFC HBAs yielded some very surprising results. While we already knew the Emulex LPe16000B delivered exceptional performance – with more than 1.2 million I/O operations per second (IOPS) on a single port1 – and is shipped by all major OEMs today, it was the QLogic performance that had us all scratching our heads. Demartek found that the newly released QLogic QLE2672 performance fell significantly short of its advertising claims—in some cases demonstrating worse performance for its 16GFC adapter than its 8GFC predecessor. So we tested them again, and again to be sure, and we could not come anywhere close to QLogic’s claimed 16GFC performance. While testing both adapters, Demartek found the following results. Here are a few of the report highlights:
Emulex LPe16000B was by far the fastest HBA evaluated, enabling 1.2 million IOPS on a single-port
- Emulex delivered 7x better IOPS than QLogic, which delivered under 200k IOPS on one port and requires both ports to reach its maximum IOPS 323,000. The Emulex architecture enables all resources to be applied to a single-port, enabling 1.2 IOPS on a single-port when needed.2
The LPe16002B runs at nearly full line rate for SQL Server/ Oracle workloads (4k and 8k block sizes) so Service Level Agreements (SLAs) are met.
- The LPe16002B is up to 124 percent faster for Oracle (4K block sizes) and up to 137 percent faster for SQL Server (8K block sizes) environments. The QLE2672 can’t achieve full line rate until it reaches 16k data block sizes. This is a significant deficiency if you happen to be running Oracle and SQL Server workloads. 2
CPU efficiency testing evaluated the amount of I/O being performed by the HBA in relation to the server CPU utilization being consumed. The Emulex HBA is more efficient by a factor of 13x for reads.
- The QLE2672’s I/O consumes more than 60 percent of the server CPU in processing I/O. The Emulex LPe16002B consumes less than 30 percent. It is most desirable to consume the fewest CPU resources for I/O processing as possible so that the server CPU is free to perform other processing, such as applications processing. 2
The Emulex LPe16000B is up to 11x more efficient in terms of IOPS/server watt.
- QLogic consumed 16,527 watts vs. Emulex at 1,487 watts. Additionally, Emulex allows the server to consume 14 percent fewer active watts (311 vs. 356 watts) while performing more than seven times the I/O traffic. Over a three year server life, the server using Emulex HBA saves 1 Megawatt of power compared to QLogic HBA. 2
The Emulex 16GFC HBA has the highest mean time between failures (MTBF), delivering enterprise class reliability.
- Emulex LPe16000B has a published MTBF of more than 10 million hours whereas QLogic has not published reliability data for the QLE2672.
To see the entire report, click here .
1. Demartek Emulex LPe16000B 16Gb Fibre Channel HBA Evaluation, October 2012
2. Demartek, Emulex LPe16000B 16 Gb Fibre Channel HBA Feature Comparison, December 2012