QSAN - high quality at a reasonable price.
In connection with granting Upgreat the status Authorized QSAN Sub-distributor in Poland, we decided to share our impressions about the products offered by this manufacturer.
But let me start with maybe general information about mass storage.
They are present in almost every large enterprise, institution or used for private purposes. Most of the Internet resources are stored on mass memory. Hence, it is important that these devices are reliable and efficient. Our experience with such devices is based mainly on such vendors as HP, IBM, NetApp. Due to the award of the status of the Authorized QSAN Sub-distributor, we received two matrix models from this manufacturer. It is a Taiwanese manufacturer that started operating in 2004. Qsan Technology (this is the full name of the company) specializes in the production of a complete line of SAN, iSCSI, Fiber Channel SAN storage network products for global enterprises, medium and small businesses.
The person who took care of QSAN arrays is me, Marcin, an engineer at Upgreat responsible for implementations related to servers, arrays and other equipment related to the proper operation of IT systems.
At the very beginning, I would like to present the hardware configuration and the method of connecting the matrix to the server.
Q500 matrix
- 4 FC ports maximum speed 8GbE,
- 2 iSCSI ports with 1GbE speed each.
- 2 cooling systems,
- 2 power supplies,
- 24 slots for hard drives (it is possible to expand to 256 drives),
- possibility of connecting two controllers.
The P600Q matrix
- iSCSI hardware controller 2x 10GbE, 2 x 1GbE and RAID.
- expansion of the cache memory up to 16GB
- 2 power supplies,
- 2 cooling systems,
- 12 slots for hard drives,
- possibility of connecting two controllers.
Test plan:
Q500 matrix
- Connecting the matrix to the server - FC (4 GbE)
- Creation of RAID 5 (4 drives - 1TB each)
- IOmeter performance tests - various block sizes.
- HDTune performance tests.
- Failover tests.
The P600Q matrix
- Connecting the matrix to the server - iSCSI - 1Gbit / s.
- Creation of RAID 0 and RAID 5 (4 drives - 300GB each)
- IOmeter performance tests - various block sizes.
- HDTune performance tests.
- Failover tests.
Q500 test results
Performance tests were performed for the matrix with iometer and Hd Tune for various block sizes. A group was created in RAID 5 consisting of 4 1000GB disks (WD).
Hd Tune - Block 512B
Hd Tune - Block 64K
HD Tune - 512K block
IOmeter - block 512B
IOmeter - block 16K
IOmeter - block 32K
P600Q test results
Performance tests with iometer and Hd Tune were also performed for the P600Q array for various block sizes. A group was created in RAID 0 and RAID 5 consisting of 4 disks of 300GB (WD). The test results for different block sizes are presented below. The matrix was connected to the server via an iSCSI port with a speed of 1Gbit / s.
RAID 0 - HD Tune - 64K block
RAID 0 - HD Tune - 512K block
RAID 0 - IOmeter - 16K block
RAID 0 - IOmeter - 32K block
RAID 5 - HD Tune - 64K block
RAID 5 - HD Tune - 512K block
RAID 5 - IOmeter - 16K block
RAID 5 - IOmeter - 32K
Conclusions
After performing performance tests and failover tests, it can be concluded that QSAN arrays can compete with companies such as IBM or HP, because their equipment has similar parameters, but it is much cheaper.
During a RAID rebuild after a simulated disk failure, the performance of the disks was measured. The access time increased by 0.7 ms and the transmission speed decreased, however, it did not have too much influence on the operation of the test system. With more drives and more data, RAID recovery can take up to several days.
According to the manufacturer, thanks to the use of QiSOE (Qsan iSCSI offload engine), the iSCSI performance increases by 30% while reducing the CPU load by 25%.
Another fact worth mentioning is that the P600Q matrix obtained a certificate of compatibility with virtualization systems (Vmware, Hyper-V, Citrix). It is worth mentioning here that Vmware is supported on the VAAI level (matrix programming interface). It consists in the fact that the Vmwate server sends the command to the matrix, which copies the image indicated by the matrix, skipping copying to local Vmware disks.