Articles: Networking
 

Bookmark and Share

(1) 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 ]

Performance

The performance of the NAS was tested with Intel NASPT 1.7.0 that replays prerecorded traces of various usage scenarios. We installed four Western Digital Caviar Black WD5001AALS HDDs (500GB, 7200rpm, 32MB cache, SATA II interface) into the TS-439 Pro and used a PC as a client. The PC had a dual-core Intel Core 2 Duo 1.8GHz processor, 4GB of system memory, a PCIe 1x Gigabit Ethernet controller, and 32-bit Windows Vista. The NAS was connected to the local network via a Gigabit Ethernet router with support for Jumbo Frames. This technology was enabled on all the devices used in the test session. The HDDs were formatted in one of the possible configurations, we created a user account and assigned it full rights to one of the common folders. We did not change anything else in the factory settings.

The first diagram compares the single-disk, single disk with encryption and iSCSI volume usages:

We’ve got interesting results here. The speed of the single-disk volume is higher than what a four-disk RAID0 built out of early HDDs would deliver. It is very good for a single disk to have a write speed of 38MBps and a read speed of 55MBps. This is even faster than USB 2.0 which has been among the fastest external interfaces just a little while ago. We can also note the familiar dependence of performance on the trace. The speed is lower with the small files of the Content Creation and Photo Album traces than with the large HD video files. Encryption slows the NAS down by 30-70%, especially with writes.

To test the iSCSI protocol we created a single virtual volume on a single-disk array and used Windows Vista’s integrated client software. To remind you, iSCSI is a kind of an extension of the disk controller’s software interface via Gigabit Ethernet and does not allow to access this disk from multiple systems simultaneously. However, the speed seems to be limited by the network adapter here: almost 80MBps for writing and reading large files. This is very good even for full-featured storage systems.

Such a big difference from classic NAS is due to the operation mechanism. Most of software network stacks are bypassed by an iSCSI connection because data transfers are performed at the low level of disk blocks. This feature helps reduce the overhead for sequential operations. On the other hand, when there is a large number of small files, a classic NAS is faster because the server-side cache memory is utilized effectively.

The second group of tests is meant to show us the highest speed the NAS can provide by means of RAID0 arrays. Although this RAID type is not fault tolerant, it can improve performance if the latter is not limited by the platform.

The overall behavior is quite standard: the top read speed is growing up along with the number of disks in the array. It is only in the sequential writing test (HDVideo_1Record) that we see an odd thing: the top performance is achieved on the two-disk array. Perhaps the CPU is not fast enough to process such a large amount of data for more than two HDDs.

The last diagram is about the fault-tolerant configurations: two-disk RAID1, three- and four-disk RAID5, and four-disk RAID6. The first two options can survive a failure of one HDD whereas the RAID6 can survive two HDDs failing at once.

The two-disk mirror is somewhat faster than the single disk in reading: the system is reading from both disks alternately in this case. Writing is expectedly worse since data is doubled, requiring some overhead. The three-disk RAID5 is faster than the RAID1 and is preferable due to the more efficient utilization of the disk space. The four-disk RAID5 performs excellently at reading and nearly hits the 80MBps barrier we have seen in the RAID0 tests. Writing requires checksum computations, so the array is only as fast as the single disk then. On the other hand, 54MBps is a very good speed, too. The RAID6 leaves a nice impression. When using four disks, it is similar to RIAD1 in useful capacity but faster by up to 40% depending on the test trace.

 
Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 ]

Discussion

Comments currently: 1
Discussion started: 12/10/09 01:08:24 PM
Latest comment: 12/10/09 01:08:24 PM

View comments

Add your Comment