Sign in with
Sign up | Sign in

Iometer Results, Cont.

Seagate vs. WD: Battle for the SOHO NAS Crown
By
Brought to you by What's this

Webserver Profile

Whereas our workstation profile only used 8K files, the webserver profile is far more complex, featuring a much broader range of data appropriate to online site traffic. Naturally, 100% of this traffic is comprised of random reads, and most of the load is 8K or smaller. The smaller the random accesses, the more time it takes the hard drive to service that data.

Clearly, the more difficult workload takes a toll on overall performance. Our workstation profile used 20% sequential traffic as opposed to 0% sequential here, and this no doubt also plays a part in our difference in throughput levels between the two tests.

Fileserver Profile

Whereas 83% of our Webserver profile used reads of 8KB size or less, here we have another entirely random workload in which 80% of files are 4K or less, which is reflective of a high-transaction application, such as in the financial sector. Also, 20% of our purely random traffic is now comprised of writes, reflecting that a file server is likely to have some new files being added as users fetch files for outside use.

Interestingly, Seagate dips a bit as queuing increases while WD sees a slight rise. Neither move is terribly significant, again reflecting a write-cache enabled state.

Database Profile

With an 8K-only traffic size and 1/3 of that entirely random traffic being writes, our database profile finally lets queuing start to have an impact.

Both drives show marked gains in moving from a depth of 1 to 2 followed by a gradual rise. Interestingly, Seagate shows some weakness here with this data load when no queuing is in play. Once queuing kicks in, Seagate leaps sharply over WD, although both drives then gain only slightly as queuing increases.

2MB Streaming Reads

Now we start to see a workload more typical of a SOHO environment: streaming reads of 2 MB files. This might correlate well to a media repository filled with lots of JPEG or MPEG files.

Again, performance jumps when the system begins piling on read instructions into a standing queue. Both drives exhibit very similar performance in our RAID 5 environment, starting at 40 MB/s and scaling up to around 100 MB/s with queuing. Note that we examine streaming tests in MB/s terms while our enterprise-oriented tests report in IOPS. These are just two sides of the same coin.

2 MB Streaming Writes

Now the inverse of our prior test. What if you have users pouring loads of 2 MB media files into a NAS? This test should give us some idea.

We ran this test twice to confirm our data was correct, but yes, the lines really do overlap that closely. Effectively, there is no difference in performance between drives with this workload.