QNAP TS-809 Pro Reviewed

[beq]

Another excellent review! Very high performance, but I'm intrigued by the way different brands and models leapfrog each other in different scenarios and applications (like iSCSI for example).

A minor point, the TS-509 Pro is supposed to have 1 eSATA port, which is one more than the 809, but not as many as QNAP's Atom models which have 2 ports. The article seems to imply that the 509 lacks any eSATA.

I wonder though, could QNAP have used a different set of SATA controllers with a higher port count, to be able to have eSATA?

[thiggins]

OOPS! Thanks beq for pointing out my error on eSATA on the 509 Pro. It does indeed have an eSATA port. I need to look at my own backup charts! D'oh.

Article has been corrected.

[Mark Adams]

I liked the article but I wanted to point out a possible issue in testing performance using link aggregation. This issue comes up when testing link aggregation on network storage devices with multiple data streams from multiple sources to a single set of disks. When using this test method, performance measurements are heavily affected by the storage device's hard drive seek times.

I found this out while I was testing high performance network storage systems that were expected to operate in the 300+ MB/S range. The performance I got using multiple computers was abysmal at 100 to 125 MB/S (abysmal in the sense I was looking for 300+ MB/S). When I realized what was occurring, I changed to a single computer with dual dual core CPUs (to eliminate the effect of processor throughput) with an Intel four port NIC using link aggregation through a switch to the four GBE inputs on the storage device. In this way I could run a single stream of data to the storage device without incurring seek times. This change did yield the expected Read/Write data rates of 300+ MB/S .

BTW I purchased both an 809 (7TB/R5) and 639 (5TB/R5) to fill my local storage requirements. I use the 809 for shared main storage w/incremental nightly backups to the 639. The 639 password is known to the 809 but not the systems that use the 809. In this way my 639 backup has a bit of protection from viruses and the like as the other systems on the net do not have the access password for it.

[the taker]

Does this class of device (integrated SAN/NAS with iSCSI target) support the SCSI-3 Persistent Reservations for Hyper-V Clustering?

[thiggins]

Thanks for the information, Mark.

If I am understanding you correctly, link aggregation is only going to produce increased performance with reads/writes to disk for a single stream of data, assuming that the devices on both ends of the read/write chain can handle the additional bandwidth.

So what sort of throughput gain have you seen for the TS-639 and 809 running link aggregation with your dual, dual test machine?

[thiggins]

Originally Posted by the taker

Does this class of device (integrated SAN/NAS with iSCSI target) support the SCSI-3 Persistent Reservations for Hyper-V Clustering?

I have no idea, sorry.

[Mark Adams]

Originally Posted by thiggins

Thanks for the information, Mark.

If I am understanding you correctly, link aggregation is only going to produce increased performance with reads/writes to disk for a single stream of data, assuming that the devices on both ends of the read/write chain can handle the additional bandwidth.

Yes, exactly . For DnD testing of the earlier referenced system I had 4x73G Raptors in Raid 0 using an 8G file as the data source . DnD performance topped out at slightly over 250 MB/S (Raptor performance was the primary limiting factor). The 300+ MB/S performance used IOMeter as the data source with an 8 Gig test area (to get around the XP cache issue). In this case the device driver CPU utilization was the limiting factor.

Originally Posted by thiggins

So what sort of throughput gain have you seen for the TS-639 and 809 running link aggregation with your dual, dual test machine?

The testing I did with link aggregation was long ago and far away (2006/7) at a company whose creativity and architectural expertise exceeded its financial skills . Currently my working system at home is a Q6600 with single 150G Raptors for my C: and D: drives. While it's good enough for my everyday use, it's woefully inadequate for aggregate testing .

The aggregation test results are not surprising, given that link aggregation needs tuning in each direction. Setting it up for load balancing on the NAS device does nothing to tune the distribution policy for packets outbound from the switch to the nas ports.

Most packet distribution policies implemented on switches are simplistic (XOR of source and/or destination mac or IP). It's entirely possible in a test of only two servers to the same NAS device, that the switch is deciding to use the same outbound link for both streams.

Some implement round robin, but thats usually a recipe for out of order packet delivery.

[beq]

I wonder why QNAP's new SS-839 Pro 8-bay model for 2.5" HDDs still manage to have 2 eSATA ports, but the 809 8-bay models (tower and rackmount) for 3.5" HDDs don't have any eSATA ports

[thiggins]

Different main board designs. Since it is Atom-based, I suspect it uses the same main board as the TS-439 and 639 Pros. Don't know why the 809 Pro main board didn't include eSATA. Not a smart choice, IMO.