Dennis Wood
Senior Member
Tim, what you've found is pretty much what we've seen based on a few weeks of testing various configurations.
It goes something like this:
1. PCI NIC limit: PCI connected NICs limited the machines regardless of OS. The best read/write rates we measured here on our 5.3GB file set were in the 40MB/s area.
2. SMB vs SMB2: Going from XP to Vista SP1 was required before we started pushing single drive limits, either to/from the NAS units, or between workstations. In Vista, drive indexing from the search feature and a host of other features affected results unless the services were turned off, or the machine left to itself for an hour or two. With XP running, the best gigabit rates we saw (between two RAID0 workstations) was an average of 58MB/s. This jumped to 93MB/s with Vista SP1 upgrade installed. We did not test pre SP1 Vista otherwise, but on 3 different workstations, performance improved a great deal over XP (unlike your results). All three have C2D processors at 2.3 to 2.7 Ghz and 2GB of RAM.
3. Gigabit max transfer rates: Once that was taken care of, we were able to hit our max of 103MB/s average transfer rates (again over a 5.3 GB file set read/write) but that was between one of the Vista SP1 workstations and the TS509 in RAID0 with 5 drives. Iozone reported over 120MB/s up until the NAS ran out of cache.
The one thing that I would suggest is running the iozone tests using the parameter 2G instead of reducing the workstation RAM. This gives a much more accurate view of the NAS performance (particularly the TS509) which in my opinion is misrepresented somewhat by the manufacturer's tests of a 1GB file. Yes, the single drive and raid0 figures are great, but using RAID5 (which most users will want I suspect) and assuming files over 1GB, the writes are positively slow. The Intel box and the Qnap box look very similar testing with iozone at 1GB, but very different as you go to 2GB. This is reflected in our measured tests with the Intel SS4200 box sustaining large file writes at nearly 3 times faster than the TS509...and that's using Vista. One thing that helps to see the cache effect is Vista's copy dialogue that indicates the current transfer rate. You can copy a 2GB file over to the TS509 and see it blast along until about 750MB (assuming again a RAID array on the test workstation to deliver a 120MB/s or better data rate) and then watch the calculated rate drop as the NAS's "real" performance is reached.
One thing I haven't posted up yet is the load testing with multiple clients, link aggregation, teamed dual gigabit and all that sort of fun stuff. The results were interesting. Btw, this: http://kb.ciprico.com/lore/article.php?id=268 is worth a go to test your hard disk read and write rates...it works well under Vista and is free..unlike the HDTach version that tests writes.
Looking forward to the balance of this series Tim
It goes something like this:
1. PCI NIC limit: PCI connected NICs limited the machines regardless of OS. The best read/write rates we measured here on our 5.3GB file set were in the 40MB/s area.
2. SMB vs SMB2: Going from XP to Vista SP1 was required before we started pushing single drive limits, either to/from the NAS units, or between workstations. In Vista, drive indexing from the search feature and a host of other features affected results unless the services were turned off, or the machine left to itself for an hour or two. With XP running, the best gigabit rates we saw (between two RAID0 workstations) was an average of 58MB/s. This jumped to 93MB/s with Vista SP1 upgrade installed. We did not test pre SP1 Vista otherwise, but on 3 different workstations, performance improved a great deal over XP (unlike your results). All three have C2D processors at 2.3 to 2.7 Ghz and 2GB of RAM.
3. Gigabit max transfer rates: Once that was taken care of, we were able to hit our max of 103MB/s average transfer rates (again over a 5.3 GB file set read/write) but that was between one of the Vista SP1 workstations and the TS509 in RAID0 with 5 drives. Iozone reported over 120MB/s up until the NAS ran out of cache.
The one thing that I would suggest is running the iozone tests using the parameter 2G instead of reducing the workstation RAM. This gives a much more accurate view of the NAS performance (particularly the TS509) which in my opinion is misrepresented somewhat by the manufacturer's tests of a 1GB file. Yes, the single drive and raid0 figures are great, but using RAID5 (which most users will want I suspect) and assuming files over 1GB, the writes are positively slow. The Intel box and the Qnap box look very similar testing with iozone at 1GB, but very different as you go to 2GB. This is reflected in our measured tests with the Intel SS4200 box sustaining large file writes at nearly 3 times faster than the TS509...and that's using Vista. One thing that helps to see the cache effect is Vista's copy dialogue that indicates the current transfer rate. You can copy a 2GB file over to the TS509 and see it blast along until about 750MB (assuming again a RAID array on the test workstation to deliver a 120MB/s or better data rate) and then watch the calculated rate drop as the NAS's "real" performance is reached.
One thing I haven't posted up yet is the load testing with multiple clients, link aggregation, teamed dual gigabit and all that sort of fun stuff. The results were interesting. Btw, this: http://kb.ciprico.com/lore/article.php?id=268 is worth a go to test your hard disk read and write rates...it works well under Vista and is free..unlike the HDTach version that tests writes.
Looking forward to the balance of this series Tim
Last edited:
