LAN speed confusion

woodson:

What gives? Why does the LAN utility give such "inflated" results?

Presuming that you have a 1 gigabit/s Ethernet connection to the NAS, the read speed of 203 Mibytes/s is impossible - that is 1624 Mibits/s, way faster than your Ethernet can do. So clearly that number is just wrong. However, LAN speed tests will often be much faster than the file copy speeds, as the LAN speed test does not use the hard disk, so is not slowed down to the speed of the disk transfers. It is only fairly recently that hard disks have become faster than gigabit Ethernet, and your NAS box is not likely to have such fast disks. And lots of NAS boxes seem to have rather slower disk speeds than the rated speeds of the drives, as they have underperforming CPUs or bad software. And the protocol used for the file copies also matters, and whether it has been tuned properly. Some protocols have higher LAN overheads than others. The protocol used for LAN speed testing is likely to be minimal, compared to that used for file transfers.

And then there is the packet size used - in a LAN, if you want to speed things up significantly it is possible to use much larger packets such as 9000 bytes or even larger, if your Ethernet cards support it and if their drivers support it and if the operating systems support it. That cuts down on the protocol overheads. But mixing LANs with large packets enabled with access to the wider Internet where only the original 1500 byte maximum packet size is supported causes all sorts of problems (especially with Windows boxes), so it is rare to be able to get that working.

Also a problem when doing file copies to test speeds is caching. Most modern operating systems use any RAM that is not in use for other things as file cache. So if you copy the same file more than once, the second copy operation will likely find lots of that file, if not all of it, available in RAM cache, and will not have to read the disk. At the other end, when writing a file, once all the file is received at the far end and is in RAM cache, the signal may well be sent back to the receiving end that the copying is done, when in fact it will take quite a while longer for the RAM cache to be written to disk. To complicate things further, hard disks also have their own cache.

Benchmarking is a very difficult thing to get right, even for experts. There are many more complicating factors.

woodson

fe31nz