Single spots of slow harddisk performance, but no bad sectors?

Question

I use a Samsung 1.5TB harddisk (EcoGreen F2) and recently tidied up quite a lot. So 500GB is written, the rest is free. Now when I am copying some huge files (gigabyte sized), the harddisk shows full speed up to 100MB/s, but there are single spots where the speed goes down severely to 2–5MB/s.

The data can and will be successfully written on the drive, I checked it. But it’s quite slow with these spots. When I wrote the file, there can be a range of 10, 20GB fullspeed area, then again slow speed. The thing is, there are no bad sectors (according to SMART data), and since the data is actually written correctly, the harddrive is not likely to mark the areas as bad sectors.

Is there a possibility to gain more overall speed or to make the drive just mark the spots as errornous, so that the next copy processes are faster? Currently I’m running an error check using HD Tune on Windows 7. But I’m not confident that it will change anything.

Answer 1

There's several factors at play here... Without more information on the drive mechanics, status, and what is already on the disk, I can only guess at some things.

First, when copying large files to the disk, it's unlikely that you'll get a singular chunk of 500gb (contiguous). It's more likely that you'll end up getting a few fragmented segments (if not many) of space all over the disk. So, as the file is being written... it's having to work back & forth over different regions of the disk.

Second, Most drives today have multiple platters and heads that have to share an axis in order to access parts of the disk. I.e., if your file is being written to the top side of the 1st platter on the first few blocks... and then the next block is allocated on the 2nd platter near the middle of the disk... you can end up where the disk's head has to jump back & forth between both segments (which adds a lot more time "seeking" between sectors) and that problem is compounded by technologies like "bus-mastering" that attempt to be smart and fill up all available bus bandwidth... by attempting to read/write in a "multi-threaded" fashion. i.e. a write operation is broken down into 5 write operations simultaneously... 1st writes blocks 1-10, 2nd writes 11-20... etc. Unfortunately, platter-based hard drives have to spend time going between each sector trying to write & verify the data... and so such a technology can actually cause more problems than it would fix. SSDs, on the other hand, have almost no seek-time... so they will fair significantly better.

Third, this is a "green" drive... which tries to be power-efficient. By design, they will "go-to-sleep" from time to time to reduce power-consumption, and spin right back up when something accesses the disk. There have been issues where the drive decides to goes to sleep (for a random unknown reason) and then has to wake back up. The spin-down time and spin-up time can generate a HUGE delay in read/write jobs. It's possible that there is a firmware for your hard drive to fix this issue... or perhaps you need to contact the manufacturer and get them to address the issue. (which may be near impossible for an 'individual/consumer' to do)

Forth, yes, the platter can go bad... but not enough bad to mark a sector as being bad. Most drives have some fuzzy-logic built-in that will write to the disk... go back & read that it was written to verify it was written properly... and if not... write it again... and then verify again... and if that fails too many times, it will finally mark the sector as being bad. But if it succeeds on the 2nd try... it's not marked as "bad" just yet.

I hope this helps you a bit... but I admit, this is far from complete. There are many MANY more reasons that can cause platter-based disk to become a "bottleneck" for performance.

Answer 2

When you remove a lot of loose items all over the hard disk, it leaves non-sequetial holes all over. Sequential writes and sequential accesses are many times faster than random on a hard drive. The MFT reserve might be being used and unused, and the unsequential mess that certian programs and even the system writes can make it a mess also.

Doing a full space consolidation defrag minimum, or going way overboard and doing a whole reorder defrag, will leave large open spaces. Depending on those programs (again), they will just find a hole and fill it , making another non-sequential mess. Best to defrag overnight some time. Cool thing about a full defrag, if you need to pull off a recovery of the data, it being sequential can come in really handy. so the pain has purpose :-)

If it is bad blocks and seek errors, then a S.M.A.R.T testing should show things occuring.
Getting as low as 2-5MB/s refer to The Comp Wizs answer also.
For everything else there is the "useless" defrag, which as anyone who defrags knows, there is nothing useless about it, when your going for max speed. It is sequential, it helps read-aheads, it can improve recovery ability, which just leaves it being a very time consuming process, best done when your not even there.

Way out: Really far out defragging can also include refreshing/clearing the journaling and resource messes using fsutil.