Disk performance in laptops?

Hi,

I'm reviewing stats for some execution times (which imply data transfer rates) of some utilities I'd run previously on laptops and, from that, see what I can only attribute to lousy disk subsystem performance.

Granted, laptop drives tend to be less performant simply because of power and space considerations (and, the Windows disk access patterns). And, of course, laptops are all over the map in terms of price/performance.

But, what sort of unbuffered,[1] single sector performance are you likely to encounter in that environment? E.g., a desktop disk can easily saturate a 100Mb link -- or a USB2 PCI i/f. I'm not sure that is true with many (?) laptop drives...

I realize the 2.5" form factor (nor the PATA/SATA distinction) should arbitrarily limit performance (though low RPMs can). E.g., I have some 2.5" drives that will do 300MB/s easily (though run very warm and draw twice the power of "typical" drives)

Thx,

--don

[1] Of course, all disk accesses are buffered -- in several places. My point is "without taking explicit and extraordinary measures"...
Reply to
Don Y
Loading thread data ...

On Wed, 19 Nov 2014 14:02:53 -0700, Don Y Gave us:

10,000 and 15,0000 rpm SAS drives.

Better... Solid State... duh. mSATA and M.2 mini-PCIe are the screamers these days. Literally 100s of MB/s even over USB links. Some laptops have a socket in them.

Reply to
DecadentLinuxUserNumeroUno

The Solid State SATA drives I have installed in my Gigabyte Brix PC claim a transfer capability of 6Gb/s. Not sure I could fill the DDR3 memory that fast though.

--
******************************************************************** 
Paul E. Bennett IEng MIET..... 
 Click to see the full signature
Reply to
Paul E Bennett

On Thu, 20 Nov 2014 01:51:36 +0000, Paul E Bennett Gave us:

The tests are things like having multiple partitions on the same drive and then copying a big 20GB movie file from one to the other.

It gets done in chunks and the copy or move routine manages that, but it all has to go through the memory a couple times as the open file gets written to and built up with a running check code to keep from having to read the file again.

I notice small stuff... like when more than one file is selected, the entire copy runs at one low rate (on mine that was some 70MB/s) thru the whole thing.

When a single large file is selected, it "bursts" several hundred megabytes at about 140MB/s for just a second or so, and then drops to the low rate for the rest of the process.

Then, I found a neat trick and some insight as to what takes place with these really big files. I used "move" instead, and it "bursted" at a super high 1+GB/s rate and then dropped to like 180MB/s for the rest of the move. So I concluded that the "move" code was probably actually done well (better than the copy code) and could be relied on for good numbers. I need to get a second SSD in there and see what the physical volume to physical volume speeds are. They should be far higher, since the other way was all partitions residing on the same physical drive.

Oh... darn... that is my spinning drive I did all that with. Please take note.

The SSD in that box is my OS drive and I will not be taxing that like this. I am sure Windows hits them way too often as it is.

So, when I get two more SSDs to put in there, I can refer to those numbers. I do intend to do that as well.

Reply to
DecadentLinuxUserNumeroUno

I've done a lot of database related performance measurements during the last months. Typically a cheap 500GB 5400RPM 2.5" SATA disk can sustain about 30-40MB/s with large (1MB) files, which are typically quite quite well located. With very small files all bets are off, as the read head will need to do a lot of work. I've seen even numbers less than 1MB/s in such cases.

When measuring disk performance, remember to read much more data (or at least new data) than you've got RAM. Modern OSs cache aggressively. I've used bonnie++ for comparing performance on different setups.

I don't have any numbers to give, but with 7200RPM similar disk my laptop felt much faster.

With SSDs you'll get much higher numbers, of course. Especially small file performance improvement will be huge. My target has been 'rust drives', as 500TB of SSD still costs a lot.

--
mikko OH2HVJ
Reply to
Mikko OH2HVJ

Yeah, big files alter the results dramatically. Especially if they can exploit contiguous location on the medium.

I've been seeing numbers in the 2MB/s range -- but, that's at the end of a significant pipeline, as well (and, with just "sector accesses" -- though the OS will obviously be buffering).

I'm not trying to measure what the disk *can* do but, rather, understand the low numbers I was seeing. E.g., even a USB2 drive would typically be moving ~15M/s (1GB/min). I was processing ~80G of data (but in sector sized pieces) and seeing total execution times in *hours* instead of "an hour or so".

No doubt. And, figure most (of my) workstations are 10K and 15K drives so there's a significant difference in expectation. Likewise, the USB2 drives I was comparing against I had tended to use in big transfers so there wasn't all these syscalls and traps to eat up cycles.

Exactly. One of the questions I'm trying to answer is "where" it makes the most sense to access a drive's content:

  • on the laptop (with a "power optimized" CPU of varying ability)
*or*
  • ship it off to a server that can copy it to a faster medium (RAM/SSD)

By way of analogy with your DBMS apps: if you had many small(er) DB's and only needed to run one at a time, it might be faster to move the entire DB onto an SSD, run it there, then ship the updated DB back to its original medium (sort of as "tertiary storage").

That would allow a small(er) SSD to give you high performance at the cost of these two "copy's" (because the DB's content is accessed many times so the cost of the copy quickly falls into the noise when compared to the relative speedup of the SSD accesses)

Reply to
Don Y

As pointed out most consumer drives are ~40Mbs. I have a WD blue 160G PATA, I see ~45Mbs. I also have a WD Scorpio Black, I see ~100Mbs on a 3yr old acer i5. My PATA laptop is now using a PATA to MSATA converter and maxes out at

90Mbs with a Msata SSD.

I've come to the conclsion that drives with a good 4k cluster size transfer rate get high ratings, like the samsung SSD drives. THeir 4k transfer rates are higher than average, but not by much.

NTFS by default uses 4k Clusters.

Cheers

Reply to
Martin Riddle

"See" when doing *what*? I.e., I was looking for "unbuffered,[1] single sector performance". So, the cost of the disk *subsystem* is reflected, not just the disk itself. I want to get a feel for how much I have to do to get this out of the "critical path" and shift the performance issue to CPU-bound issues.

And, with "typical" laptops, not laptops tuned for specific performance, etc.

Reply to
Don Y

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.