Significant performance drop after úpdating to 490
Hello,
I noticed a significant performance drop after updating to 490 from version 487. Please compare these pictures, first from version 487:
Memory card SanDisk Extreme 4GB speed 30x (20 MB/s)
http://img178.imageshack.us/my.php?image=eboostr304871st6.jpg
Memory card SanDisk Extreme 8GB speed 30x (20 MB/s)
http://img178.imageshack.us/img178/5989/eboostr304872vc6.jpg
OK, that was excellent performance from eBoostr. Almost maximum speed my memory cards can achieve.
*****************************************
After updating to 490 results show the following:
Memory card SanDisk Extreme 4GB speed 30x (20 MB/s)
http://img178.imageshack.us/img178/2254/eboostr304901ju6.jpg
Memory card SanDisk Extreme 8GB speed 30x (20 MB/s)
http://img178.imageshack.us/img178/5594/eboostr304902cg0.jpg
*****************************************
What you say? Speed has dropped so much I can’t even believe it. Maybe I have to downgrade to 487. I love this program but I do not know what can cause this.’
Best regards
Crisse
Jan 25, 2009
Hello again,
I formatted my memory cards, regenerated cache and got the same speed as before, maybe even better:
Memory card SanDisk Extreme 4GB speed 30x (20 MB/s)
Random read speed 19342 Kb/s
Memory card SanDisk Extreme 8GB speed 30x (20 MB/s)
Random read speed 19982 Kb/s
Earlier upgrades have not changed random read speeds, but if you encounter same kind of problems, just format your flash drives (if they do not have something you need). Regenerating the cache (without re-format) did not help, at least in my case.
Jan 25, 2009
Personally, I don’t see how you would ever see ANY performance gain with only 20 MB/s access times. Spindle drives typically get 40 minimum on average. You should seriously consider researching a new USB flash drive which gets at least that amount! Have you ran HDTACH on any or all of your I/O devices for comparisons?
Jan 26, 2009
The main ‘improvements’ achieved using eBoostr are *not* because the USB-sticks have a high data-throughput, but because they have very low latency!
Simply put : your hard-disk *can* feed data at very high speeds *if* that data is nicely ordered on the disk. However, if the data is ‘all over the place’, the disk-heads need to switch from track to track, which is rather time-consuming but cannot be avoided because of the mechanical limitations of hard-disks. USB-sticks have the advantage that they have no moving parts, it’s just a bunch of electronics and as such offer much lower seek-times. As a result of this difference in hardware, reading large files from a (defragmented) hard-disk will be faster indeed, but reading lots of small files will be much quicker when getting it from the USB stick. If eBoostr does it’s job well, it will try to fetch data from where it can be accessed the fastest, thus offering best of both worlds.
I suggest to read the wikipedia article about ReadyBoost. It’s not 100% identical to eBoostr, but generally spoken the same ideas apply.
Jan 26, 2009
My note above was not to say “don’t use USB” it was to say “get a faster stick.” Are you going to say this wouldn’t improve performance?
Next, see my later post with speed testing screenshots… which do RANDOM READ ACCESS measurements… and tell me a faster medium isn’t beneficial?
I measured little performance gains on my system when using eBoostr on a USB flash drive. That is why I switched the cache to the i-RAM disk.
I’d sure like to see AnandTech do similar testing on eBoostr… because the eBoostr authors don’t give us any comprehendible statistics from within the application itself.
Jan 26, 2009
P.S. Read the actual article about ReadyBoost testing. It simply proves adding RAM, and/or using a RAM drive for caching, is way better than using a USB flash drive for caching.
So why limit oneself more with a *slow* USB flash drive? There are plenty of models which read at least 40 MB/sec.
http://www.anandtech.com/systems/showdoc.aspx?i=2917&p=6
Jan 26, 2009
Ah, sorry, one more thing. Let’s do some math!
USB thumb drive seek times are typically 1 ms — but that’s PER data block (let’s say that’s 4096 bytes). That would mean 1048576 bytes (1 MB) requires 256ms to read, MINIMUM.
Spindle drives, on the other hand, average 8 ms per SEEK (not per data block) and they have build-in cache. A fragmented file of 6 MB can be read in <40ms.
We can fairly assume the file would not be fragmented as much as we can assume the USB flash drive has the file you need cached already.
So, even though the USB flash drive benefits with very very small data blocks, how many of these very small files are saved in your eBoostr cache?
So, it all comes down to faster USB flash drives… a 40 MB/s thumb drive would have half the seek time, about .5ms.
Even so, I don’t really see the benefit of using USB flash as the booster unless the disk drive you have is very VERY slow, constantly beaten down by pagefile access, and is never ever defragmented!
TTFN
Jan 26, 2009
I fully agree that the faster the USB stick, the more effect… however, I read your reply as : don’t bother using that USB stick, it’s too slow.
There’s at least 2 reasons why (even a slow) USB stick will help :
* much lower seek times help out reading small files. In (simplified) numbers : assume a disk takes 10ms to get to the right position and another 3ms to read 128k (= 40Mb/s), that means we needed 13ms to get the data. The USB has a seek time of 1ms and needs 6ms to feed the data (= 20Mb/s), or a total of 7ms => you’ve just won 6ms or almost 50% !
* having 2 places where the data can come from is akin to having your data in a RAID-1 environment, and there’s plenty of stats around that will show you this is beneficial.
Jan 27, 2009
deroby,
You are correct that even a slower USB device might beat out in very small data reads, but the problem is this device is not JUST caching small files.. it’s caching larger ones which have reduced performace, slower than the system drive! So, what you might gain in small files is likely lost (or most of it) in the larger ones. Your ‘50%’ example above is agreeable, but it’s not the full picture.
The best way to take advantage of the slower flash media is to somehow configure eBoostr to use the flash media for smaller files and RAM for larger ones. At this time, however, devices cannot be configured separately from each other (to my knowledge) — the same exclusion list applies to all devices.
So, my recommendation is still to get faster flash media, or use RAM instead, and not bother with the slower ones — but I will caveat that with “unless you configure it to cache only small files.”
I think someone mentioned how to exclude larger files in another post.