The Arachne Minimalist- Vol 2 Num 1 Clarence Verge ( ) Přečteno 157230x, 12 komentářů, rubrika: (en) Archive Disk caching and Arachne. What works best ?
Some caching software compared...
The last article of Vol#1 barely scratched the surface when it touched on
disk caches. There are not only many products to choose from, there are
also many ways to use the cache.
Some users recommend delayed writes to make the cache function as a short
term ramdisk. Let me assure you that HD caching with delayed writes or
otherwise, doesn't come close to the speed performance of a ramdisk.
Tests that seem to indicate otherwise are flawed in that they don't test
for the kind of disk activity required by Arachne. (See below)
In addition, the risk associated with delayed writes is far greater than
the risk associated with a ramdisk. The ramdisk user PLANS for failure
because he/she KNOWS the contents are volatile. The worst that can happen
with a lock-up or power outage while using the ramdisk is the loss of a
download or some mail.
When using a HD cache with delayed writes, your worst nightmare can come
true. If your application crashes, locking both you and the cache software
out, you are likely to get the most dreaded of all damage - crosslinked
files with different copies of the fats PLUS missing data.
Surfing the web with ANY browser means you are continuously dealing with
both newly acquired files and small files. Arachne additionally uses many
small files in its internal operation. A ramdisk will always be superior
to a large cache because the first access of a file on the ramdisk is as
fast as the second, while with a HD cache the first access of a file
requires that the (possibly long) cache index be searched first before
making a slow access to the HD.
In spite of the above, a good HD cache is not only necessary, it is vital
for fast and smooth operation of Arachne and any other software. Files
must be transferred to and from the ramdisk and safety dictates that mail
and download directories should be kept on the hard disk.
Keeping in mind the above warning that delayed writes might be dangerous,
I selected a number of caching programs for speed testing. The test
program, SPEEDCHK.bat,
was specifically designed to mimic the kind of activity Arachne generates
on the hard disk. To isolate the tests from the effects of DOS buffering
and possible disk fragmentation, I used a totally empty 46meg uncompressed
partition (2k clusters) and ran the tests with DOS's buffers set at ONE.
The tests were done with both MSDOS 6.22 and MSDOS 5.0 on a P90.
Note that some of the software tested will not support FAT32.
The following cache software was compared for speed and memory footprint:
PC-Cache.com V5.1
Smartdrv.sys V3.13
Smartdrv.exe V4.0
Smartdrv.exe V5.01
Ncache.exe V6.01
Ncache2.exe V8.0
NWcache.exe V1.02
Lightnin.com V1.21
The short story is that the fastest caches were Lightning and Ncache2, and
the smallest footprints were Ncache2 and PC-Cache. This seems to make Ncache2
the best overall; but Smartdrv wasn't terrible, so don't feel too bad.
The detailed test results take more space that I have here, but I have
included a link to the complete data.
To appease those living on the edge and using delayed writes, I've also
included the best delayed write results and a ramdisk test for laughs.
You can find all the results and detail in CACHSTUF.txt
Have fun. 8-)
Formátovat pro tisk bez komentářůFormátovat pro tisk s komentářiVšechny články tohoto autora Všechny články v rubrice (en) Archive |