Don't know specifically for SSD, but if it shows-up as a "normal" hd, this will be possible, since AFAIK the biggest "bottleneck" is the (unneccessary) expansion of qcaux and it's 278 MB worth of files for each and every task.

Note: For most projects except CEP2, there's no point of moving the slots-directory to a different hd, since appart for stderr.txt that often is empty, the slots-directory often includes only links to the actual files that is residing under the project-directory.

Under Windows, you'll need linkd.exe (or similar). If this is included or not AFAIK depends on windows-version you're using. If assumes BOINC data-directory is c:\programdata\boinc and you'll want to use d:\boinc as the 2nd. hd, it will be like this:

0: You must use NTFS on the hd's.
1: Exit BOINC-client.
2: To guard against problems, it can be an idea to backup the BOINC data-directory initially, in case you'll later-on runs into problems...
3: If neccessary, create d:\boinc
4: Navigate to c:\programdata\boinc, select the "Slots"-directory, and move this directory (including sub-directories) to d:\boinc
5: Open-up a cmd-window (just select start, run, and type cmd (enter)), and type following commands:

c:
cd \programdata\boinc
mkdir slots
linkd slots d:\boinc\slots
dir slots (This should now show the contents of the slots-directory residing under d:\boinc\slots, if it's not, something's wrong...)

Most common error, you'll get a message about linkd isn't a known program. If this is the case, search for linkd on the internet and download a version.

6: When successfully moved & linked slots-directory, uninstall BOINC & re-install BOINC.
7: Startup BOINC (while offline), and verify everything's working as it should...


If your current BOINC data-directory is located anywhere else, just replace as neccessary.

Hello everybody,
During 4 weeks in November and December 2010, I ran CEP2 only. The performance decreasing was more than significant - around 20% in comparison to HCMD2 or C4W - and the IO rate was really high.
Because I wished to test SSDs, I installed 3 weeks ago SSD for the boincdata directory. At the same time I took benefit of the new options for CEP2. I defined a limit of one WU per host for quad core with conventional HD and of two WUs for hosts using SSD (in this case: two Phenom II x6). Indeed the performance decreasing seems to be less significant with SSD. Nonetheless, even if SSD are used, Windows based hosts generate much more IO than Linux based hosts (Ubuntu 10.04 64 bits server).
The next step will be to limit CEP2 to Linux based hosts only, in order to treat the (expensive) SSDs with care. However, I have still the feeling that the CEP2 performance could (should) be improved to become more efficient and finally to complete more work per CPU hours.
My problem is not the earned credits - since I am crunching only for WCG and no real "competition" could be done against other projects, in particular with those using GPUs - but much more to consider earned credits as a representative performance indicator.
Briefly summarized, the question is not only to win more participants but to use the available computation performance more efficiently.
Cheers,
Yves
---
PS: @hypernova: for my-self, I did not experience many problems with errored WUs (excepted at the project start); are you maybe overclocking your hosts?

Ingleside, I think doing it your way would cause issues as sekerob pointed out. Boinc would need to be programed to allow this option to work. Similar to how video editing, ( input to one disk, output to another), kind of like a scratch disk. I think this would work great. I thought about using a ssd but presume it would die allot sooner than normal use. One thing I gave noticed is even crunching one wu along with other projects, it effects the other projects CPU time verses actual time. For now I will crunch other projects but will check here periodically to see if a better solution becomes available.

For tripple channel RAM to be used in tripple channel mode, either 3 modules or 6 modules need to be installed. Using 2 X 2GB means you are running in dual channel mode which has 33% less memory bandwidth. I'm just saying as having tripple channel improves the IO performance, of importance for CEP2.

While running just one CEP2 task initially caused other tasks on an 8core or 16core system to slow down due to the CPU time - actual time delta, it should be less now than it was when the project was first released (due to some I/O fixes). It's also likely to be less if you use tripple channel RAM, Linux and/or a fast hard drive system but dependent on what other projects you are running. HCC has been reported by some to be the best match for CEP2. I don't know how this would scale, for running multiple CEP2 tasks but I'm sure it would depend on the amount of RAM you have as well as the memory bandwidth and disk r/w speeds.

Have you noticed any improvements lately?