Enhancements and Improvements.

[Haitch]

So what data would you like to see in the Database, and why ?

Currently I pick up physical CPU's, cores, threads, Windows memory, CPU cache sizes and numbers, WU ID and min/max/avg TPF. Working on the DRAM speed, but thats going to take a little while. Also working on porting it to Linux so we can record for both OS's. Linux version will be Ubuntu specific - but should be compatible with other versions. If there is a need for a specific distribution to be supported, I can do that.

One thing I was looking at adding was a free form notation field for the database. Create a text file with you comments about the system config, and it gets added to the database for each run.

Thoughts? Comments? Requests?

H.

 

[Haitch]

First obvious enhancement - Get the OS. Getting the NUMA status - not sure, but might be possible on Linux.

H.

 

[musky]

The more, the better.
CPU - # sockets, # physical cores, # logical cores (or HT enabled), clock speed, architecture
Memory - amount, # dimms, speed, channel (dual, triple, quad), NUMA status, interleaving status
OS (Windows) - pretty much just knowing it is Windows is fine - there is no folding difference between XP/Vista/7 - maybe 32 bit or 64 bit
OS (Linux) - scheduler in use (BFS/CFS), kernel version, distro
Other - tear's "The Kraken" wrapper

I don't know how much of this you can grab, however.

 

[Haitch]

The more, the better.
CPU - # sockets, # physical cores, # logical cores (or HT enabled), clock speed, architecture
Memory - amount,

This I've got already,

OS (Windows) - pretty much just knowing it is Windows is fine - there is no folding difference between XP/Vista/7 - maybe 32 bit or 64 bit
OS (Linux) - scheduler in use (BFS/CFS), kernel version, distro
Other - tear's "The Kraken" wrapper

This I can get.

# dimms, speed, channel (dual, triple, quad), NUMA status, interleaving status
I don't know how much of this you can grab, however.

This I'm working on, but it could be a while coming.

H.

 

[musky]

Short-term, just let them be user-entered until you figure it out.

 

[Patrick]

Or... do not capture # of DIMM data and use N/A for the time being.

One other thought is to capture benchmark version, FAH client version, WU.

My suggestion on the time reported, we should do like 5 frames with min/max reviewed and average the remaining three.

Another thought is to not have user edited values at all. This allows us to skip user management.

 

[Haitch]

Or... do not capture # of DIMM data and use N/A for the time being.

Can do it either way - preference ?

One other thought is to capture benchmark version, FAH client version, WU.

I'll add those, along with number of frames processed.

My suggestion on the time reported, we should do like 5 frames with min/max reviewed and average the remaining three.

That's how I'm doing it - report the min and max, then average the remaining values.

Another thought is to not have user edited values at all. This allows us to skip user management.

For the # DIMMs, timings etc that currently need to be manually entered, I can prompt for those in the program. That way you only need DB users with insert and read access.

H.

 

[jebo_4jc]

so, you can or can not capture the number of DIMMs? With that data alone we could guesstimate pretty accurately about the number of channels.

Having a user defined "memory" field would be fine, but as you can see from the benchmark spreadsheet, the results are less than consistent.

Just to double check, you are capturing CPU clock speed? You didn't mention that in your first post.

 

[Haitch]

so, you can or can not capture the number of DIMMs? With that data alone we could guesstimate pretty accurately about the number of channels.

Currently can't get any DIMM info.

Having a user defined "memory" field would be fine, but as you can see from the benchmark spreadsheet, the results are less than consistent.

Trying to avoid user entered data completely

Just to double check, you are capturing CPU clock speed? You didn't mention that in your first post.

CPU type, clockspeed, caches .....

H.

 

[Haitch]

Currently can't get any DIMM info.

Now I can - number of DIMMs, capacity each and speed. Rather than add fields for each DIMM - those 4P's could have 16 DIMMs, record number of modules, and the slowest speed ?

H.

 

[Haitch]

This is the current output from the Windows Bench - now working on writing it to the database.

H.

CPU Vendor : Intel
CPU Name : Intel Core i7 860
CPU Model : Lynnfield
CPU Spec : Intel(R) Core(TM) i7 CPU 860 @ 2.80GHz
CPU Frequency : 3610 MHz
CPU Multiplier : 19.0
CPU FSB : 190.0
CPUID : 6.E.5
eCpUID : 6.1E
Package : Socket 1156 LGA (0x1)
Stepping : B1
Instruction Sets : MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, EM64T, VT-x

Num Processors : 1
Num Cores : 4
Num Threads : 8

MotherBoard Vendor: Gigabyte Technology Co., Ltd.
MotherBoard Model : P55A-UD4P

Num L1 I Cache(s) : 4
Size L1 I Cache(s): 32 KB
Num L1 D Cache(s) : 4
Size L1 D Cache(s): 32 KB
Num L2 Cache(s) : 4
Size L2 Cache(s) : 256 KB
Num L3 Cache(s) : 1
Size L3 Cache(s) : 8192 KB

Number of Dimms : 4
Mem Frequency : 570.0 MHz (2:6)
Total Memory : 8192 MBytes
Memory Type : DDR3
Memory Channels : Dual
CAS Latency : 9.0
RAS 2 CAS Delay : 9
RAS Precharge Time: 9
Cycle Time : 24
Row Refresh Time : 64
Command Rate : 2T

Operating System : Microsoft Windows 7 (6.1) Ultimate Edition (Build 7600)
TPF Frame 1 - 2549 seconds
TPF Frame 2 - 2399 seconds
TPF Frame 3 - 2397 seconds
TPF Frame 4 - 2374 seconds
TPF Frame 5 - 2449 seconds
TPF Frame 6 - 2542 seconds
TPF Frame 7 - 2449 seconds
TPF Frame 8 - 2370 seconds
TPF Frame 9 - 53339 seconds
TPF Frame 10 - 5899 seconds

FAHBench Ver : 1.03
FAH Client Ver : 6.34
Work Unit : 6900

Fah Arguments : -smp
Minimum TPF : 2370
Maximum TPF : 53339
Average TPF : 2882

 

[Patrick]

Pretty cool!

One perhaps dumb thing is to add a date.

Also, that is an odd bench on the 9th and 10th frames.

 

[Haitch]

The date/time of the run is in the database. The 9th and 10th frames were manually adjusted so I could test that it correctly calculated the TPF when the frame passed midnight - ie Frame 8 finished at 11:xx PM and frame 9 at 00:xx am.

H.