Home Forums Discussions General Discussion Benchmark vs Real World, HUH!

My old computer had an AMD Fx Quad Processor one Samsung 830 SSD AC V8.2.4 took 8 seconds to open a chart, my new server has a Xeon V3 Quad Processor Raid 10 Samsung 850 Pro SSDs AC V8.2.4 takes about 6 seconds to open a chart it does seem a little faster overall except for prescribing.

Ran benchmarks read/writes on old system twice as fast as new; what the ???? (see below)

Could it be the benchmark testing is only a function of drive performance and Raid 10 is a hardware bottleneck for SSD's yet real world difference is due to Xeon processor being 40% faster per single core than AMD Fx???

Makes me wonder about just using a single SSD one for OS one for AC and be more anal about nightly backups.

Bump
I don't have an answer?
Anyone else?

Server design can be tricky. Trust me, I know...

With the single drive testing, you are actually seeing the Samsung Drive buffer the data. A single Samsung SSD transfers at about 500 MB/s sequential. Buffering is tricking the software since it is reading and writing the same file in its on-board memory. It's common issue for these benchmarking utilities. People's hard drives will look like they are working at 1GB/s but in reality it's much slower. If you want to see what the real speed is, use something like CrystalDiskMark and choose a file size that is like 5-10 GB. That way there is no way that amount of data can be cached. You'll see the real speed. There's also an option to disable the buffer (Samsung Magician) to get accurate results when benchmarking.

There is no software buffer for this benchmark with the RAID 10, but you can see the write penalty in your RAID data. You are getting 1GB/s, but the writes are half. As expected for RAID 10. 2 X Read Speed, 1 X Write Speed (relative to a single drive).

But what really matters is the 4K Random Read (IOPS in another form). Your system is usually reading really small files not large ones. It's why a 5 drive RAID 0 of hard drives is still very slow compared to a single SSD. They will both transfer about 500 MB/s but that's not what matters. It's also why without a good RAID card for RAID 10, it will seem just as fast as single SSD for day to day activities. You should re-run the tests and compare the 4K benchmarks. If they are the same, there's not much benefit over a single SSD other than reliability and better sequential transfer rates.

Actually, a LOT more anal.

When that hard drive crashes and that hot spare kicks in instantaneously, and the hard drive which is sitting next to the server can be popped in, that's awfully helpful. Still need a good backup, but that just saved you a ton of work and worry.

Once again, Sandeep, reveals knowledge that Bill Gates wishes he has. But, this just shows that you buy a $10,000 Xeon processor, and if the 8.2.4 cache isn't working, it doesn't matter how fast you pull the data over.

I'm getting ready to upgrade to 8.2.4 (ala ICD-10). Does this mean I will have 6 second waits for every chart I open??

Hi Boondoc,

Yeah, we are on about the same schedule....Still 9 days to relish 6.3.3 and ICD-9, can't rush these things.....

AC 8.2.4 is slower. AC has available 8.2.5, other board members have reported that this version is faster. The link is available from tech support, I emailed them and they sent the links.

I have not actually tried this version yet.

Good luck!

Gene

If you are lucky, YMMV!

We have been told in every CAB meeting, that 6.2.5 fixes about three issues, but has NO increase in performance.

Ditto, AC Support advised me no change 8.2.4 to 8.2.5