In my raw IO benchmark of Intel 520 SSD we saw that the drive does not provide uniform throughput and response time, but it is interesting how does it affect workload if it comes from MySQL.
I prepared benchmarks results for Sysbench OLTP workload with MySQL running on Intel 520.
You can download it there.
There I want to publish graphs to compare Intel 520 vs regular RAID10.
Throughput:
Response time:
So despite big variation in raw IO, it seems it does not affect MySQL workload significantly, and single Intel 520 SSD gives much better throughput and response time comparing with traditional SAS RAID, and what is interesting it also much cheaper.
What’s bad with Intel 520 is that this card does not have capacitor to protect write cache, so if you worry about data protection in case of power outage it is better to disable write cache on this card and use write cache from RAID controller (i.e. LSI-9260).
Benchmarks specification, hardware, scripts and raw results are available in the full report.
Is 520 significantly faster than 320?
320 has capacitors. But I wondered if it is slower than 520.
Also in your comparison what is the configuration of RAID10? 8 15k SAS with BBU write cache?
Andy,
Actually it appears that 520 is noticeable faster then 320.
You can get an idea comparing
https://www.percona.com/blog/2011/07/intel-320-ssd-read-performance/
https://www.percona.com/blog/2012/05/testing-intel-ssd-520/
RAID10 is 8 15k SAS with BBU, that is correct.
These graphs use what tool to produce?
hongbin,
http://www.mysqlperformanceblog.com/2012/02/23/some-fun-with-r-visualization/
Could you please also show specs (or something like lspci -vvv output) for the RAID controller? It performs too slow for 8 HDs.
Alexey,
Here it is:
RAID bus controller: Hewlett-Packard Company Smart Array G6 controllers (rev 01)
Subsystem: Hewlett-Packard Company Smart Array P410i
Was it the same setup you used for raw IO benchmark?
According to specs from official website, neither this server, nor this raid controller support SATA-III (6-gbit one). 380 mb/s throughput you’re hitting is fully saturated SATA-II. No wonder – according to published benchmarks this SSD drive should do 520-550 mb/s (without any problems since our block size is 16k or bigger).
Oh and by the way, the kernel release you’ve used was a buggy one (and it sucks in terms of performance too – both overall and mysql-specific).
I think newer LSI cards (2108-based) should support 6 gigabit for SATA-III drives (and they also have recent drivers backported to RHEL6 kernels).
Was the testing performed on the Intel 520 with the write cache disabled on the drive in favor of the cache on the controller?
Mark,
The write cache was enabled on the Intel 520, and the raid controller has no write cache.
The 520 has no write cache.
Look at pictures of the board, no DRAM.
http://en.wikipedia.org/wiki/SandForce
“SandForce controllers do not use DRAM for caching[2] which reduces cost and complexity compared to most other SSD controllers.”
You can also verify the above by measuring the performance with cache disabled. No real difference.
This does not mean that it cannot loose data on a powerloss. That would depend on it guaranteeing that writes are completed to nand before it returns ok to the controller which is a completely different thing, especially given the compression that the 520 implements.
I have no idea how “guaranteed” the writes are or how big the risk is for dataloss, but it is defintely much smaller than for some controllers with a lot of DRAM cache.
I found MySQL to run much slower on my Intel SSD 520 than i did on my WD 320GB SATA II. Shouldnt everything run much faster on an SSD?