We`ve received alot of feedback on our 100G packet capture system and most customers don't believe we can do sustained 100Gbps capture to storage! We understand the skepticism as its taken alot of time and resources to achieve sustained peek write performance. We explain a little on our SSD journey.
SSD in a nutshell
You might think the biggest problem with SSD`s is the fixed number of PE(Program/Erase)cycles however our experience shows this is the least important problem. First and most importantly function for us is the sequential write performance across the entire drive. Unfortunately sustained SSD write performance is a very unique workload, and almost never seen in real world applications expect our 10G 40G 100G packet capture systems. Thus few SSD vendors optimize for it.
Most SSD`s are optimized for high IOPs (IO Operations per Second) for random(ish) read/writes address all over the drive. This covers the majority of workloads such as desktop usage, general server usage and massive databases. As such most SSD vendors optimize for high random IOPS which usually has a negative effect on sustained sequential write performance.
In addition to less emphasis on sequential write performance, most SSD`s operate in a "Cached Mode". The assumption is general workloads operate on a smallish subset of data with an SSD Cache architecture providing an excellent performance boost. However, for 100G packet capture once the working data set exceeds the total cache size the performance drops off a cliff. To give you an idea here`s some graphs of sequential write performance across an entire drive.
amd radeon r7 256G
toshiba q hd pro 256G
Plexor m5 Pro 256G
Samsung 850 pro 256G
Samsung 850 Evo 256G
Crucial mx100 256G
Crucial MX200 256
Corsair Neuton XTS SAFC 250G
Kingston HyperX Savage 240G
OCZ 100 256
OCZ Vector 256G
SanDisk Extreme II SSD 240G
SanDisk Ultra II Pro SSD 240G
Segate 600 240G
.... and the list goes on. The above tests are the most basic benchmarks to achieve sustained sequential write performance for our 100G packet capture systems. The performance gets more complicated once fragmentation and garbage collection kicks, unleashing a whole different set of problems and challenges. In short we know our SSD`s beyond the level required of data recovery specialists. After all our SSD storage backend is the coal burning furnace that drives at the very heart of every 10G 20G 40G 80G 100G packet capture system.
enterprise SSD i hear you say?
The sad truth of "enterprise" class SSD`s is they are expensive and slow. Their designed to be fast HDD replacements, not retail SSD`s for the enterprise. This means the total storage size is low, with very high PE cycle counts (compared to retail SSD`s) and performance significantly improved over a 2.5" 15K SAS HDD ... but not even close to the bleeding edge retail SSD market... and a x5 price tag.
Our capture systems raw cost is dominated by the storage cost. As such we use retail SSD`s which have significant price and performance advantages, then layer smart software on this of cost effective hardware. This approach is different to the traditional "Big Iron" IBM ... expensive approach of engineering hardware for extreme reliability with basic software. Compare IBM to leading tech companies such as Amazon, Facebook, Google .. all use cost effective hardware with smarts and redundancy built into the software. We agree using smart software on cost effective hardware is the superior approach and our customers agree too, our pricing is fantastic!
100Gbps Write Performance
Getting back to how to get 100Gbps of sustained write performance, we use many optimizations and performance methodologies to ensure we get not only maximum performance, but maximum sustained write performance across an entire array, always.
FMADIO100 12TB Storage Write performance
... and yes we`re doing around 160Gbps of raw sequential write bandwidth across the entire 12TB of our lowest end storage configuration. Its not just the low end config, the FMAD100 100G capture system performance is the same for 24TB and 48TB models. This level of bandwidth is some what insane but extremely cool!
Our customers and competitors have doubts about our ability to capture and write 100Gbps to storage. We`ve been there and understand why, its hard to get 100Gbps of sustained write performance. Whats required to achieve 100Gbps of sustained sequential write performance is decades of experience building high performance systems, deep understanding of system architecture and much blood, sweat and tears.
Contact us for an eval if you still don't believe us :)