Announcement press release was February 5, 2015. First ship was
sometime in March. This is the next generation follow-on to
the VDX 6740. A snippet from the
datasheet shows the info on packet buffers:
Brocade VDX 6940 Switches offer a unique balance between two conflicting
attributes—-buffer and latency. Niche products with very high buffer
often suffer from high latency, and products with ultra-low latency
are not a good fit for data centers with bursty traffic. Brocade VDX
6940 Switches, with a purpose-built data center chip, excel in
optimizing buffer and latency to deliver better application performance.
These switches deliver 700 ns any-port-to-any-port latency. In
addition, they offer an industry-leading 24 MB deep buffer. This
provides the buffering capacity to handle increases in traffic,
especially during peak times when ports are congested, allowing
traffic to be distributed across the ports. Brocade VDX 6940
Switches also feature a single ASIC design, instead of multiple
ASIC designs commonly found in other switches. This improves
performance since all ports communicate via the one ASIC.
A
cautionary note about vendor claims in general and this
switch in particular appears in Ivan Pepelnjak's blog.
The claim that big buffers cause high latency is [cough] bullshit.
A comparison
of the 6940 with switches pegged as the competition was offered
by Brocade. A nit-pick note: system buffer is a poorly
chosen name. System buffer is memory used by the switch CPU
for control plane functions. This is packet buffer data plane memory.
The Command Reference Manual page
for one of the buffer control commands shows the per port limit.
Optimized Packet Buffer
and Latency for Better
Application Performance
