Teaser-time!
- Thread starter RMerlin
- Start date
shrike4242
Occasional Visitor
And what a tease you are showing it off to us. Looks good.
applesnowleo
Regular Contributor
367.28! Uau, when???
RMerlin
Asuswrt-Merlin dev
What can you expect out of an ARM A9 dual-core CPU?
OpenVPN throughput benchmark:
iperf -c 10.16.0.1 -M 1400 -N -l 64K -t 30
The iperf and OpenVPN servers were on the router, and clients were a Win7 laptop connected to the WAN port.
For OpenVPN, this router is a beast
Here's the raw OpenSSL performance. openssl speed test:
That's right. OpenSSL raw performance is nearly 3x faster on the RT-AC56U.
OpenVPN throughput benchmark:
iperf -c 10.16.0.1 -M 1400 -N -l 64K -t 30
The iperf and OpenVPN servers were on the router, and clients were a Win7 laptop connected to the WAN port.
Code:
RT-AC66U (FW 3.0.0.4.270.25):
AES-128-CBC [152] 0.0-30.0 sec 79.5 MBytes 22.2 Mbits/sec
RT-AC56U
AES_128-CBC [156] 0.0-30.0 sec 217 MBytes 60.7 Mbits/secFor OpenVPN, this router is a beast
Here's the raw OpenSSL performance. openssl speed test:
Code:
RT-AC66U:
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-128 cbc 9398.56k 9941.17k 10138.71k 10195.09k 10234.54k
RT-AC56U (with ARM ASM optimizations backported from OpenSSL 1.0.1):
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
aes-128 cbc 25596.63k 28151.82k 29251.64k 29363.88k 29587.09kThat's right. OpenSSL raw performance is nearly 3x faster on the RT-AC56U.
RMerlin
Asuswrt-Merlin dev
The FW version isn't relevant. It's the other things on that page that are
This is developed on a separate (private) code branch because it contains unreleased code that I can't mix with the public Asuswrt-Merlin code. This is why this is still based on .28 and not .29 for instance.
Once there's a public release of the source code, the new GPL code will be merged in the public branch, and all the model-specific code will be re-applied.
RMerlin
Asuswrt-Merlin dev
I haven't had much time to do too many performance tests so far as I have been focusing on implementing HW support in Asuswrt-Merlin. The OpenVPN test was in part to ensure that my OpenSSL and OpenVPN builds were working correctly on the ARM architecture.
I'll try to find some time in the near future to do more performance tests. Keep in mind however that this is all based on pre-release firmware code and drivers, so actual results can vary once the product is released with its final firmware. Also, I don't have Tim's experience at benchmarking routers. But I'll try to still get at least a basic throughput test done with and without CTF enabled and see how it compares to an RT-AC66U.
Wireless is something I can't really test at the moment however, as the router is currently disassembled outside of its case to allow me to connect it to my laptop while debugging code on it. A PC board laying flat on a living room table ain't the best way to test radio coverage
The future has arrived to domestic routers, raw throughput at it's best.
Asus got you some very nice presents before official launch, one up for Asus for them supporting you and your firmware project.
Can't wait for even more goodies.
Do you have any idea on price and/or availability?
Asus got you some very nice presents before official launch, one up for Asus for them supporting you and your firmware project.
Can't wait for even more goodies.
Do you have any idea on price and/or availability?
RMerlin
Asuswrt-Merlin dev
I expect Asus to announce more details next week at Computex. The latest figure I saw posted on the web was late June or early July (I think that was on guru3d's announcement?)
As for pricing, nothing announced yet. I saw a placeholder entry on NCIX, but the price there could be pure speculation on their part, and totally change by the time they actually get stocks.
How much do you want? Will probably be in the 500+ Mbps region if it is like the other Broadcom BCM4708 powered routers I've just reviewed (Linksys EA6300, D-Link DIR-868L, NETGEAR R6250).
RMerlin
Asuswrt-Merlin dev
I did some basic WAN-LAN tests using iperf. Unfortunately I'm not setup to do a proper test - my main desktop is in a different room and behind a cheap switch (which saw a major throughput drop when I tested from it), so I had to settle for the laptop and the Atom-based HTPC.
With CTF enabled, the HTPC CPU was the bottleneck while running iperf, so the number was kinda meaningless (it was a bit above 600 Mbits).
With CTF disabled, I was getting 240 Mbits, which seems to be directly scaled from the 600 MHz clock rate of the RT-AC66. Unfortunately the dual-core doesn't help much there.
Again, this is with a pretty simple setup here, and with pre-release firmware, so these numbers are only rough ballparks - they might change between now and release time.
With CTF enabled, the HTPC CPU was the bottleneck while running iperf, so the number was kinda meaningless (it was a bit above 600 Mbits).
With CTF disabled, I was getting 240 Mbits, which seems to be directly scaled from the 600 MHz clock rate of the RT-AC66. Unfortunately the dual-core doesn't help much there.
Again, this is with a pretty simple setup here, and with pre-release firmware, so these numbers are only rough ballparks - they might change between now and release time.
RMerlin
Asuswrt-Merlin dev
Kernel is 2.6.36, so it's newer. No idea what Asus has backported from the 3.x versions - keep in mind that 2.6.36 is just the base version. Asus tends to backport patches and security fixes from newer versions. The 2.6.22.19 used by the RT-N66U for example has the USB code from 2.6.32 or 2.6.33 (I forgot the actual kernel version), and various Netfilter bits also from newer 2.6.xx versions.
Thank you for reply. At least we get ext4 with "new" kernel.
What's the CPU load with that? If it seems linear to CPU speed, that seems to suggest only one core being used.
Well.... being able to handle symmetric 1gig wouldn't hurt and provide some headroom for speeds that are just around the corner. I'm waiting for the day where these boxes can work on a pro level, so activating features doesn't slow things down to a crawl. 240 might not be enough for the next speed bump.
RMerlin
Asuswrt-Merlin dev
One core was indeed topped. The current bottleneck lies at the irq level, so I doubt that SMP can help with that. Someone told me that this is an inherent problem with gigabit Ethernet, which generate a lot of interrupts to process. The closed-source ctf module must be doing something about it beyond just bypassing netfilter steps.
Might be worth digging through more recent kernels to see if there are any optimizations done at Linux's level to help with these.
On the other hand, having SMP does mean that while you have one core heavily loaded with traffic, the other core can take care of other things like Samba or Transmission. I believe that dual-core will be a great benefit to modern routers who are now designed to do much more than just route packets.
