Revisiting SG-5100 ipsec in the real world
-
These are the settings I have on both boxes (an SG-5100 and an i3-4130t-powered box):
AES128-GCM (128 bits)
SHA256
14 (2048 bit)FWIW, the i3 seems to use even less of the cpu than the C3558-powered SG-100.
Both boxes have hardware acceleration enabled. And this is over a local gbps LAN with barely measurable latency.
-
I will run top -aSH as soon as I am back in that location, later on today.
-
If you're using AES-GCM you don't need an authentication hash since it's an AEAD cipher so that's unnecessary cpu cycles.
Enabling asynchronous crypto will usually accelerate things significantly. It cannot be enabled on all systems but the SG-5100 should be OK. That's an advanced ipsec setting.Steve
-
I just checked and asynchronous crypto was enabled.
-
How do I remove the authentication hash? the UI seems to force at least SHA1.
-
At phase 2, which is where it matters, it should not.
-
Problem completely solved. With crypto settings at suggested, ipsec throughput is 900 mbps on gbps LAN.
-
Nice!
-
I spoke too fast.
ipsec throughput is 900 mbps flowing from the SG-5100 (C3558) to the i3-4130t box, but maxes out at only 100 mbps in the other direction. What could cause such asymmetry? could it be asymmetry in encryption vs decryption? on which side?
At 900 mbps (SG-5100 to i3 box), the C3558 is at 75-80%, the i3 at 25%. in other direction, at 100 mbps, the C3558 is at 20%, the i3 at 9%.
any thoughts?
-
Generally I expect the encryption side to be more cpu intensive.
On some systems enabling async-crypto can actually limit throughput severely which is why we don't enable it by default in CE. You might try disabling that in the 4130T.Though those cpu usage ratios look quite similar in both situations, like something else is limiting it to 100Mbps. I assume you have tested directly, outside the tunnel, and you can see >900Mbps?
Steve
-
Yes, both boxes were able to handle gbps outside the tunnel.
I was finally able to get 900 mbps in both ipsec directions. Once the i3's hardware encryption was enabled, there was no problem. Interestingly, the C3558 maxed out at 95% on the receiving end, while the i3 flowed along at about 30% encrypting on the upload side. It's just a much more capable chip, even though it draws just 35W and runs only 2 cores (4 threads).
-
Hmm, just enabling AES-NI on the i3 end? Interesting, that's a significant step up, more than I would expect there.
Steve
