x86 and x86_64 woes
Posted: Tue Apr 26, 2016 9:10 am
I have been working on setting up an x86 based router as I want qos that can handle more than 50 mbit up/down. The commercial routers have at most arm 1.xx ghz cpu's which aren't even comparable to a cheapo low end x86 cpu.
But as I have found, building is a hassle- especially when you want to do x86_64 (to get proper SMP support). Despite the weirdness in the gargoyle build environment - as it uses openwrt build env as a base- I finally got the packages I need to work out.
However, I hit another frustration which occurs whether I use my x86_64 or 32 bit builds or other premade gargoyle x86 builds:
If traffic is going on up, down, or both- changing anything in the qos down or up page will crash the router. There's no message because it becomes either a hard freeze or reboots automatically (depending on kernel options). With little or no traffic, there is no crash and qos updates fine.
Even maxing out the qos at 300/300mbit (yes, wow and on a dual core 1.2 ghz core 2!) the cpu load is below 50% total!
VM's were notoriously slow, whether using vmxnet, virtio, or e1000e drivers. I have tried all the NIC tweaks on the host and guest os.
I have the same issue with pfsense- and I think it's because VMs add latency to the network stack which throws off the queuing capability of both systems.
Any idea what could be causing this or how I could get an actual trace that could help me find out? I suppose I could try to sacrifice a 120Gb sata drive to load an OS that needs less than 100mb to eliminate the usb flash issue (I have used different ports and flash drives).
But as I have found, building is a hassle- especially when you want to do x86_64 (to get proper SMP support). Despite the weirdness in the gargoyle build environment - as it uses openwrt build env as a base- I finally got the packages I need to work out.
However, I hit another frustration which occurs whether I use my x86_64 or 32 bit builds or other premade gargoyle x86 builds:
If traffic is going on up, down, or both- changing anything in the qos down or up page will crash the router. There's no message because it becomes either a hard freeze or reboots automatically (depending on kernel options). With little or no traffic, there is no crash and qos updates fine.
Even maxing out the qos at 300/300mbit (yes, wow and on a dual core 1.2 ghz core 2!) the cpu load is below 50% total!
VM's were notoriously slow, whether using vmxnet, virtio, or e1000e drivers. I have tried all the NIC tweaks on the host and guest os.
I have the same issue with pfsense- and I think it's because VMs add latency to the network stack which throws off the queuing capability of both systems.
Any idea what could be causing this or how I could get an actual trace that could help me find out? I suppose I could try to sacrifice a 120Gb sata drive to load an OS that needs less than 100mb to eliminate the usb flash issue (I have used different ports and flash drives).