| Message ID | 20200724080254.8389-1-rsalvaterra@gmail.com |
|---|---|
| State | Accepted |
| Delegated to: | Daniel Golle |
| Headers | show |
| Series | wireguard-tools: allow compiling with MIPS16 instructions | expand |
On Fri, Jul 24, 2020 at 10:03 AM Rui Salvaterra <rsalvaterra@gmail.com> wrote: > > The wg utility compiles and runs without issues in MIPS16 mode, despite setting > PKG_USE_MIPS16:=0 in the makefile. Let's remove this, allowing for a substantial > size reduction of the wg executable. Since wg is a just a configuration utility, > it shouldn't be performance-critical, as the crypto heavy-lifting is done on the > kernel side. > > wg sizes for both modes: > > MIPS32: 64309 bytes > MIPS16: 42501 bytes Seems probably fine to me, but would you let me know if the performance of `wg pubkey` suffers as a result? I'm far from home at the moment without any access to MIPS hardware to test myself. Jason
> On Jul 24, 2020, at 1:14 AM, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > On Fri, Jul 24, 2020 at 10:03 AM Rui Salvaterra <rsalvaterra@gmail.com> wrote: >> >> The wg utility compiles and runs without issues in MIPS16 mode, despite setting >> PKG_USE_MIPS16:=0 in the makefile. Let's remove this, allowing for a substantial >> size reduction of the wg executable. Since wg is a just a configuration utility, >> it shouldn't be performance-critical, as the crypto heavy-lifting is done on the >> kernel side. >> >> wg sizes for both modes: >> >> MIPS32: 64309 bytes >> MIPS16: 42501 bytes > > Seems probably fine to me, but would you let me know if the > performance of `wg pubkey` suffers as a result? Of course. I imagine it’s very tiny. Logging in to the router with ssh is much slower. > I'm far from home at > the moment without any access to MIPS hardware to test myself. > > Jason > > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Well… On Fri, 24 Jul 2020 at 09:32, Rosen Penev <rosenp@gmail.com> wrote: > > > On Jul 24, 2020, at 1:14 AM, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > > > Seems probably fine to me, but would you let me know if the > > performance of `wg pubkey` suffers as a result? > Of course. I imagine it’s very tiny. Logging in to the router with ssh is much slower. root@heimdal:/tmp# time wg genkey | tee privatekey | wg pubkey > publickey real 0m 0.00s user 0m 0.00s sys 0m 0.00s root@heimdal:/tmp# … yeah, I don't think performance will be an issue. :) (This is a 775 MHz 74Kc router, mind you.) Thanks, Rui
On Fri, Jul 24, 2020 at 10:39 AM Rui Salvaterra <rsalvaterra@gmail.com> wrote: > > Well… > > On Fri, 24 Jul 2020 at 09:32, Rosen Penev <rosenp@gmail.com> wrote: > > > > > On Jul 24, 2020, at 1:14 AM, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > > > > > Seems probably fine to me, but would you let me know if the > > > performance of `wg pubkey` suffers as a result? > > Of course. I imagine it’s very tiny. Logging in to the router with ssh is much slower. > > root@heimdal:/tmp# time wg genkey | tee privatekey | wg pubkey > publickey > real 0m 0.00s > user 0m 0.00s > sys 0m 0.00s > root@heimdal:/tmp# > > … yeah, I don't think performance will be an issue. :) > (This is a 775 MHz 74Kc router, mind you.) > > Thanks, > Rui Testing the process once like that isn't a good testing methodology representative of anything at all.
On Fri, 24 Jul 2020 at 09:53, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > Testing the process once like that isn't a good testing methodology > representative of anything at all. I completely agree, this wasn't an objective test at all. I was merely illustrating what a normal user will do, generate a key pair to configure a VPN. If it took more than two seconds, I'd be worried, but it "feels" instantaneous. I could, of course, time 1000 key generations in a loop, on a completely idle system, but that wouldn't represent the typical use case. :)
On Fri, Jul 24, 2020 at 11:05 AM Rui Salvaterra <rsalvaterra@gmail.com> wrote: > > On Fri, 24 Jul 2020 at 09:53, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > > > Testing the process once like that isn't a good testing methodology > > representative of anything at all. > > I completely agree, this wasn't an objective test at all. I was merely > illustrating what a normal user will do, generate a key pair to > configure a VPN. If it took more than two seconds, I'd be worried, but > it "feels" instantaneous. I could, of course, time 1000 key > generations in a loop, on a completely idle system, but that wouldn't > represent the typical use case. :) This might be a typical use case for you, but some people are running scripts that generate lots of keys. There may be a good argument that if you're doing that kind of thing, a tiny MIPS router isn't the right hardware for your use case. But, before going down that route, I'd still be interested in knowing if it _actually_ makes any difference at all. Maybe it's negligible, for example.
Hi again, Jason, On Fri, 24 Jul 2020 at 10:12, Jason A. Donenfeld <Jason@zx2c4.com> wrote: > > This might be a typical use case for you, but some people are running > scripts that generate lots of keys. There may be a good argument that > if you're doing that kind of thing, a tiny MIPS router isn't the right > hardware for your use case. But, before going down that route, I'd > still be interested in knowing if it _actually_ makes any difference > at all. Maybe it's negligible, for example. Here are more objective numbers. I timed 1000 runs of wg genkey | wg pubkey > /dev/null. Each timing was done four times for both MIPS16 and MIPS32 and the worst results were chosen. MIPS32 results: real 0m 10.06s user 0m 8.12s sys 0m 1.34s MIPS16 results: real 0m 11.96s user 0m 10.14s sys 0m 1.15s I'd say the difference is, indeed, negligible. Thanks, Rui
diff --git a/package/network/utils/wireguard-tools/Makefile b/package/network/utils/wireguard-tools/Makefile index 3232060bd1..2137a8b4a1 100644 --- a/package/network/utils/wireguard-tools/Makefile +++ b/package/network/utils/wireguard-tools/Makefile @@ -22,7 +22,6 @@ PKG_LICENSE:=GPL-2.0 PKG_LICENSE_FILES:=COPYING PKG_BUILD_PARALLEL:=1 -PKG_USE_MIPS16:=0 include $(INCLUDE_DIR)/package.mk include $(INCLUDE_DIR)/package-defaults.mk
The wg utility compiles and runs without issues in MIPS16 mode, despite setting PKG_USE_MIPS16:=0 in the makefile. Let's remove this, allowing for a substantial size reduction of the wg executable. Since wg is a just a configuration utility, it shouldn't be performance-critical, as the crypto heavy-lifting is done on the kernel side. wg sizes for both modes: MIPS32: 64309 bytes MIPS16: 42501 bytes Signed-off-by: Rui Salvaterra <rsalvaterra@gmail.com> --- package/network/utils/wireguard-tools/Makefile | 1 - 1 file changed, 1 deletion(-)