| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <1609010645.20121221122100@eikelenboom.it>
Date: Fri, 21 Dec 2012 12:21:00 +0100
From: Sander Eikelenboom <linux@...elenboom.it>
To: Eric Dumazet <erdnetdev@...il.com>
CC: Ian Campbell <Ian.Campbell@...rix.com>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>,
annie li <annie.li@...cle.com>,
"xen-devel@...ts.xensource.com" <xen-devel@...ts.xensource.com>
Subject: Re: [PATCH] xen/netfront: improve truesize tracking
Thursday, December 20, 2012, 4:39:28 PM, you wrote:
> On Thu, 2012-12-20 at 13:51 +0100, Sander Eikelenboom wrote:
>> Eric:
>> From the warn_on_once, delta should be smaller than len, but probably they should be as close together as possible.
>> When you say "accurate estimation", what would be a acceptable difference between DELTA and LEN ?
> I would use the most exact value, which is :
> skb->truesize += nr_frags * PAGE_SIZE;
> Then, if we can spot later a regression in some stacks, adapt the
> limiting parameters. I did a lot of work in GRO and TCP stack to reduce
> the memory, and further changes are possible.
> We really want to account memory, because we want to control how memory
> is used on our machines and don't let some users use more than the
> amount that was allowed to them.
Hi Eric and Ian,
I have ran some perfnet tests (although i'm not an expert, so i'm not sure i have done the right tests).
I you have better tests, please do say so ..
"current" is with netfront as is (skb->truesize += skb->data_len - RX_COPY_THRESHOLD;)
"patched" is with IanC's latest patch (skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags;)
Tested between domU and dom0 (bridged) on a system with only one guest. The results don't seem to differ very much.
+ netperf -H 192.168.1.1 -t TCP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. KBytes/sec
current 87380 16384 16384 60.00 954438.38
patched 87380 16384 16384 60.00 975236.19
+ netperf -H 192.168.1.1 -t TCP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 50
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. KBytes/sec
current 87380 2048 2048 60.00 17614.79
patched 87380 2048 2048 60.00 17207.46
+ netperf -H 192.168.1.1 -t TCP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 50 -M 1432 -m 1432
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. KBytes/sec
current 87380 2048 1432 60.00 35.28
patched 87380 2048 1432 60.00 35.28
+ netperf -H 192.168.1.1 -t TCP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 9000
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. KBytes/sec
current 87380 18000 18000 60.00 157762.45
patched 87380 18000 18000 60.00 158606.02
+ netperf -H 192.168.1.1 -t TCP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 9000 -M 1432 -m 1432
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. KBytes/sec
current 87380 18000 1432 60.00 78567.39
patched 87380 18000 1432 60.00 78329.98
+ netperf -H 192.168.1.1 -t UDP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # KBytes/sec
current 212992 65507 60.00 248771 0 265238.24
current 212992 60.00 214267 228450.01
patched 212992 65507 60.00 251188 0 267814.90
patched 212992 60.00 235101 250662.67
+ netperf -H 192.168.1.1 -t UDP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 50
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # KBytes/sec
current 2048 2048 60.00 1329653 0 44321.73
current 212992 60.00 1329650 44321.62
patched 2048 2048 60.00 1363257 0 45441.68
patched 212992 60.00 1363253 45441.57
+ netperf -H 192.168.1.1 -t UDP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 50 -M 1432 -m 1432
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # KBytes/sec
current 2048 1432 60.00 1516249 0 35339.61
current 212992 60.00 1516247 35339.56
patched 2048 1432 60.00 1483705 0 34581.11
patched 212992 60.00 1483701 34581.01
+ netperf -H 192.168.1.1 -t UDP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 9000
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # KBytes/sec
current 18000 18000 60.00 540410 26 158322.98
current 212992 60.00 540349 158305.24
patched 18000 18000 60.00 555449 32 162728.98
patched 212992 60.00 555392 162712.28
+ netperf -H 192.168.1.1 -t UDP_STREAM -fK -i10,5 -l 60 -I95,5 -P1 -v2 -- -s 9000 -M 1432 -m 1432
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # KBytes/sec
current 18000 1432 60.00 5144189 0 119896.95
current 212992 60.00 5138354 119760.96
patched 18000 1432 60.00 5104540 0 118972.85
patched 212992 60.00 5099802 118862.44
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists