[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: [Xen-users] low network performance
> -----Original Message----- > From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx > [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of > funk menera > Sent: 31 May 2007 17:32 > To: xen-users@xxxxxxxxxxxxxxxxxxx > Subject: [Xen-users] low network performance > > Hi @list ! > > This is my first posting so please have patience with > me ;) > > As the titel says, i have some problems with the > network performance. > I'm using Xen3.0.3 with Debian on domO and all over > domUs > Maybe the Hardware is interesting too... > AthlonXP 1700+ on MSI.K7T266Pro2 with 1GB RAM, 3 > IDE-HDs without RAID and two Networkdevices with > Realtek chipset: > NIC1: 1GBit r8169 -> for LAN > NIC2: 100MBit 8139 -> for DSL > on the LAN-device is a D-LINK GBit 8-Port Switch > connected. > > Well i experienced the problems while listening to a > mp3 file on a domU using Samba. Or when extracting > sth. from/to a domU form/to my Laptop. Same for a > transport from domU <-> domU > Basically it feels laggy or not as fast as i'm used to > (as comparison: the same hardware on a debian-srv > withou XEN) and there are hang-ups > > > So a short test with netio shows me sth. like this: > Laptop -> domO : 11000 KByte/s for Tx and Rx > domO -> domU-mainsrv : > TCP connection established. > Packet size 1k bytes: 1870 KByte/s Tx, 12293 > KByte/s Rx. > Packet size 2k bytes: 1659 KByte/s Tx, 13503 > KByte/s Rx. > Packet size 4k bytes: 1893 KByte/s Tx, 14072 > KByte/s Rx. > Packet size 8k bytes: 1869 KByte/s Tx, 16355 > KByte/s Rx. > Packet size 16k bytes: 1810 KByte/s Tx, 14836 > KByte/s Rx. > Packet size 32k bytes: 1636 KByte/s Tx, 16829 > KByte/s Rx. > Done. > > as i go on > domU -> domU-efw21(Endian Firewall 2.1): > TCP connection established. > Packet size 1k bytes: 6753 KByte/s Tx, 5481 KByte/s > Rx. > Packet size 2k bytes: 6005 KByte/s Tx, 5532 KByte/s > Rx. > Packet size 4k bytes: 6216 KByte/s Tx, 5271 KByte/s > Rx. > Packet size 8k bytes: 5811 KByte/s Tx, 5491 KByte/s > Rx. > Packet size 16k bytes: 6236 KByte/s Tx, 5603 KByte/s > Rx. > Packet size 32k bytes: 5846 KByte/s Tx, 5588 KByte/s > Rx. > Done. > > That's terrific if i compare it with the speed which i > had before ... 28MByte/s ~ 50MB/s > > So i read some pages in the mailing list archive and > found sth. about para-virtual drivers which looked > like it could give me more bandwidth. > Anyhow i just got confused ... > Is para-virtualsiation meanwhile standard for > Linux-domUs ? > If not where can i get these drivers ? You are already running para-virtualization, since you're on an older Ahtlon XP 1700+, these are a few years away from the current generation that has the AMD-V feature to allow non-para-virtual OS's to run on Xen. > > Maybe the reason for the low network performance is > somewhere else ? Yes, I suspect the real problem here is "CPU congestion". Every packet you send out goes from the DomU to Dom0, then from Dom0 to the othe DomU (in the DomU -> DomU case). Since all three of these domains run on the same processor, the overall latency is much higher than comparing individual systems. Add to this the fact that the virtual system will have a higher overhead simply because there is a hypervisor that want's to control things too [although this is probably only a small part of the puzzle]. There is nothing you can really do about this, other than getting a new system that has more CPU(-core)s to allow the processing of packets to be done more in parallel. > Could someone give me good options for the (tcp_rmem > and so) variables in /et/sysctl.conf ? I found a howto > therefore but changing the values didn't had a > reliable performance enhancement. Probably not worth much, as you've found. > > A other reason could be that the domO is 'sucking' in > network traffic ... ?! > Or that i run the system with 100MHz instead of 133MHz > busfrequency ?! But i don't think so cause i did that > before too. No, that will have a small effect in the whole scheme of things. Of course, Xen and multiple guests will increase the memory bus-traffic, so higher speed processors/memory controller will help to some extent, but the 25% increase in bus-frequency will not give 25% performance improvement here - probably more like 5-10% (or maybe none - depends on how well the caching works in this case). Do you want to have a guess why AMD and Intel are so interested in virtualization? Nothing better for a processor manufacturer than something that sucks up a bit more CPU power and makes customers buy new processors, eh? ;-) -- Mats > > For any hints, tips or links i will be very thankfully > ! > > > Regards, > funki > > > > > > > > > ___________________________________________________________ > Telefonate ohne weitere Kosten vom PC zum PC: > http://messenger.yahoo.de > > _______________________________________________ > Xen-users mailing list > Xen-users@xxxxxxxxxxxxxxxxxxx > http://lists.xensource.com/xen-users > > > _______________________________________________ Xen-users mailing list Xen-users@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-users
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |