Re: Performance of open-iscsi over TCP
Mike Christie escribió: > Miguel Gonzalez Castaños wrote: > >> Erez Zilber escribió: >> >>> Hi, >>> >>> I'm running some performance tests with open-iscsi (iops & throughput). >>> With open-iscsi over TCP, I see very low numbers: >>> >>> * iops: READ - 2, WRITE - 13000 >>> * throughput: READ - 185, WRITE - 185 >>> >>> In open-iscsi.org, I see much higher numbers. Did anyone measure >>> open-iscsi performance lately? Can you share your numbers? Which >>> iscsid.conf file did you use? >>> >>> Thanks, >>> Erez >>> >>> >>> >> I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. >> >> I have tried using open-iscsi and samba mounts and I'm getting around 10 >> MB/s of performance. I've asked in the HP forums: >> >> > > If I have not already, try this > > http://www.open-iscsi.org/bits/open-iscsi-2.0-868-test1.tar.gz > > I am not sure exactly why yet, but we found that with someone's setup we > saw IO taking 6 or 7 seconds to get sent. With that tarball the problem > is fixed and performance is normal. > Apparently there is a big difference between measuring with dd or doing it with disktest. I know the latter is a raw disk test while dd is testing it in the filesystem level but how come there are such big differences? I'm getting 40 MB/s in writes while disktest gets 100 MB/s in writes. Another question, using RAID 1+0 with iSCSI targets could improve the writes performance? The iSCSI server has RAID 5 and that makes the performance really low: 100 MB/s writes 1000 MB/s for reads. Thanks, Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Mike Christie escribió: > Miguel Gonzalez Castaños wrote: > >> Erez Zilber escribió: >> >>> Hi, >>> >>> I'm running some performance tests with open-iscsi (iops & throughput). >>> With open-iscsi over TCP, I see very low numbers: >>> >>> * iops: READ - 2, WRITE - 13000 >>> * throughput: READ - 185, WRITE - 185 >>> >>> In open-iscsi.org, I see much higher numbers. Did anyone measure >>> open-iscsi performance lately? Can you share your numbers? Which >>> iscsid.conf file did you use? >>> >>> Thanks, >>> Erez >>> >>> >>> >> I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. >> >> I have tried using open-iscsi and samba mounts and I'm getting around 10 >> MB/s of performance. I've asked in the HP forums: >> >> > > If I have not already, try this > > http://www.open-iscsi.org/bits/open-iscsi-2.0-868-test1.tar.gz > > I am not sure exactly why yet, but we found that with someone's setup we > saw IO taking 6 or 7 seconds to get sent. With that tarball the problem > is fixed and performance is normal. > Apparently there is a big difference between measuring with dd or doing it with disktest. I know the latter is a raw disk test while dd is testing it in the filesystem level but how come there are such big differences? I'm getting 40 MB/s in writes while disktest gets 100 MB/s in writes. Another question, using RAID 1+0 with iSCSI targets could improve the writes performance? The iSCSI server has RAID 5 and that makes the performance really low: 100 MB/s writes 1000 MB/s for reads. Thanks, Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
On Jan 21, 5:28 pm, Mike Christie <[EMAIL PROTECTED]> wrote: > Miguel Gonzalez Castaños wrote: > > Erez Zilber escribió: > >> Hi, > > >> I'm running some performance tests with open-iscsi (iops & throughput). > >> With open-iscsi over TCP, I see very low numbers: > > >> * iops: READ - 2, WRITE - 13000 > >> * throughput: READ - 185, WRITE - 185 > > >> In open-iscsi.org, I see much higher numbers. Did anyone measure > >> open-iscsi performance lately? Can you share your numbers? Which > >> iscsid.conf file did you use? > > >> Thanks, > >> Erez > > > I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. > > > I have tried using open-iscsi and samba mounts and I'm getting around 10 > > MB/s of performance. I've asked in the HP forums: > > If I have not already, try this > > http://www.open-iscsi.org/bits/open-iscsi-2.0-868-test1.tar.gz > > I am not sure exactly why yet, but we found that with someone's setup we > saw IO taking 6 or 7 seconds to get sent. With that tarball the problem > is fixed and performance is normal. Apparently there is a big difference between measuring with dd or doing it with disktest. I know the latter is a raw disk test while dd is testing it in the filesystem level but how come there are such big differences? I'm getting 40 MB/s in writes while disktest gets 100 MB/s in writes. Another question, using RAID 1+0 with iSCSI targets could improve the writes performance? The iSCSI server has RAID 5 and that makes the performance really low: 100 MB/s writes 1000 MB/s for reads. Thanks, Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Miguel Gonzalez Castaños wrote: > Erez Zilber escribió: >> Hi, >> >> I'm running some performance tests with open-iscsi (iops & throughput). >> With open-iscsi over TCP, I see very low numbers: >> >> * iops: READ - 2, WRITE - 13000 >> * throughput: READ - 185, WRITE - 185 >> >> In open-iscsi.org, I see much higher numbers. Did anyone measure >> open-iscsi performance lately? Can you share your numbers? Which >> iscsid.conf file did you use? >> >> Thanks, >> Erez >> >> > I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. > > I have tried using open-iscsi and samba mounts and I'm getting around 10 > MB/s of performance. I've asked in the HP forums: > If I have not already, try this http://www.open-iscsi.org/bits/open-iscsi-2.0-868-test1.tar.gz I am not sure exactly why yet, but we found that with someone's setup we saw IO taking 6 or 7 seconds to get sent. With that tarball the problem is fixed and performance is normal. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
On Wed, Jan 16, 2008 at 10:45:34AM -0500, Miguel Gonzalez Castaños wrote: > > Pasi Kärkkäinen escribió: > > On Wed, Jan 09, 2008 at 11:13:26AM -0500, Miguel Gonzalez Castaños wrote: > > > >> > >>> Well, this is your problem. Fix it first. > >>> > >>> If you can't get better throughput with FTP/HTTP/CIFS, how could you with > >>> iSCSI? > >>> > >>> Sounds like you're running at 100 Mbit/sec. > >>> > >>> > >> I'm researching on this matter. I have tested from the machine where > >> Virtual Server is running using hrping: > >> > >> hrping -l 4096 -t 10.0.6.41 > >> > >> Statistics for 10.0.6.41: > >> [Aborting...] > >> Packets: sent=70, rcvd=70, error=0, lost=0 (0% loss) in 34.500313 sec > >> RTTs of replies in ms: min/avg/max: 0.304 / 0.331 / 0.426 > >> > >> from your calculations that means around 13 MB/s > >> > >> Even at my home LAN where I have a very cheap gigabit switch and I use > >> CAT 5e (so I'd meant to get around 200 Mb/s) and I'm getting with hrping > >> around 15 MB/s > >> > >> Maybe We have something wrong with our switches (my boss says that all > >> are gigabit) or with our cabling. > >> > >> Thanks! > >> > >> > > > > OK. > > > > Remember that 4k I mentioned was just an example, because that's the often > > used block size by many filesystems.. > > > > If you do (or if your application does) larger requests, you can easily get > > much more throughput.. for example 128k requests will give you much more > > throughput.. > > > > And then you can have many outstanding io's active at the same time.. > > depending on the used queue depth, io elevator etc.. > > > > iometer (on windows) let's you choose number of outstanding io's.. > > > > But yeah, first fix the throughput to be good with FTP/HTTP/CIFS and then > > start playing with iSCSI. > > > I've connected a cross-over cable and created an ad-hoc network. hrping > is not being very meaningful: > > hrping -L 4096 -t 10.0.7.41 > > RTTs of replies in ms: min/avg/max: 0.228 / 0.266 / 1.121 > > However, I have installed bing. The performance it gives is around 500 > Mbps in the cross-over network. > 500 Mbps sounds OK, but not very good.. you should get more. Have you done any tcp/ip stack option tweaking? There is a lot of network memory/socket options to tune for gigabit links/transfers.. (at least in Linux). Also some network driver settings/parameters affect the performance. > The problem here is both machines are running Windows, the open-iscsi > client would be running in a Virtual Server machine running Debian. > Ouch. I have never measured performance of Linux VM under MS Virtual server.. so no idea about that. > Running bing from the host Virtual Server against the debian virtual > machine reduces the performance to around 60 Mbps. The tulip driver in > Debian for Virtual Server configures a Fast Ethernet network card. > Although Microsoft claims that the real performance is limited by the > physical network, apparently is not the case for Debian virtual machines. > Well, here you go.. if you only get 60 Mbit/sec between the host and the Linux VM, that's the problem.. Can you change the emulated NIC to something else from the Linux VM? Does MS have optimized (=paravirtualized) NIC drivers available for use in the Linux VM? Have you tried VMware Server or ESX? > About the MTU, I get 1500 in the client part, in the initiator. Do I > need a bigger MTU? Is it possible to change that? > I don't know what MS Virtual Server supports.. you might get better performance with jumbo frames (9000 bytes), but not necessarily.. it depends a lot of the switches used etc. Good ethernet flow control implementation is more important for iSCSI than jumbo frames. If you need choose between flow control and jumbo frames, choose flow control.. (some switches can't do both at the same time - and many switches have bad flow control implementation - so be careful). -- Pasi --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Pasi Kärkkäinen escribió: > On Wed, Jan 09, 2008 at 11:13:26AM -0500, Miguel Gonzalez Castaños wrote: > >> >>> Well, this is your problem. Fix it first. >>> >>> If you can't get better throughput with FTP/HTTP/CIFS, how could you with >>> iSCSI? >>> >>> Sounds like you're running at 100 Mbit/sec. >>> >>> >> I'm researching on this matter. I have tested from the machine where >> Virtual Server is running using hrping: >> >> hrping -l 4096 -t 10.0.6.41 >> >> Statistics for 10.0.6.41: >> [Aborting...] >> Packets: sent=70, rcvd=70, error=0, lost=0 (0% loss) in 34.500313 sec >> RTTs of replies in ms: min/avg/max: 0.304 / 0.331 / 0.426 >> >> from your calculations that means around 13 MB/s >> >> Even at my home LAN where I have a very cheap gigabit switch and I use >> CAT 5e (so I'd meant to get around 200 Mb/s) and I'm getting with hrping >> around 15 MB/s >> >> Maybe We have something wrong with our switches (my boss says that all >> are gigabit) or with our cabling. >> >> Thanks! >> >> > > OK. > > Remember that 4k I mentioned was just an example, because that's the often > used block size by many filesystems.. > > If you do (or if your application does) larger requests, you can easily get > much more throughput.. for example 128k requests will give you much more > throughput.. > > And then you can have many outstanding io's active at the same time.. > depending on the used queue depth, io elevator etc.. > > iometer (on windows) let's you choose number of outstanding io's.. > > But yeah, first fix the throughput to be good with FTP/HTTP/CIFS and then > start playing with iSCSI. > I've connected a cross-over cable and created an ad-hoc network. hrping is not being very meaningful: hrping -L 4096 -t 10.0.7.41 RTTs of replies in ms: min/avg/max: 0.228 / 0.266 / 1.121 However, I have installed bing. The performance it gives is around 500 Mbps in the cross-over network. The problem here is both machines are running Windows, the open-iscsi client would be running in a Virtual Server machine running Debian. Running bing from the host Virtual Server against the debian virtual machine reduces the performance to around 60 Mbps. The tulip driver in Debian for Virtual Server configures a Fast Ethernet network card. Although Microsoft claims that the real performance is limited by the physical network, apparently is not the case for Debian virtual machines. About the MTU, I get 1500 in the client part, in the initiator. Do I need a bigger MTU? Is it possible to change that? Thanks, Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
On Wed, Jan 09, 2008 at 11:13:26AM -0500, Miguel Gonzalez Castaños wrote: > > > > > > Well, this is your problem. Fix it first. > > > > If you can't get better throughput with FTP/HTTP/CIFS, how could you with > > iSCSI? > > > > Sounds like you're running at 100 Mbit/sec. > > > I'm researching on this matter. I have tested from the machine where > Virtual Server is running using hrping: > > hrping -l 4096 -t 10.0.6.41 > > Statistics for 10.0.6.41: > [Aborting...] > Packets: sent=70, rcvd=70, error=0, lost=0 (0% loss) in 34.500313 sec > RTTs of replies in ms: min/avg/max: 0.304 / 0.331 / 0.426 > > from your calculations that means around 13 MB/s > > Even at my home LAN where I have a very cheap gigabit switch and I use > CAT 5e (so I'd meant to get around 200 Mb/s) and I'm getting with hrping > around 15 MB/s > > Maybe We have something wrong with our switches (my boss says that all > are gigabit) or with our cabling. > > Thanks! > OK. Remember that 4k I mentioned was just an example, because that's the often used block size by many filesystems.. If you do (or if your application does) larger requests, you can easily get much more throughput.. for example 128k requests will give you much more throughput.. And then you can have many outstanding io's active at the same time.. depending on the used queue depth, io elevator etc.. iometer (on windows) let's you choose number of outstanding io's.. But yeah, first fix the throughput to be good with FTP/HTTP/CIFS and then start playing with iSCSI. -- Pasi --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
> > Well, this is your problem. Fix it first. > > If you can't get better throughput with FTP/HTTP/CIFS, how could you with > iSCSI? > > Sounds like you're running at 100 Mbit/sec. > I'm researching on this matter. I have tested from the machine where Virtual Server is running using hrping: hrping -l 4096 -t 10.0.6.41 Statistics for 10.0.6.41: [Aborting...] Packets: sent=70, rcvd=70, error=0, lost=0 (0% loss) in 34.500313 sec RTTs of replies in ms: min/avg/max: 0.304 / 0.331 / 0.426 from your calculations that means around 13 MB/s Even at my home LAN where I have a very cheap gigabit switch and I use CAT 5e (so I'd meant to get around 200 Mb/s) and I'm getting with hrping around 15 MB/s Maybe We have something wrong with our switches (my boss says that all are gigabit) or with our cabling. Thanks! Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
On Wed, Jan 09, 2008 at 06:07:10AM -0500, Miguel Gonzalez Castaños wrote: > > > > 10 MB/sec sounds really poor. > > > > What kind of disk performance do you get locally on the target/server from > > the SATA RAID array? You can try iometer or some other disk benchmarking > > tools. > > > I've downloaded iometer, can you tell me what do I have to test and how? You can try running tests with different block/request sizes.. with small requests (512 bytes) you can test the maximum IOPS you can get, and with large request sizes (64 kB and more) you can test the maximum throughput you can get.. > > What kind of latency (ping roundtrip with 4k packets) do you have from the > > initiator to the target? latency is one of the key factors limiting IOPS > > you will get. > > > ping -s 4048 10.0.6.41 > > 4056 bytes from 10.0.6.41: icmp_seq=179 ttl=128 time=0.795 ms > > --- 10.0.6.41 ping statistics --- > 179 packets transmitted, 178 received, 0% packet loss, time 178261ms > rtt min/avg/max/mdev = 0.789/1.429/1.990/0.534 ms > You should try with 4096 bytes packets.. but anyway, that's close enough. 1000ms / 0.795 ms == 1258 IOPS That's your maximum IO operations per second you can get (assuming 1 outstanding IO): 1258 * 4k = 5032k That's around 5 MB/sec. > > How's your network between the target and initiator? gigabit? jumbo frames? > > flow control? what kind of switch are you using? > > > gigabit in both ends. The switch I'm not in front of the machine so I > can't tell you until next week but my guess is that's a regular switch. > I think that I'm going to try a crossover cable of CAT 6 to avoid any > other issue. I think jumbo frames are deactivated but how can I know? Yep, testing with straight cable between the target and the initiator is a good idea. > > How's the performance with FTP/HTTP/CIFS between the target and the > > initiator? > > > It is the same as mentioned before: 10 MB/s > Well, this is your problem. Fix it first. If you can't get better throughput with FTP/HTTP/CIFS, how could you with iSCSI? Sounds like you're running at 100 Mbit/sec. -- Pasi --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
RE: Performance of open-iscsi over TCP
Use ifconfig to check the MTU on your eth device If it is bigger than 1500 than u use jumbo. You will need to check on the switch as well. Make sure all the MTU's in your path are the same. You can also check the mss (MTU-headers) -Original Message- From: open-iscsi@googlegroups.com [mailto:[EMAIL PROTECTED] On Behalf Of Miguel Gonzalez Casta?os Sent: Wednesday, January 09, 2008 1:07 PM To: open-iscsi@googlegroups.com Subject: Re: Performance of open-iscsi over TCP > 10 MB/sec sounds really poor. > > What kind of disk performance do you get locally on the target/server > from the SATA RAID array? You can try iometer or some other disk benchmarking tools. > I've downloaded iometer, can you tell me what do I have to test and how? > What kind of latency (ping roundtrip with 4k packets) do you have from > the initiator to the target? latency is one of the key factors > limiting IOPS you will get. > ping -s 4048 10.0.6.41 4056 bytes from 10.0.6.41: icmp_seq=179 ttl=128 time=0.795 ms --- 10.0.6.41 ping statistics --- 179 packets transmitted, 178 received, 0% packet loss, time 178261ms rtt min/avg/max/mdev = 0.789/1.429/1.990/0.534 ms > How's your network between the target and initiator? gigabit? jumbo frames? > flow control? what kind of switch are you using? > gigabit in both ends. The switch I'm not in front of the machine so I can't tell you until next week but my guess is that's a regular switch. I think that I'm going to try a crossover cable of CAT 6 to avoid any other issue. I think jumbo frames are deactivated but how can I know? > How's the performance with FTP/HTTP/CIFS between the target and the > initiator? > It is the same as mentioned before: 10 MB/s Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
> 10 MB/sec sounds really poor. > > What kind of disk performance do you get locally on the target/server from > the SATA RAID array? You can try iometer or some other disk benchmarking > tools. > I've downloaded iometer, can you tell me what do I have to test and how? > What kind of latency (ping roundtrip with 4k packets) do you have from the > initiator to the target? latency is one of the key factors limiting IOPS > you will get. > ping -s 4048 10.0.6.41 4056 bytes from 10.0.6.41: icmp_seq=179 ttl=128 time=0.795 ms --- 10.0.6.41 ping statistics --- 179 packets transmitted, 178 received, 0% packet loss, time 178261ms rtt min/avg/max/mdev = 0.789/1.429/1.990/0.534 ms > How's your network between the target and initiator? gigabit? jumbo frames? > flow control? what kind of switch are you using? > gigabit in both ends. The switch I'm not in front of the machine so I can't tell you until next week but my guess is that's a regular switch. I think that I'm going to try a crossover cable of CAT 6 to avoid any other issue. I think jumbo frames are deactivated but how can I know? > How's the performance with FTP/HTTP/CIFS between the target and the > initiator? > It is the same as mentioned before: 10 MB/s Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
On Tue, Jan 08, 2008 at 07:37:14PM -0500, Miguel Gonzalez Castaños wrote: > > Erez Zilber escribió: > > Hi, > > > > I'm running some performance tests with open-iscsi (iops & throughput). > > With open-iscsi over TCP, I see very low numbers: > > > > * iops: READ - 2, WRITE - 13000 > > * throughput: READ - 185, WRITE - 185 > > > > In open-iscsi.org, I see much higher numbers. Did anyone measure > > open-iscsi performance lately? Can you share your numbers? Which > > iscsid.conf file did you use? > > > > Thanks, > > Erez > > > > > I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. > > I have tried using open-iscsi and samba mounts and I'm getting around 10 > MB/s of performance. I've asked in the HP forums: > > http://forums12.itrc.hp.com/service/forums/questionanswer.do?admit=109447627+1199836532175+28353475&threadId=1155370 > > And they say that this is the normal performance of iSCSI with SATA > drives, is that true? > 10 MB/sec sounds really poor. What kind of disk performance do you get locally on the target/server from the SATA RAID array? You can try iometer or some other disk benchmarking tools. What kind of latency (ping roundtrip with 4k packets) do you have from the initiator to the target? latency is one of the key factors limiting IOPS you will get. How's your network between the target and initiator? gigabit? jumbo frames? flow control? what kind of switch are you using? How's the performance with FTP/HTTP/CIFS between the target and the initiator? -- Pasi --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Erez Zilber escribió: > Hi, > > I'm running some performance tests with open-iscsi (iops & throughput). > With open-iscsi over TCP, I see very low numbers: > > * iops: READ - 2, WRITE - 13000 > * throughput: READ - 185, WRITE - 185 > > In open-iscsi.org, I see much higher numbers. Did anyone measure > open-iscsi performance lately? Can you share your numbers? Which > iscsid.conf file did you use? > > Thanks, > Erez > > I'm having a HP AiO 600 Storage Server with SATA drives in RAID 5. I have tried using open-iscsi and samba mounts and I'm getting around 10 MB/s of performance. I've asked in the HP forums: http://forums12.itrc.hp.com/service/forums/questionanswer.do?admit=109447627+1199836532175+28353475&threadId=1155370 And they say that this is the normal performance of iSCSI with SATA drives, is that true? Thanks, Miguel --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Did anybody use the open-iscsi from the Fedora distro ? if is it positive, how was your impression of it ?? Thank´s! -- Marcos G. M. Santos SysAdmin - DIGILAB S.A. Tel: 55 48 3234 4041 www.digilab.com.br --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Sagi Rotem wrote: > Iops on such large messages ? It was probably low ~4K > I had used default iscsid.conf and mapped nullio lun in the target. > Used iscsitarget-0.4.15 over RH5 Intel cpu 2Ghz. > Did you check your cpu utilization ? > CPU utilization is low on both sides (initiator & target). Erez --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Mike Christie wrote: > Erez Zilber wrote: > >> Hi, >> >> I'm running some performance tests with open-iscsi (iops & throughput). >> With open-iscsi over TCP, I see very low numbers: >> >> * iops: READ - 2, WRITE - 13000 >> * throughput: READ - 185, WRITE - 185 >> >> In open-iscsi.org, I see much higher numbers. Did anyone measure >> open-iscsi performance lately? Can you share your numbers? Which >> > > Yeah, I have been running tests for a while. What test program are you > using, what io sizes, and what io scheduler and kernel and what nic module? > Now, I'm running the following benchmarks: * For BW checks: sgp_dd bs=512 of=/dev/null if=/dev/sg2 bpt=1024 thr=8 count=2048 time=1 dio=1 (and a similar command for write ops) * For small iops: sgp_dd bs=512 of=/dev/null if=/dev/sg2 bpt=2 thr=8 time=1 count=1000k time=1 (and a similar command for write ops) I'm using v2.0-865.15 (user & kernel) from OFED 1.3 on SLES 10 sp1. I'm running over IPoIB on a ConnectX IB HCA. Now, I see the following numbers: * BW: read - 260 MB/sec, write - 190 MB/sec * iops: read - 27000, write - 17000 Actually, I'm trying to improve the performance of open-iscsi over iSER. I justed wanted to compare it to open-iscsi over TCP (because the numbers in open-iscsi.org look very impressive). > And what are the throughput values in? Is that 185 KB/s. > No, it was MB/sec. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Mike Christie wrote: > Mike Christie wrote: >> Erez Zilber wrote: >>> Hi, >>> >>> I'm running some performance tests with open-iscsi (iops & throughput). >>> With open-iscsi over TCP, I see very low numbers: >>> >>> * iops: READ - 2, WRITE - 13000 >>> * throughput: READ - 185, WRITE - 185 >>> >>> In open-iscsi.org, I see much higher numbers. Did anyone measure >>> open-iscsi performance lately? Can you share your numbers? Which >> Yeah, I have been running tests for a while. What test program are you >> using, what io sizes, and what io scheduler and kernel and what nic module? >> > > Oh yeah what kernel are you running? I ported linux-iscsi to a recent > kernel so I could test if there are a problem with open-iscsi or > something in the network, block or scsi layer (this is why I tried > switching open-iscsi to use a recv thread like how linux-iscsi does > instead of running from the network softirq). > > I will try to tar up my linux-iscsi code, so you can try it out on your > kernel. > Oh ignore that. I am not seeing any write problems (have to switch to noop io sched to get things going sometimes though). I just see slow reads, so we are not hitting the same problem since both your numbers look to be the same. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Mike Christie wrote: > Erez Zilber wrote: >> Hi, >> >> I'm running some performance tests with open-iscsi (iops & throughput). >> With open-iscsi over TCP, I see very low numbers: >> >> * iops: READ - 2, WRITE - 13000 >> * throughput: READ - 185, WRITE - 185 >> >> In open-iscsi.org, I see much higher numbers. Did anyone measure >> open-iscsi performance lately? Can you share your numbers? Which > > Yeah, I have been running tests for a while. What test program are you > using, what io sizes, and what io scheduler and kernel and what nic module? > Oh yeah what kernel are you running? I ported linux-iscsi to a recent kernel so I could test if there are a problem with open-iscsi or something in the network, block or scsi layer (this is why I tried switching open-iscsi to use a recv thread like how linux-iscsi does instead of running from the network softirq). I will try to tar up my linux-iscsi code, so you can try it out on your kernel. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Erez Zilber wrote: > Hi, > > I'm running some performance tests with open-iscsi (iops & throughput). > With open-iscsi over TCP, I see very low numbers: > > * iops: READ - 2, WRITE - 13000 > * throughput: READ - 185, WRITE - 185 > > In open-iscsi.org, I see much higher numbers. Did anyone measure > open-iscsi performance lately? Can you share your numbers? Which Yeah, I have been running tests for a while. What test program are you using, what io sizes, and what io scheduler and kernel and what nic module? And what are the throughput values in? Is that 185 KB/s. With smaller IOs I get really bad iop numbers. We just talked about this on the list. For throughput if I use larger IOs I get this though (this is from some tests I did when I was testing what I was putting in git): disktest -PT -T30 -h1 -K32 -B256k -ID /dev/sdb -D 0:100 | 2007/11/18-12:54:17 | STAT | 4176 | v1.2.8 | /dev/sdb | Write throughput: 117615274.7B/s (112.17MB/s), IOPS 454.0/s. disktest -PT -T30 -h1 -K32 -B256k -ID /dev/sdb | 2007/11/18-12:49:58 | STAT | 3749 | v1.2.8 | /dev/sdb | Read throughput: 96521420.8B/s (92.05MB/s), IOPS 374.6/s. Normally for reads I also get 112 MB/s. For some reason with my home setup where I tool those numbers though, I am getting really bad read numbers. I did a patch to switch the read path to always use a thread and then the throughput went back to 112. Not sure if you are hitting that problem, but I also noticed with some targets and workloads I need to switch to noop instead of cfq or throughput drops to 3-12 MB/s even with large IOs like above. This is using the default values in the iscsid.conf. --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
RE: Performance of open-iscsi over TCP
Iops on such large messages ? It was probably low ~4K I had used default iscsid.conf and mapped nullio lun in the target. Used iscsitarget-0.4.15 over RH5 Intel cpu 2Ghz. Did you check your cpu utilization ? -Original Message- From: open-iscsi@googlegroups.com [mailto:[EMAIL PROTECTED] On Behalf Of Erez Zilber Sent: Monday, December 31, 2007 6:56 PM To: open-iscsi@googlegroups.com Subject: Re: Performance of open-iscsi over TCP Sagi Rotem wrote: > What message size ? What kind of link/nic do u use ? What BM ? > I had measured ~900MB/sec reads and writes over 10Gig with 256K > messages using disktest BM. > > -Original Message- > From: open-iscsi@googlegroups.com [mailto:[EMAIL PROTECTED] > On Behalf Of Erez Zilber > Sent: Monday, December 31, 2007 6:16 PM > To: open-iscsi@googlegroups.com > Subject: Performance of open-iscsi over TCP > > You're talking only about throughput (not iops). Here's more info: * benchmark: sgp_dd with 512K messages. * Target: IPStor with a RAM disk (so the storage isn't the bottleneck) * NIC: DDR IB HCAs (running IPoIB) * I'm using the default iscsid.conf that comes with open-iscsi Which target are you using? Are you suing the deafult iscsid.conf? Anything else? Did you measure iops? Thanks, Erez --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
Re: Performance of open-iscsi over TCP
Sagi Rotem wrote: > What message size ? What kind of link/nic do u use ? What BM ? > I had measured ~900MB/sec reads and writes over 10Gig with 256K messages > using disktest BM. > > -Original Message- > From: open-iscsi@googlegroups.com [mailto:[EMAIL PROTECTED] > On Behalf Of Erez Zilber > Sent: Monday, December 31, 2007 6:16 PM > To: open-iscsi@googlegroups.com > Subject: Performance of open-iscsi over TCP > > You're talking only about throughput (not iops). Here's more info: * benchmark: sgp_dd with 512K messages. * Target: IPStor with a RAM disk (so the storage isn't the bottleneck) * NIC: DDR IB HCAs (running IPoIB) * I'm using the default iscsid.conf that comes with open-iscsi Which target are you using? Are you suing the deafult iscsid.conf? Anything else? Did you measure iops? Thanks, Erez --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---
RE: Performance of open-iscsi over TCP
What message size ? What kind of link/nic do u use ? What BM ? I had measured ~900MB/sec reads and writes over 10Gig with 256K messages using disktest BM. -Original Message- From: open-iscsi@googlegroups.com [mailto:[EMAIL PROTECTED] On Behalf Of Erez Zilber Sent: Monday, December 31, 2007 6:16 PM To: open-iscsi@googlegroups.com Subject: Performance of open-iscsi over TCP Hi, I'm running some performance tests with open-iscsi (iops & throughput). With open-iscsi over TCP, I see very low numbers: * iops: READ - 2, WRITE - 13000 * throughput: READ - 185, WRITE - 185 In open-iscsi.org, I see much higher numbers. Did anyone measure open-iscsi performance lately? Can you share your numbers? Which iscsid.conf file did you use? Thanks, Erez --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "open-iscsi" group. To post to this group, send email to open-iscsi@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/open-iscsi -~--~~~~--~~--~--~---