Thread: Bad RAID1 read performance
Hi,
after doing the "dd" tests for a server we have at work I obtained:
Read: 47.20 Mb/s
Write: 39.82 Mb/s
Some days ago read performance was around 20Mb/s due to no readahead in md0
so I modified it using hdparm. However, it seems to me that being it a RAID1
read speed could be much better. These are SATA disks with 3Gb of RAM so I
did 'time bash -c "dd if=/dev/zero of=bigfile bs=8k count=786432 && sync"'.
File system is ext3 (if read many times in the list that XFS is faster), but
I don't want to change the file system right now. Modifing the readahead from
the current 1024k to 2048k doesn't make any difference. Are there any other
tweaks I can make?
This sounds like a bad RAID controller - are you using a built-in hardware RAID? If so, you will likely want to use Linux software RAID instead. Also - you might want to try a 512KB readahead - I've found that is optimal for RAID1 on some RAID controllers. - Luke On 5/30/07 2:35 AM, "Albert Cervera Areny" <albert@sedifa.com> wrote: > Hi, > after doing the "dd" tests for a server we have at work I obtained: > Read: 47.20 Mb/s > Write: 39.82 Mb/s > Some days ago read performance was around 20Mb/s due to no readahead in md0 > so I modified it using hdparm. However, it seems to me that being it a RAID1 > read speed could be much better. These are SATA disks with 3Gb of RAM so I > did 'time bash -c "dd if=/dev/zero of=bigfile bs=8k count=786432 && sync"'. > File system is ext3 (if read many times in the list that XFS is faster), but > I don't want to change the file system right now. Modifing the readahead from > the current 1024k to 2048k doesn't make any difference. Are there any other > tweaks I can make? > > > ---------------------------(end of broadcast)--------------------------- > TIP 4: Have you searched our list archives? > > http://archives.postgresql.org >
Hardware isn't very good I believe, and it's about 2-3 years old, but the RAID is Linux software, and though not very good the difference between reading and writing should probably be greater... (?) Would you set 512Kb readahead on both drives and RAID? I tried various configurations and none seemed to make a big difference. It seemed correct to me to set 512kb per drive and 1024kb for md0. A Dimecres 30 Maig 2007 16:09, Luke Lonergan va escriure: > This sounds like a bad RAID controller - are you using a built-in hardware > RAID? If so, you will likely want to use Linux software RAID instead. > > Also - you might want to try a 512KB readahead - I've found that is optimal > for RAID1 on some RAID controllers. > > - Luke > > On 5/30/07 2:35 AM, "Albert Cervera Areny" <albert@sedifa.com> wrote: > > Hi, > > after doing the "dd" tests for a server we have at work I obtained: > > Read: 47.20 Mb/s > > Write: 39.82 Mb/s > > Some days ago read performance was around 20Mb/s due to no readahead in > > md0 so I modified it using hdparm. However, it seems to me that being it > > a RAID1 read speed could be much better. These are SATA disks with 3Gb of > > RAM so I did 'time bash -c "dd if=/dev/zero of=bigfile bs=8k count=786432 > > && sync"'. File system is ext3 (if read many times in the list that XFS > > is faster), but I don't want to change the file system right now. > > Modifing the readahead from the current 1024k to 2048k doesn't make any > > difference. Are there any other tweaks I can make? > > > > > > ---------------------------(end of broadcast)--------------------------- > > TIP 4: Have you searched our list archives? > > > > http://archives.postgresql.org > > ---------------------------(end of broadcast)--------------------------- > TIP 6: explain analyze is your friend -- Albert Cervera Areny Dept. Informàtica Sedifa, S.L. Av. Can Bordoll, 149 08202 - Sabadell (Barcelona) Tel. 93 715 51 11 Fax. 93 715 51 12 ==================================================================== ........................ AVISO LEGAL ............................ La presente comunicación y sus anexos tiene como destinatario la persona a la que va dirigida, por lo que si usted lo recibe por error debe notificarlo al remitente y eliminarlo de su sistema, no pudiendo utilizarlo, total o parcialmente, para ningún fin. Su contenido puede tener información confidencial o protegida legalmente y únicamente expresa la opinión del remitente. El uso del correo electrónico vía Internet no permite asegurar ni la confidencialidad de los mensajes ni su correcta recepción. En el caso de que el destinatario no consintiera la utilización del correo electrónico, deberá ponerlo en nuestro conocimiento inmediatamente. ==================================================================== ........................... DISCLAIMER ............................. This message and its attachments are intended exclusively for the named addressee. If you receive this message in error, please immediately delete it from your system and notify the sender. You may not use this message or any part of it for any purpose. The message may contain information that is confidential or protected by law, and any opinions expressed are those of the individual sender. Internet e-mail guarantees neither the confidentiality nor the proper receipt of the message sent. If the addressee of this message does not consent to the use of internet e-mail, please inform us inmmediately. ====================================================================
As there is no 'continuous space' option on ext3/ext2 (or probably "-f fragment_size" may do a trick?) - I think after some filesystem activity you simply loose continuous space allocation and rather expected sequential reading may be transformed into random seeking of 'logically' sequentual blocks... Rgds, -Dimitri On 5/30/07, Albert Cervera Areny <albert@sedifa.com> wrote: > Hardware isn't very good I believe, and it's about 2-3 years old, but the > RAID > is Linux software, and though not very good the difference between reading > and writing should probably be greater... (?) > > Would you set 512Kb readahead on both drives and RAID? I tried various > configurations and none seemed to make a big difference. It seemed correct > to > me to set 512kb per drive and 1024kb for md0. > > A Dimecres 30 Maig 2007 16:09, Luke Lonergan va escriure: > > This sounds like a bad RAID controller - are you using a built-in hardware > > RAID? If so, you will likely want to use Linux software RAID instead. > > > > Also - you might want to try a 512KB readahead - I've found that is > optimal > > for RAID1 on some RAID controllers. > > > > - Luke > > > > On 5/30/07 2:35 AM, "Albert Cervera Areny" <albert@sedifa.com> wrote: > > > Hi, > > > after doing the "dd" tests for a server we have at work I obtained: > > > Read: 47.20 Mb/s > > > Write: 39.82 Mb/s > > > Some days ago read performance was around 20Mb/s due to no readahead in > > > md0 so I modified it using hdparm. However, it seems to me that being it > > > a RAID1 read speed could be much better. These are SATA disks with 3Gb > of > > > RAM so I did 'time bash -c "dd if=/dev/zero of=bigfile bs=8k > count=786432 > > > && sync"'. File system is ext3 (if read many times in the list that XFS > > > is faster), but I don't want to change the file system right now. > > > Modifing the readahead from the current 1024k to 2048k doesn't make any > > > difference. Are there any other tweaks I can make? > > > > > > > > > ---------------------------(end of broadcast)--------------------------- > > > TIP 4: Have you searched our list archives? > > > > > > http://archives.postgresql.org > > > > ---------------------------(end of broadcast)--------------------------- > > TIP 6: explain analyze is your friend > > -- > Albert Cervera Areny > Dept. Informàtica Sedifa, S.L. > > Av. Can Bordoll, 149 > 08202 - Sabadell (Barcelona) > Tel. 93 715 51 11 > Fax. 93 715 51 12 > > ==================================================================== > ........................ AVISO LEGAL ............................ > La presente comunicación y sus anexos tiene como destinatario la > persona a la que va dirigida, por lo que si usted lo recibe > por error debe notificarlo al remitente y eliminarlo de su > sistema, no pudiendo utilizarlo, total o parcialmente, para > ningún fin. Su contenido puede tener información confidencial o > protegida legalmente y únicamente expresa la opinión del > remitente. El uso del correo electrónico vía Internet no > permite asegurar ni la confidencialidad de los mensajes > ni su correcta recepción. En el caso de que el > destinatario no consintiera la utilización del correo electrónico, > deberá ponerlo en nuestro conocimiento inmediatamente. > ==================================================================== > ........................... DISCLAIMER ............................. > This message and its attachments are intended exclusively for the > named addressee. If you receive this message in error, please > immediately delete it from your system and notify the sender. You > may not use this message or any part of it for any purpose. > The message may contain information that is confidential or > protected by law, and any opinions expressed are those of the > individual sender. Internet e-mail guarantees neither the > confidentiality nor the proper receipt of the message sent. > If the addressee of this message does not consent to the use > of internet e-mail, please inform us inmmediately. > ==================================================================== > > > > > ---------------------------(end of broadcast)--------------------------- > TIP 1: if posting/reading through Usenet, please send an appropriate > subscribe-nomail command to majordomo@postgresql.org so that your > message can get through to the mailing list cleanly >
Albert, On 5/30/07 8:00 AM, "Albert Cervera Areny" <albert@sedifa.com> wrote: > Hardware isn't very good I believe, and it's about 2-3 years old, but the RAID > is Linux software, and though not very good the difference between reading > and writing should probably be greater... (?) Not for one thread/process of I/O. Mirror sets can nearly double the read performance on most RAID adapters or SW RAID when using two or more thread/processes, but a single thread will get one drive worth of performance. You should try running two simultaneous processes during reading and see what you get. > Would you set 512Kb readahead on both drives and RAID? I tried various > configurations and none seemed to make a big difference. It seemed correct to > me to set 512kb per drive and 1024kb for md0. Shouldn't matter that much, but yes, each drive getting half the readahead is a good strategy. Try 256+256 and 512. The problem you have is likely not related to the readahead though - I suggest you try read/write to a single disk and see what you get. You should get around 60 MB/s if the drive is a modern 7200 RPM SATA disk. If you aren't getting that on a single drive, there's something wrong with the SATA driver or the drive(s). - Luke > A Dimecres 30 Maig 2007 16:09, Luke Lonergan va escriure: >> This sounds like a bad RAID controller - are you using a built-in hardware >> RAID? If so, you will likely want to use Linux software RAID instead. >> >> Also - you might want to try a 512KB readahead - I've found that is optimal >> for RAID1 on some RAID controllers. >> >> - Luke >> >> On 5/30/07 2:35 AM, "Albert Cervera Areny" <albert@sedifa.com> wrote: >>> Hi, >>> after doing the "dd" tests for a server we have at work I obtained: >>> Read: 47.20 Mb/s >>> Write: 39.82 Mb/s >>> Some days ago read performance was around 20Mb/s due to no readahead in >>> md0 so I modified it using hdparm. However, it seems to me that being it >>> a RAID1 read speed could be much better. These are SATA disks with 3Gb of >>> RAM so I did 'time bash -c "dd if=/dev/zero of=bigfile bs=8k count=786432 >>> && sync"'. File system is ext3 (if read many times in the list that XFS >>> is faster), but I don't want to change the file system right now. >>> Modifing the readahead from the current 1024k to 2048k doesn't make any >>> difference. Are there any other tweaks I can make? >>> >>> >>> ---------------------------(end of broadcast)--------------------------- >>> TIP 4: Have you searched our list archives? >>> >>> http://archives.postgresql.org >> >> ---------------------------(end of broadcast)--------------------------- >> TIP 6: explain analyze is your friend
As you suggested with two threads I get 42.39 Mb/s in one and 40.70 Mb/s in the other one, so that's more than 80Mb/s. That's what I expected with a single thread, so thanks for the information. It seems I will have to buy better hard drives if I want increased performance... A Dimecres 30 Maig 2007 22:13, Luke Lonergan va escriure: > Not for one thread/process of I/O. Mirror sets can nearly double the read > performance on most RAID adapters or SW RAID when using two or more > thread/processes, but a single thread will get one drive worth of > performance. > > You should try running two simultaneous processes during reading and see > what you get.