vBulletin Search Engine Optimization
| |||||||
| Register | FAQ | Members List | Calendar | Search | Today's Posts | Mark Forums Read |
| ||||
| Hey all... Does anyone here have any experience with running DB2 on a NetApp filer (storage)? If so, what volumes did you use (how many disks per volume). For those who do not know what a Netapp Filer is: It's a storage device created by a select group of Cisco employees. It connects to a server using NAS or iSCSI. What makes it different from other networked storage arrays is that it has a x GB buffer for writes. So for the database write to it, it seems that it's performing writes on a Ramdisk. In the buffer, data is preformatted in Raid4 order and than flushed to disk. reads are performed buffered from the raid-4 array. We're planning on using such a device for storing data, and I was curious if anyone has had any positive or negative experiences with such a setup. Note: I already read a few best-practice papers, but I'd like to know from real people in real situations how it performs. We're using DB2 for a lot of small concurrent transactions. -R- |
| |||
| Jurgen Haan wrote: > Hey all... Does anyone here have any experience with running DB2 on a > NetApp filer (storage)? If so, what volumes did you use (how many disks > per volume). > > For those who do not know what a Netapp Filer is: > It's a storage device created by a select group of Cisco employees. > It connects to a server using NAS or iSCSI. > What makes it different from other networked storage arrays is that it > has a x GB buffer for writes. So for the database write to it, it seems > that it's performing writes on a Ramdisk. In the buffer, data is > preformatted in Raid4 order and than flushed to disk. > > reads are performed buffered from the raid-4 array. > > We're planning on using such a device for storing data, and I was > curious if anyone has had any positive or negative experiences with such > a setup. > > Note: I already read a few best-practice papers, but I'd like to know > from real people in real situations how it performs. I've used NetApps at a few different customer sites. A couple of things: 1) Assuming you're using it as a NAS (with SMB or NFS), make *sure* you have the appropriate patches on your database server. Work with NetApp to find the latest suggested patch levels. 2) Make sure you have sufficient, dedicated network bandwidth between the database server and the Filer. If you are thinking about using Etherchannel (bonding multiple individual network connections into 1 virtual connection), make sure you understand your vendor's implementation of Etherchannel before going down this path. 3) Volumes configuration: I've only worked on Filers for business intelligence applications (tuning for read performance), so I'm not sure if my experience is applicable to an OLTP environment. However, we got good performance using 11+P RAID groups, and even better performance when we combined multiple RAID groups into a single volume (i.e. the volumes we were using had 3 11+P RAID groups). For OLTP, I suspect you may find that you have better experience using multiple volumes, each built on separate, smaller (5+P) RAID groups. Good luck, |
| |||
| Ian wrote: > > I've used NetApps at a few different customer sites. A couple of > things: > > 1) Assuming you're using it as a NAS (with SMB or NFS), make *sure* you > have the appropriate patches on your database server. Work with > NetApp to find the latest suggested patch levels. Well... iSCSI has crossed my mind Are there any signifigant performance differences between iSCSI and NAS? > > 2) Make sure you have sufficient, dedicated network bandwidth between > the database server and the Filer. If you are thinking about using > Etherchannel (bonding multiple individual network connections into > 1 virtual connection), make sure you understand your vendor's > implementation of Etherchannel before going down this path. I was thinking of trunking the 2 1Gbit channels into 1 2Gbit. What use is a fast storage solution if you use a poor connection. > 3) Volumes configuration: I've only worked on Filers for business > intelligence applications (tuning for read performance), so I'm not > sure if my experience is applicable to an OLTP environment. > > However, we got good performance using 11+P RAID groups, and even > better performance when we combined multiple RAID groups into a > single volume (i.e. the volumes we were using had 3 11+P RAID > groups). Ah... this is very useful information. > For OLTP, I suspect you may find that you have better experience > using multiple volumes, each built on separate, smaller (5+P) RAID > groups. > Thank you for your detailed response, I will certainly keep it in mind. -R- |
| |||
| Jurgen Haan wrote: > Ian wrote: > >> >> I've used NetApps at a few different customer sites. A couple of >> things: >> >> 1) Assuming you're using it as a NAS (with SMB or NFS), make *sure* you >> have the appropriate patches on your database server. Work with >> NetApp to find the latest suggested patch levels. > > > Well... iSCSI has crossed my mind > Are there any signifigant performance differences between iSCSI and NAS? There should be pretty significant performance differences. I haven't used iSCSI, but iSCSI is a method for attaching the storage directly to your server, like Fibre Channel, SATA. NAS is a fileserver on your network that you connect to using TCP/IP protocols like NFS or SMB. >> 2) Make sure you have sufficient, dedicated network bandwidth between >> the database server and the Filer. If you are thinking about using >> Etherchannel (bonding multiple individual network connections into >> 1 virtual connection), make sure you understand your vendor's >> implementation of Etherchannel before going down this path. > > > I was thinking of trunking the 2 1Gbit channels into 1 2Gbit. > What use is a fast storage solution if you use a poor connection. Again, I can't speak to iSCSI, but it is probably different than normal ethernet. Generally I have found that when using Etherchannel, the particular NIC within the trunk is selected based on the destination IP address. Therefore, if all traffic is going to 1 IP address (i.e. the NAS server), then all traffic goes through one interface. Etherchannel does have a different routing mechanism (round-robin), but using it has never resulted in performance gains you might expect. |
| |||
| Ian wrote: > > > Generally I have found that when using Etherchannel, the particular NIC > within the trunk is selected based on the destination IP address. > Therefore, if all traffic is going to 1 IP address (i.e. the NAS > server), then all traffic goes through one interface. Etherchannel > does have a different routing mechanism (round-robin), but using it > has never resulted in performance gains you might expect. > iSCSI is a fairly new method with both the advantages of NAS and SAN combined. It uses stock Ethernet components to tunnel SCSI blocks through TCP/IP. In a linux environment you can use special Kernel modules to transform a NIC into a HBA. But I don't know whether both the linux kernel as well as the netapp can trunk multiple connections, but that's an issue I can address at the local netapp vendor. How is your overall opinion on the netapp as DB2 storage? Thanks -R- |
| |||
| Jurgen Haan wrote: > Ian wrote: > >> >> >> Generally I have found that when using Etherchannel, the particular NIC >> within the trunk is selected based on the destination IP address. >> Therefore, if all traffic is going to 1 IP address (i.e. the NAS >> server), then all traffic goes through one interface. Etherchannel >> does have a different routing mechanism (round-robin), but using it >> has never resulted in performance gains you might expect. >> > > iSCSI is a fairly new method with both the advantages of NAS and SAN > combined. It uses stock Ethernet components to tunnel SCSI blocks > through TCP/IP. In a linux environment you can use special Kernel > modules to transform a NIC into a HBA. But I don't know whether both the > linux kernel as well as the netapp can trunk multiple connections, but > that's an issue I can address at the local netapp vendor. > > How is your overall opinion on the netapp as DB2 storage? > > Thanks > > -R- From an Oracle standpoint NetApp is one of the best storage solutions available. The one thing I would suggest you check out too is the Apple Xserve RAID. They are faster than both NetApp and EMC, are priced well below either, and mount on Linux. -- Daniel A. Morgan http://www.psoug.org damorgan@x.washington.edu (replace x with u to respond) |
| ||||
| DA Morgan wrote: > > From an Oracle standpoint NetApp is one of the best storage solutions > available. The one thing I would suggest you check out too is the > Apple Xserve RAID. They are faster than both NetApp and EMC, are > priced well below either, and mount on Linux. Aha... Interesting device. Thank you. -R- |
| Thread Tools | |
| Display Modes | |
|
|