Comparison of 3par peer persistence and XP7 in RAC environments

par does not support the strict meaning of the storage of double live, but its peerpersistence features are very interesting to store in a dual data center configuration. Here we take a RAC Application as an example to analyze it.

Let's start with an architecture diagram to see how this architecture is going.

650) this.width=650; "title=" 3par pp1.png "src=" http://s4.51cto.com/wyfs02/M01/80/7E/ Wkiol1dderbgdngraafvwdhjwuw793.png "alt=" Wkiol1dderbgdngraafvwdhjwuw793.png "/>

There are two modules in this architecture: one is the Server +oracle RAC composed by rac provide services outside and manage server clusters for possible physical device failures and load balancing allocations; the other is the 3par store and san network switches and the quorum server running on the virtual machine constitute the storage space through peer persistence show to server and rac software.

If you read my previous article , there will be a feeling: it and hp XP7 > The double-work plan is like yes! xp7 The virtual disk array is the middle tier that is stored on the host, 3par Peer persistence

It's going to be about 3par Peer Persistence with the HP XP7 to the host IO the routes were assigned:

LXP7Virtual Disk array , the receivedIOaccording to certain (such as local storage first)IOsent to two disk arrays. ALUNsimages on two disk arrays can be processed at the same time by the host .IO, there is no master-slave difference. XP7The fact that the LUNs are mirrored on two arrays can handle host IO, but their wwn is the same, so how does the host know which LUNs are local?

l  lun There is a master-slave difference between the host and the io alua

The picture below shows the two io Different on the path, for , two yellow The path is inactive in normal state without io xp7

650) this.width=650; "title=" 3par pp2.png "src=" http://s3.51cto.com/wyfs02/M02/80/7E/ Wkiol1ddevdw0tweaadnsb7xgdi232.png "alt=" Wkiol1ddevdw0tweaadnsb7xgdi232.png "/>

3parof thePeer PersistenceHow do you manage an array without a yellow path? It throughALUA(Asymmetriclogical Unit AccessAsynchronous Logical Unit Access) protocol that sets the primary array to the active/Optimization (Active/optimized) state, set the from array to active/Non-optimized (Active/unoptimized) state. So the host naturallyIoare sent to the main array. When the primary array fails,peerpersistenceand it will turn the arrayset to Active/Optimization (Active/optimized) state, set the primary array to active/Non-optimized (Active/unoptimized) state. This is illustrated below:

650) this.width=650; "title=" 3par pp3.png "src=" http://s2.51cto.com/wyfs02/M01/80/80/ Wkiom1ddeiqydoycaadxwlrzie4596.png "alt=" Wkiom1ddeiqydoycaadxwlrzie4596.png "/>

at this point, is it possible to say XP7 whether the double work is necessarily better than 3par of the Peer Persistence it? On the surface, yes, for the following reasons:

1. Since the two arrays are active at the same time, the switchover will be quick and the theoretical host and RAC software are not aware.

2. in the case of XP7, the two arrays handle the host IOat the same time, and theload is balanced.

Now, we have a deeper analysis of the two situations:

1. Storage Switching speed problem, need two kinds of situation discussion

L Normal manual switching: Double live switching speed, basic in a few seconds to complete, the front-end business will not be aware; ALUA the switching speed of the mode is slightly slower, the delay can reach dozens of seconds, the front-end business will be slightly sluggish.

L Abnormal Automatic Switching: Both methods take time to determine the state of the peer disk array (in order to prevent repeated switching, this time threshold is not very short), to query and set up an arbitration body, these actions will take a few 10 seconds to complete, so the time difference is not too large.

2. two arrays handle host IO at the same time , balancing load issues:

Lwhile both arrays can handle the hostIO, but in order to keep the data consistent, theseIOYou must synchronize the write operation to another array. In other words, the first2and Section3steps cannot be saved. Only the first1and Section4steps Save the transfer time. However, the double-live is due to the complex locking mechanism which brings the internal software processing time delay in the array. Whether this delay can offset the first1and Section4the transfer time of the step is saved, it is debatable. It is also important to note that as the system load grows, IOthe larger the flow, the longer the processing time.

L In general , an array often contains multiple application data, grouping these application data, setting different primary storage can achieve the effect of manual load balancing. Note, however , that the data for an application must be set up in a group, and it is not possible to set up different primary storage by grouping.

650) this.width=650; "Title=" XP7 double live lock mechanism. png "src=" http://s1.51cto.com/wyfs02/M02/80/80/ Wkiom1ddelqcrzd8aacmul0y54a322.png "alt=" Wkiom1ddelqcrzd8aacmul0y54a322.png "/>

This article is from the "splinters nonsense" blog, make sure to keep this source http://leesbing.blog.51cto.com/1344594/1782307

Comparison of 3par peer persistence and XP7 in RAC environments