topic Re: Performance VRAID question in HPE EVA Storage

Performance VRAID question

Oscar_Maimó — Tue, 10 Mar 2009 17:49:04 GMT

Hi guys:
I have some questions about performance:

1.Mix different VRAID levels in the same DiskGroup affect performance?

2. What VRAID level offers the best performance at reading

Greetings
Oscar

Re: Performance VRAID question

Víctor Cespón — Tue, 10 Mar 2009 18:25:44 GMT

1) No
2) VRAID 0, followed by VRAID 5 and VRAID 1

Re: Performance VRAID question

IBaltay — Tue, 10 Mar 2009 18:27:02 GMT

hi,
ftp://ftp.compaq.com/pub/products/storageworks/whitepapers/5982-9140EN.pdf

Re: Performance VRAID question

Oscar_Maimó — Tue, 10 Mar 2009 19:13:25 GMT

Thanks for quick response:
I was reading EVA best practices white paper but there is no information about mixing different VRAID levels in the same Group.

Should I assume there is no problem from the point of performance and availability?.

Greetings
Oscar

Re: Performance VRAID question

IBaltay — Tue, 10 Mar 2009 19:18:22 GMT

this is the availability best practices:
http://h71028.www7.hp.com/ERC/downloads/4AA1-4202ENW.pdf?jumpid=reg_R1002_USEN

Re: Performance VRAID question

Oscar_Maimó — Tue, 10 Mar 2009 19:26:42 GMT

Thanks iBaltay:
I have read that white paper and I know each VRAID characteristics but i talk abot "mixing VRAID levels in the same Disk Group" more specifically.

Greetings
Oscar

Re: Performance VRAID question

Oscar_Maimó — Tue, 10 Mar 2009 19:35:02 GMT

This topic is not specified in white papers

Re: Performance VRAID question

IBaltay — Tue, 10 Mar 2009 19:39:30 GMT

a)from the performance point of view, it is better to have the fewest possible number of DGs. Therefore creating one big DG and mix the RAID1/RAID5 in it is the most used method. Thus all the RAID1&RAID5 VDISKs are distributed across as many physical disks as possible which is giving us the power at the back-end.
b)To have reasonable availability the double disk protection should be set which equals to 4 virtual spare drives, but single disk protection (2 virtual spare drives) are most often sufficient.
c) if we split the EVA backend into the DG of RAID5 which is usualy bigger and RAID1 which is usualy smaller, the advantage of the separation is given back with the smaller size of the RAID1 DG...

Re: Performance VRAID question

Oscar_Maimó — Tue, 10 Mar 2009 19:44:05 GMT

Thanks a lot iBaltay:
your answers will be pointed and vcespon's to
Greetings
Oscar

Re: Performance VRAID question

Víctor Cespón — Tue, 10 Mar 2009 19:44:38 GMT

Each VRAID type has it's availability and speed characteristics.

VRAID 0 is the fastest but has no redundancy
VRAID 5 is fast in reads but slow on writes, you can lose one disk from each RSS
VRAID 1 is fast, very redundant, you can lose several disks as long as they are not a mirror pair.

The EVA uses all physical disks for each vdisk, the information is divided in segments, the redundancy added, and that is written to the disks.

Of course, looking at the performance, it's better to have all vdisks in VRAID 0, if you create vdisks in VRAID 1, the total performance will be lower, but not because of a problem on the EVA controllers, but because the VRAID 1 has to write each segment twice, by definition.

The same for VRAID 5, calculating the parity takes I/Os and time, so writting to a VRAID 5 vdisk is slower than writting to a VRAID 0.