RAID10 Performance on newer hardware slower then RAID5 performance on older hardware?

ADY

9/5/23, 1:54 PM

I am currently trying to figure out a performance issue with a new server compared to a old server. Performance should be better but it's not and I'm having trouble figuring out why.

Old server is a Dell R640 (2x Xeon Gold 5218, 128Gb RAM) with a PERC H740P Mini and 8Gb cache. 6x Samsung Pm1645 1.6Gb 12Gb/s SAS mixed use enterprise SSDs in a RAID 5 virtual disk

New server is a Dell R650 (2x Xeon Gold 6330, 256Gb RAM) with a PERC H755 and 8Gb cache. 6x Seagate Nytro 2532 1.92Tb 12Gb/s SAS mixed use enterprise SSDs in a RAID 10 virtual disk

The new drives are physically slower in both reads and writes but going from a 6 drive RAID5 array to a 6 drive RAID10 I thought that would make up a good deal of that. The old drives are rated at 1Gb sequential read and 1.2Gb sequential write while the new drives are at 840Mb and 650Mb. In a simple test creating a new 200Gb vhdx on both machines: the old server creates the file in about 46 seconds, the new server takes 102 seconds.

Should it really take twice as long even with the performance gain of the newer RAID? How is the old RAID5 doing it so quickly or am I just not understanding something.

0 + 0

raid

ssd

dell-perc

Score:2

Server

wazoox

9/5/23, 3:01 PM

A RAID-10 array is as fast as a RAID-0 array half its size. It can be somewhat faster than that in some configurations or for some workloads (reading alternatively from both drives of each pair). So your 6 drives RAID-10 should be 3 times as fast as one of your newer drives: 3 x 840MB/s (not Mb!) equals ~2500MB/s.

A RAID-5 array is as fast (for sequential operations only, it's much slower at random writes) as a RAID-0 with one less drive. So your 6 drives RAID-5 should be 5 times as fast as your older drives: 5 x 1GB/s = 5 GB/s. Which is about twice as fast as your newer setup, and match quite well your numbers.

0 + 0

ADY

9/5/23, 3:10 PM

So if the write performance of the new drives is 650 then in a RAID10 it's 1950. And writing a 200Gb file / 1950 = 102.5 seconds which is what I'm getting. Does that mean if I reconfigure the new server to RAID5 it should be 5 * 650 = 3250 sequential write? Also this server will host 8 - 10 virtual machines (SQL, File, WSUS, etc) so maybe this doesn't matter since I will have a lot of random reads and writes and the RAID10 will have better performance for that then RAID5?

wazoox

9/5/23, 4:24 PM

@ADY yes if you reconfigure the array as RAID-5 it should be faster at sequential operations, but it will be probably slower at random writes which are pretty common on VM servers. So it's impossible to make a general statement without benchmarking. You could use the RAID-10 and compare with the older system to see which one performs best in your particular setup.

ADY

9/5/23, 8:18 PM

Thank-you. Gonna leave it as RAID10 and see how things go. We only create or copy the VM's once so much more concerned about the constant little reads and writes.

Elon Musk

I sit in a Tesla and translated this thread with Ai:

EN: RAID10 Performance on newer hardware slower then RAID5 performance on older hardware?

TH: ประสิทธิภาพของ RAID10 บนฮาร์ดแวร์รุ่นใหม่ช้ากว่าประสิทธิภาพของ RAID5 บนฮาร์ดแวร์รุ่นเก่าหรือไม่

RO: Performanța RAID10 pe hardware mai nou mai lentă decât performanța RAID5 pe hardware mai vechi?

RU: Производительность RAID10 на новом оборудовании ниже, чем производительность RAID5 на старом оборудовании?

VI: Hiệu suất RAID10 trên phần cứng mới hơn chậm hơn hiệu suất RAID5 trên phần cứng cũ hơn?

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.