RAID Caching and SSD Drives
-
@scottalanmiller said
You mean the SSD's own cache?
No, the RAID cache. I thought that is what was being discussed.
-
-
Hopefully xByte will be by soon
-
@aaronstuder said in RAID Caching and SSD Drives:
Hopefully xByte will be by soon
I found e-mailing the person you are looking for and referencing a thread (by the link) s the best way to get them to respond quickly.
Since the notifications don't really work on ML, it's the only good way.
-
To get the best performance enable write-through cache, and no read ahead in the controller BIOS. Dell calls Cut-Through IO specifically to enhance the performance of SSD arrays.
-
@BradfromxByte Thanks brad! Can you PM me your contact details?
-
@BradfromxByte said in RAID Caching and SSD Drives:
To get the best performance enable write-through cache, and no read ahead in the controller BIOS. Dell calls Cut-Through IO specifically to enhance the performance of SSD arrays.
So my comment that the PERC takes care of it was kind of correct.
-
@BradfromxByte said in RAID Caching and SSD Drives:
To get the best performance enable write-through cache, and no read ahead in the controller BIOS. Dell calls Cut-Through IO specifically to enhance the performance of SSD arrays.
http://www.computerweekly.com/feature/Write-through-write-around-write-back-Cache-explained
Normally write through does not disable the cache, just changes it for safety. Do you know what aspect of write through would improve performance?
-
From the article:
"Write-through cache directs write I/O onto cache and through to underlying permanent storage before confirming I/O completion to the host"
"Write-through cache is good for applications that write and then re-read data frequently as data is stored in cache and results in low read latency". -
@BradfromxByte said in RAID Caching and SSD Drives:
From the article:
"Write-through cache directs write I/O onto cache and through to underlying permanent storage before confirming I/O completion to the host"
"Write-through cache is good for applications that write and then re-read data frequently as data is stored in cache and results in low read latency".I didn't know this. Valuable information. Thanks
-
@BradfromxByte said in RAID Caching and SSD Drives:
From the article:
"Write-through cache directs write I/O onto cache and through to underlying permanent storage before confirming I/O completion to the host"
"Write-through cache is good for applications that write and then re-read data frequently as data is stored in cache and results in low read latency".How does write-back not do that as well, though?
-
They are both directed to cache, but the difference is when the I/O is confirmed to the host:
Write-Back Cache - is where write I/O is directed to cache and completion is IMMEDIATELY confirmed to the host. (mixed work loads)
Write-through cache- directs write I/O onto cache and through to underlying permanent storage BEFORE confirming I/O completion to the host.
-
@BradfromxByte said in RAID Caching and SSD Drives:
They are both directed to cache, but the difference is when the I/O is confirmed to the host:
Write-Back Cache - is where write I/O is directed to cache and completion is IMMEDIATELY confirmed to the host. (mixed work loads)
Write-through cache- directs write I/O onto cache and through to underlying permanent storage BEFORE confirming I/O completion to the host.
Exactly, so the write-back would logically be faster. At least based on that fact there. Since in both cases the data is cached, what would make write-though faster here? The benefit listed for it isn't a benefit, it's just not a deficit. So we must be missing the reason for why it is recommended.
-
As mentioned earlier, Dell recommends Cut-Through IO. The Cut-Though IO is an IO accelerator for SSD arrays that boosts the throughput of devices connected to the PERC Controller. It is enabled through disabling the write-back cache (enable write-through cache) and disabling Read Ahead.
-
@BradfromxByte said in RAID Caching and SSD Drives:
As mentioned earlier, Dell recommends Cut-Through IO. The Cut-Though IO is an IO accelerator for SSD arrays that boosts the throughput of devices connected to the PERC Controller. It is enabled through disabling the write-back cache (enable write-through cache) and disabling Read Ahead.
Right, but logically that makes it slower based on everything that we know. That Dell "calls it" an accelerator tells us nothing. why do they recommend it is really the question as their documentation would suggest that this is not the right setup.
There has to be something being missed. Disabling read-ahead, that probably makes sense. but turning off write-back?
-
Dell also recommends really low end, fragile SANs with stability problems where they aren't appropriate at all and they call things RAID 10 that obviously are not. Dell sometimes even confuses themselves with misused terms, so we have to watch them carefully.
-
@scottalanmiller said in RAID Caching and SSD Drives:
Dell also recommends really low end, fragile SANs with stability problems where they aren't appropriate at all and they call things RAID 10 that obviously are not. Dell sometimes even confuses themselves with misused terms, so we have to watch them carefully.
ROFLOL - so true!
-
Asked the same question in SW. The results i got using CrystalDiskMark didn't make sense as well.
Write-though, 4K Q32T1 - write at 90000 IOPS, 369MB/s
Write-back, 4K Q32T1 - write at 37000 IOPS, 151MB/s -
@Kris_K said in RAID Caching and SSD Drives:
Asked the same question in SW. The results i got using CrystalDiskMark didn't make sense as well.
Write-though, 4K Q32T1 - write at 90000 IOPS, 369MB/s
Write-back, 4K Q32T1 - write at 37000 IOPS, 151MB/sClearly there's more tech involved that we don't know about.
-
@Kris_K said in RAID Caching and SSD Drives:
Asked the same question in SW. The results i got using CrystalDiskMark didn't make sense as well.
Write-though, 4K Q32T1 - write at 90000 IOPS, 369MB/s
Write-back, 4K Q32T1 - write at 37000 IOPS, 151MB/sWhat controller, what sized cache?
Very strange, so what is the controller doing wrong to make this happen?
