On Sun, 7 Nov 1999, Matthew Clark wrote:
> What would be a decent write performance from a SCSI H/W RAID array?
>
> We have 5 x Ultra Wide 9Gb 10000RPM SCSI Disks (So supposedly 40MB per SCSI
> channel?) 4 of which are ONLINE with 1 Hot Spare..(RAID 5, 26Gb Logical
> drive, MegaRAID controller, Single PIII 500 Xeon processor)
>
> Now I would expect at least 15+ MB/sec from this array (Bonnie, Write Chars)
> as I get around 8MB per sec from a single IDE drive..
For sequential IOs, recent 10000 RPM drives are able to sustaint 20 MB/s
and even more. These drives are happy on LVD SCSI BUSes (80 MB/s).
Using 5 of these drives on a single UW (40 MB/s) SCSI BUS is probably a
bottleneck for sequential IOs.
Basically if you use 3 10K RPM drives on a single U2W SCSI BUS you can
expect at least 50 MB/s systaint.
May-be what you want to check is not sequential IO performances but
server-like IO pattern performances, that are mostly random small IOs
than sequential large IOs.
Anyway, if we consider the following:
1) Average IO chunk for real applications is mostly less that 8KB
2) On a U2W BUS it lasts about 100 micro-seconds to burst 8KB.
3) Hardware RAID add significant latency to SCSI IOs.
Then it gets clear to me that using hardware raid leads to theorical
drive performances to be killed.
If you add that the RAID hardware may want to actually write the data
before telling you the IOs is completed and have to deal with
redundancies, then write performances can only be highly affected,
especially for sequential IO patterns.
Basically, if you want hardware raid, then you want to be slow, notably
for sequential IOs and especially for sequential writes.
> Theoretically, it should be 20+ MB/sec shouldn't it?
May-be this way low sequential write throughput should be possible.
G�rard.
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
