The new S2000Online server (geeks only)
Registered User
Joined: Mar 2001
Posts: 38
Likes: 0
From: Santa Cruz
Mmmm..the 3ware in mode 10 actually has all drives working in parallel on reads (it's unusual in this respect) and as a result is faster than 5 with the same number of drives configured. Of course, you end up with less total disk space.
For a review and lotsa benchmarks, look here:
http://www.storagereview.com/articles/2001...Escalade_1.html
The test are under windows, but since they are testing the storage subsystem the results under linux should be similar.
For a review and lotsa benchmarks, look here:
http://www.storagereview.com/articles/2001...Escalade_1.html
The test are under windows, but since they are testing the storage subsystem the results under linux should be similar.
Registered User
Joined: Dec 2000
Posts: 4,127
Likes: 0
From: Orange Park
Cthree,
Thanks for keeping us informed and for future upgrades, maybe you should consider asking for contributions from board members. I would be more than happy to contibute monetarily or otherwise. I work for a software company and we are VAR's for most of the big names and I do alot of testing on various RAID and hardware configurations.
Krazik,
Not arguing here, just spirited conversation. The BURST rates of transfer of IDE versus SCSI is one thing but sustained rates is something else. Most companies (IBM, HP, etc.) don't offer you the option of putting IDE drives in a server for many reasons. They are more prone to failure and fragmentation and they can't maintain the same sustained transfer rates as SCSI, in general. Ultra 160 versus UDMA100 is no contest, the Ultra 160 will win every time. We (my company) run on Progress. Progress doesn't have the name recognition that someone like Oracle has, does that mean it's not as good. No, Progress kicks Oracles ass across the board and there is a reason for that. Progress spends a great deal of time researching various configurations such RAID, etc. RAID 0 + 1 when set up properly will kick the crapp out of a RAID 5 array every time. That being said, I understand that if you don't have enough channels to do the 'proper' set-up this argument is no longer valid.
[Edited by derryck on 03-20-2001 at 08:01 AM]
Thanks for keeping us informed and for future upgrades, maybe you should consider asking for contributions from board members. I would be more than happy to contibute monetarily or otherwise. I work for a software company and we are VAR's for most of the big names and I do alot of testing on various RAID and hardware configurations.
Krazik,
Not arguing here, just spirited conversation. The BURST rates of transfer of IDE versus SCSI is one thing but sustained rates is something else. Most companies (IBM, HP, etc.) don't offer you the option of putting IDE drives in a server for many reasons. They are more prone to failure and fragmentation and they can't maintain the same sustained transfer rates as SCSI, in general. Ultra 160 versus UDMA100 is no contest, the Ultra 160 will win every time. We (my company) run on Progress. Progress doesn't have the name recognition that someone like Oracle has, does that mean it's not as good. No, Progress kicks Oracles ass across the board and there is a reason for that. Progress spends a great deal of time researching various configurations such RAID, etc. RAID 0 + 1 when set up properly will kick the crapp out of a RAID 5 array every time. That being said, I understand that if you don't have enough channels to do the 'proper' set-up this argument is no longer valid.
[Edited by derryck on 03-20-2001 at 08:01 AM]
Registered User
Joined: Mar 2001
Posts: 38
Likes: 0
From: Santa Cruz
Sustained rate of the 3ware in a 4 drive raid 0 config is 103MB/sec! This is substantially faster than each individual drive.
As krazik mentioned, the 3ware controller really does access all drives simultaneously then efficiently combines the result.
It's raid 5 scores aren't as impressive as raid 1/0, but it was not originally designed to be a raid 5 controller. They added that on later using just firmware so they could have that bullet item in their marketing material. The raid 0/1 stuff is supported by custom chips to make it very efficient and scale real well.
The reason SCSI has ruled in the server market is largely historical and it will probably be replaced in the cost-sensitive segment of the market by IDE (3ware). If money is no object, well, that's another matter.
As krazik mentioned, the 3ware controller really does access all drives simultaneously then efficiently combines the result.
It's raid 5 scores aren't as impressive as raid 1/0, but it was not originally designed to be a raid 5 controller. They added that on later using just firmware so they could have that bullet item in their marketing material. The raid 0/1 stuff is supported by custom chips to make it very efficient and scale real well.
The reason SCSI has ruled in the server market is largely historical and it will probably be replaced in the cost-sensitive segment of the market by IDE (3ware). If money is no object, well, that's another matter.
Thread Starter
Administrator
Joined: Oct 2000
Posts: 20,274
Likes: 4
From: Toronto, Canada
hehe, if money is no object, please call me immediately! 
Actually, thanks 89 for pointing me to that article. I've actually switched to RAID 10 based on what I read. Good tip.
Actually, thanks 89 for pointing me to that article. I've actually switched to RAID 10 based on what I read. Good tip.
Registered User
Joined: Dec 2000
Posts: 4,127
Likes: 0
From: Orange Park
RAID 0 is and should always be faster, however, it doesn't provide the main objective in most RAID arrays which is redundancy. I agree that a portion of the server market will move towards IDE. DELL already offers a lower end server with IDE drives only. However, this will not be acceptable in most any enterprise environment currently. SCSI is faster and more reliable and there just isn't any way around that fact. To those people that cost is more important than their data, sure, IDE will do fine. I have customers that are 24/7 operations and cannot afford any down time. For people like that it would not be wise for them to have a server based on IDE drives.
Thread Starter
Administrator
Joined: Oct 2000
Posts: 20,274
Likes: 4
From: Toronto, Canada
In theory, you are correct. However, this site ran for 4 months on a AMDK63 with 256MB and an single IDE drive without any downtime or data loss. We get 2 million hits/week. So while you are right to some degree about the enterprise arena, many environments don't require that level security or performance. I wouldn't run a bank on the new server but to run S2000Online it's more than enough. I guess the key is that there are a wide range of options available and price points for each. The trick is to design a system that meets the specific need of the consumer.
I should also point out that for the past year I have been the lead engineer on a NAS project that is exclusively SCSI and FCAL. I have seen more dead and near-dead SCSI and FC drives in the past year than I ever thought I would in my lifetime. Just last week I had six FCAL 36GB drives from Seagate that were working just fine. Then I rebooted the system and one of the drives failed and it brought down both FCAL loops and the RAID controller it was connected to. I low-level formatted the drive and reinserted it and it worked again. Here is a situation where the greatest of enterprise technology knocked out an entire array due to the bizarre behavior of a single, brand new, latest, greatest hard drive.
So while in theory you are correct I have personally witnessed and been involved in a non-stop parade of failed SCSI and FCAL equipment. During that time, not a single IDE drive I came across failed, even once.
I've had problems with Hitachi, IBM, Seagate and Quantum. FC is especially bad and would not recommend it to anyone. That includes drives from all manufacturers and controllers from Interphase, Qlogic, VMIC, Mylex and ICP Vortex and switches from brocade and others. Simply put FCAL is a POS I would not go near again. It simply doesn't work.
I should also point out that for the past year I have been the lead engineer on a NAS project that is exclusively SCSI and FCAL. I have seen more dead and near-dead SCSI and FC drives in the past year than I ever thought I would in my lifetime. Just last week I had six FCAL 36GB drives from Seagate that were working just fine. Then I rebooted the system and one of the drives failed and it brought down both FCAL loops and the RAID controller it was connected to. I low-level formatted the drive and reinserted it and it worked again. Here is a situation where the greatest of enterprise technology knocked out an entire array due to the bizarre behavior of a single, brand new, latest, greatest hard drive.
So while in theory you are correct I have personally witnessed and been involved in a non-stop parade of failed SCSI and FCAL equipment. During that time, not a single IDE drive I came across failed, even once.
I've had problems with Hitachi, IBM, Seagate and Quantum. FC is especially bad and would not recommend it to anyone. That includes drives from all manufacturers and controllers from Interphase, Qlogic, VMIC, Mylex and ICP Vortex and switches from brocade and others. Simply put FCAL is a POS I would not go near again. It simply doesn't work.
Registered User
Joined: Dec 2000
Posts: 4,127
Likes: 0
From: Orange Park
Cthree,
You're right but I think part of the problem there is that the newer 'generation' of hardware is becoming so finely tuned that it sometimes causes problems. Example: I recently set up a Citrix Metaframe Farm (2 servers) for a customer. Both servers were of the high quality materials but on one of the servers, the Mylex RAID controller continually failed upon the first reboot after setting the ARRAY to hardware. Couldn't for the life of me figure out what the problem was. Spoke to Mylex repeatedly about the problem and it turned out that of the 4 'identical' Ultra 160 IBM Ultrastar Drives one of them was manufactured in a different facility than the rest. Turned out this drive was the culprit. I thought it was a long shot, making sure all drives were from the same facility, but it worked. The tech I spoke to said this was actually a problem that they had encountered on many occassions. Odd that supposedly 'identical' drives didn't play well together.
You're right but I think part of the problem there is that the newer 'generation' of hardware is becoming so finely tuned that it sometimes causes problems. Example: I recently set up a Citrix Metaframe Farm (2 servers) for a customer. Both servers were of the high quality materials but on one of the servers, the Mylex RAID controller continually failed upon the first reboot after setting the ARRAY to hardware. Couldn't for the life of me figure out what the problem was. Spoke to Mylex repeatedly about the problem and it turned out that of the 4 'identical' Ultra 160 IBM Ultrastar Drives one of them was manufactured in a different facility than the rest. Turned out this drive was the culprit. I thought it was a long shot, making sure all drives were from the same facility, but it worked. The tech I spoke to said this was actually a problem that they had encountered on many occassions. Odd that supposedly 'identical' drives didn't play well together.
